CN117312518A

CN117312518A - Intelligent question-answering method and device, computer equipment and storage medium

Info

Publication number: CN117312518A
Application number: CN202311281707.XA
Authority: CN
Inventors: 郑汉锋
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2023-09-28
Filing date: 2023-09-28
Publication date: 2023-12-29

Abstract

The embodiment of the application discloses an intelligent question-answering method, an intelligent question-answering device, computer equipment and a storage medium, which are applicable to scenes such as artificial intelligence, intelligent car machines and the like. The method comprises the following steps: and responding to the question-answer request, acquiring question information from the question-answer request, and respectively calling a plurality of intelligent question-answer models to intelligently answer the question information to obtain candidate answer information obtained by each intelligent question-answer model aiming at the question information, wherein the candidate answer information comprises one or more text fragments. And then, determining the matching degree between each text segment and the question information, taking the text segments with the matching degree larger than or equal to the matching degree threshold value as candidate text segments to obtain one or more candidate text segments, and finally generating answer information containing at least one candidate text segment for the question information. By adopting the embodiment of the application, when the questioning object initiates the questioning information into any knowledge field, the answer information highly matched with the questioning information can be obtained.

Description

Intelligent question-answering method and device, computer equipment and storage medium

Technical Field

The present disclosure relates to the field of artificial intelligence technologies and natural language processing technologies, and in particular, to an intelligent question-answering method, apparatus, computer device, and storage medium.

Background

With the rapid development of artificial intelligence technology and natural language processing technology, an intelligent question-answering model has also been developed. However, the answer capability of the intelligent question-answer model in each knowledge field is limited by the selection of training data of the intelligent question-answer model, so that when a question object initiates question information in different knowledge fields to the same intelligent question-answer model, the intelligent question-answer model is difficult to obtain answer information which is highly matched with each question information. In view of this, how to obtain answer information highly matching the question information by using an intelligent question-answer model when the question object initiates the question information into any knowledge domain becomes the current hot spot research topic.

Disclosure of Invention

The embodiment of the application provides an intelligent question and answer method, an intelligent question and answer device, computer equipment and a storage medium, wherein when a question object initiates question information in any knowledge domain, an intelligent question and answer model is adopted to obtain answer information which is highly matched with the question information.

In one aspect, an embodiment of the present application provides an intelligent question-answering method, including:

responding to a question-answer request, and acquiring question information from the question-answer request;

respectively calling a plurality of intelligent question-answering models to intelligently answer the question information to obtain candidate answer information obtained by each intelligent question-answering model aiming at the question information, wherein the candidate answer information comprises one or more text fragments;

determining the matching degree between each text segment and the question information, and taking the text segments with the matching degree larger than or equal to the matching degree threshold as candidate text segments to obtain one or more candidate text segments;

and generating answer information of the questioning information, wherein the answer information comprises at least one candidate text segment.

In still another aspect, an embodiment of the present application provides an intelligent question-answering device, including:

the response unit is used for responding to the question-answer request and acquiring question information from the question-answer request;

the calling unit is used for calling a plurality of intelligent question-answering models to intelligently answer the question information respectively to obtain candidate answer information obtained by each intelligent question-answering model aiming at the question information, wherein the candidate answer information comprises one or more text fragments;

The determining unit is used for determining the matching degree between each text segment and the question information, and taking the text segment with the matching degree larger than or equal to the matching degree threshold value as a candidate text segment so as to obtain one or more candidate text segments;

and the generating unit is used for generating answer information of the question information, wherein the answer information comprises at least one candidate text segment.

In yet another aspect, embodiments of the present application provide a computer device comprising:

a processor adapted to implement one or more computer programs;

a storage medium storing one or more computer programs adapted to be loaded by the processor and to perform the intelligent question-answering method according to the first aspect.

In yet another aspect, embodiments of the present application provide a storage medium storing one or more computer programs adapted to be loaded by the processor and to perform the intelligent question-answering method according to the first aspect.

In yet another aspect, embodiments of the present application provide a program product comprising a computer program adapted to be loaded by a processor and to perform the intelligent question-answering method as in the first aspect.

In the embodiment of the application, the computer equipment invokes a plurality of intelligent question-answering models to respectively answer the question information, and selects a text segment with the matching degree with the question information being greater than or equal to a matching degree threshold value from text segments contained in answer results obtained by the intelligent question-answering models aiming at the question information so as to generate the answer information based on the selected text segment. Since the question-answering capability of each intelligent question-answering model in the real application may be irregular for the same knowledge domain, and the knowledge domains related to different question information may be various, the multiple intelligent question-answering models are utilized to intelligently answer, so that the question-answering capability of different capability levels and/or different knowledge domains can be used. On the basis, the answer information is determined based on the answers of the intelligent question-answering models, so that the generated answer information is obtained after the answers of a plurality of knowledge fields are comprehensively referred, the matching degree between the answer information and the question information is effectively ensured, the answer information can be more in accordance with the question intention of the question object, and the question-answering experience of the question object is further improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic diagram of an intelligent question-answering system according to an embodiment of the present application;

FIG. 2 is a schematic diagram of a multi-model intelligent question-answering principle provided in an embodiment of the present application;

FIG. 3 is a schematic diagram of yet another multi-model intelligent question-answering provided in an embodiment of the present application;

FIG. 4 is a schematic flow chart of an intelligent question-answering method provided in an embodiment of the present application;

fig. 5 is a schematic flowchart of a specific flow of responding to a question-answer request according to an embodiment of the present application;

FIG. 6 is a schematic flow chart diagram of yet another intelligent question-answering method provided by an embodiment of the present application;

fig. 7 is a schematic diagram of a calculation principle of matching degree between a text segment and question information according to an embodiment of the present application;

FIG. 8 is a schematic diagram of a calculation principle of matching degree between a text segment and question information according to another embodiment of the present application;

Fig. 9 is a schematic structural diagram of an intelligent question-answering device according to an embodiment of the present application;

fig. 10 is a schematic structural diagram of a computer device according to an embodiment of the present application.

Detailed Description

It should be noted in advance that, in order to enable those skilled in the art to better understand the technical solutions provided by the embodiments of the present application, the embodiments of the present application will be clearly and completely described in terms of implementation manners of the technical solutions provided by the embodiments of the present application with reference to one or more drawings. Moreover, the drawings shown in the embodiments of the present application are only exemplary, and for example, the execution sequence of each step in the drawings may be adaptively adjusted according to the actual application scenario. Furthermore, in the embodiments of the present application, the block diagrams shown in the drawings are merely functional entities, and do not necessarily correspond to physically independent entities. That is, the functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor devices and/or microcontroller devices.

In addition, various technical schemes provided by the application refer to artificial intelligence technology and natural language processing technology. Then, in order to clearly understand the implementation principle of each technical solution, the following description will be related to the artificial intelligence technology and the natural language processing technology.

1. Artificial intelligence techniques.

Artificial intelligence technology is also known as AI technology, which is a shorthand for Artificial Intelligence. It is an integrated technology in the field of computer science, mainly dedicated to the study of the essence of intelligence to produce machines or devices that can react in a similar way to human intelligence. Specifically, the artificial intelligence technology is mainly used for researching design principles and implementation methods of various intelligent machines so as to enable the machines to have the functions of sensing, reasoning and decision making by utilizing the artificial intelligence technology on the machines. In practice, artificial intelligence techniques may utilize a digital computer or a machine controlled by a digital computer to simulate, extend and extend human intelligence so that the digital computer or related machine can sense the environment and acquire knowledge. In view of this, i.e., based on artificial intelligence techniques, one can implement the theory, method, technique, and application system that uses the knowledge learned by a digital computer or related machine to obtain the best results.

As artificial intelligence technology research and advances, artificial intelligence technology has developed research and applications in a variety of areas of life, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned, autopilot, unmanned, robotic, smart medical, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will find application in more fields and will be of increasing value. It will be appreciated, then, that the art of artificial intelligence involves not only hardware-level technology (or artificial intelligence hardware technology) but also software-level technology (or artificial intelligence software technology). Specifically, artificial intelligence hardware technologies generally refer to technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. Artificial intelligence software technology generally refers to computer vision technology, speech processing technology, natural language processing technology, machine learning/deep learning technology, and the like.

2. Natural language processing techniques.

Natural language processing (Natural Language Processing, NLP) technology is an important branch of artificial intelligence technology, and is mainly dedicated to research on various theories and methods capable of realizing effective communication between a person and a computer by using natural language (i.e., language used by people in daily life), so that the natural language processing technology becomes a scientific technology integrating linguistics, computer science and mathematics. In practical applications, natural language processing techniques are often combined with machine learning/deep learning techniques to build neural network models (e.g., intelligent question-answering models in this application) for implementing natural language processing. The neural network used for constructing the neural network model can be used for reflecting the behaviors of human brains, and allows a computer program to learn the recognition mode based on training data, so that the common problems in the fields of artificial intelligence, machine learning and deep learning are solved. Based on this, it can be understood that the intelligent question-answering model mentioned in the subsequent embodiments of the present application may specifically be a neural network model with a semantic understanding function obtained by training with training data.

The method mainly provides an intelligent question-answering scheme based on the artificial intelligence technology and the natural language processing technology, in the scheme, when the computer equipment acquires the question information, a plurality of intelligent question-answering models are respectively called to intelligently answer the question information, answers obtained by each intelligent question-answering model are used as candidate answer information of the question information, and each candidate answer information comprises at least one text segment. Based on this, further, the computer device selects a text segment having a matching degree greater than a matching degree threshold as a candidate text segment by calculating a matching degree between each text segment and the question information, and finally generates answer information including at least one candidate text segment for the question information. It will be appreciated that in this embodiment, both the question information and the answer information include text content, that is, both the question information and the answer information in this embodiment include one or more text characters.

Because model parameters and/or training data of different intelligent question-answering models are different, the processing capacity of each intelligent question-answering model on question information in each knowledge domain is uneven, and the scheme adopts a plurality of intelligent question-answering models to question and answer the same question information, so that answer information of the question information is selected from answers fed back by each intelligent question-answering model. Under the condition, the answers of the intelligent question-answering models of a plurality of capability levels are comprehensively referred to in the answer information determination, so that the determined answer information can have higher matching degree with the question information, and under the condition that a plurality of intelligent question-answering models are adopted for intelligent question-answering, the question information in each knowledge domain has higher probability to obtain high-quality answer information. Therefore, by adopting the intelligent question-answering scheme provided by the application in the intelligent question-answering service, when a question object initiates question information related to any knowledge field, answer information with higher matching degree with the question information can be efficiently generated, so that the question-answering experience of the question object is improved, and the user viscosity of a question-answering product developed based on the intelligent question-answering scheme can be effectively improved.

In a specific embodiment, the intelligent question-answering scheme proposed by the present application may be applied to the intelligent question-answering system shown in fig. 1, and executed cooperatively by each device in the intelligent question-answering system. In particular, the intelligent question and answer system may illustratively include n question clients labeled 101, question and answer service providers labeled 102, and m intelligent question and answer models labeled 102 in fig. 1. Wherein n is a positive integer and m is an integer greater than 1. On this basis, the n question clients may refer to question clients adopted by n different question objects, or may refer to n different clients that provide intelligent question and answer services by calling the question and answer interface or the question and answer page. For example, the n question clients may include a question client 1 and a question client 2, each of which may be running in one or more terminal devices. In addition, each of the questioning client 1 and the questioning client 2 may call a question-answer interface or a question-answer page to provide an intelligent question-answer service to the respective client-use object, and the questioning client 1 may also be used to provide an internet service a, and the questioning client B may be used to provide an internet service B different from the internet service a.

In a possible implementation manner of this embodiment, the specific implementation principle of this solution when applied to the intelligent question-answering system shown in fig. 1 may be shown in fig. 2. Based on fig. 2, it is apparent that each of the n question clients may be used to provide a service page of an intelligent question-answering service to a question object, where the intelligent question-answering service is specifically provided by a question-answering service provider. That is, any question client may be used by the question object to communicate the question information to the question and answer service provider, which may implement intelligent question and answer to the question information using the principles provided by the present solution. Specifically, after receiving the question information sent by any question client, the question service provider can call at least two intelligent question-answering models in the m intelligent question-answering models to intelligently answer the question information, and then select answer information of the question information from answers obtained by the at least two intelligent question-answering models.

It will be readily seen that in the principle shown in fig. 2, the question and answer service provider may be regarded as one question and answer interface or question and answer page that is accessible from outside, and that n question clients may provide intelligent question and answer services by calling the question and answer interface or question and answer page. That is, in the practical application of the scheme, the intelligent question-answering program developed based on the intelligent question-answering principle of the scheme can be integrated into one interface (or SDK, namely, a software tool development kit), so that the relevant client can realize the intelligent question-answering function by calling the interface, and the development efficiency of the intelligent question-answering service of the client is effectively improved.

In yet another possible implementation manner of this embodiment, the specific implementation principle of this solution when applied to the intelligent question-answering system shown in fig. 1 may also be shown in fig. 3. Based on fig. 3, in the practical application of the present solution, the questioning object may also directly initiate questioning information to the questioning and answering service provider, so as to obtain answer information fed back by the questioning and answering service provider. In this case, the question-answering service provider may be composed of one or more of a question client, a terminal device, and a server in particular. Wherein the questioning client may be running in a terminal device, and the server may be configured to provide data support services (such as data storage, data calculation, data loading, etc.) to each function implemented by the questioning client.

Based on the above description of the embodiments, it is to be understood that, in a specific embodiment, all technical solutions provided in the present application may be executed by a computer device. The computer device may include one or both of a terminal device and a server.

When the computer device comprises a terminal device, the terminal device may specifically include, but is not limited to: smart phones, tablet computers, notebook computers, desktop computers, vehicle terminals, smart televisions, game consoles, and the like. In practical application, an application program (or a client) for realizing the intelligent question-answering service can be run in the terminal equipment, and the application program is developed based on the multi-model question-answering principle proposed by the intelligent question-answering scheme provided by the application. Of course, the terminal device may also run other various applications, such as image processing applications, multimedia playing applications, navigation applications, etc.

When the computer device includes a server, the server may establish a communication connection with an application (or client) providing the intelligent question-answering service to provide support services such as a data computing service and a data storage service to the application. Alternatively, the server may be a separate server, or may be a server cluster or a distributed system formed by a plurality of servers. The server may include one or more of a physical server and a cloud server. The cloud server may be used to provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDNs (Content Delivery Network, content delivery networks), and basic cloud services such as big data and artificial intelligence platforms, which are not particularly limited in this embodiment of the present application.

Based on the principle of the above-mentioned intelligent question-answering scheme, the application specifically proposes an intelligent question-answering method, which can still be executed by the above-mentioned computer device. Referring to fig. 4, fig. 4 is a schematic flow chart of the intelligent question-answering method. As shown in fig. 4, the method at least includes steps S401 to S404:

S401, acquiring question information from a question and answer request in response to the question and answer request.

In particular embodiments, the question and answer request may be a request to trigger the question client to generate and send to the computer device after the question object enters the question information within the question client. Optionally, the question and answer request may be automatically generated by acquiring, in real time, question information input by the question object after the computer device detects that the question object performs the question operation on the relevant page. The question and answer request at least comprises question information input by a question object, and the input mode of the question information can include one or more of voice input, keyboard input, gesture input and the like. In addition, the question information includes at least one or more text characters. That is, in practical application, the question information may be plain text content, or may be picture content including text content.

S402, respectively calling a plurality of intelligent question-answering models to intelligently answer the question information, and obtaining candidate answer information obtained by each intelligent question-answering model aiming at the question information, wherein the candidate answer information comprises one or more text fragments.

In a specific embodiment, the intelligent question-answering model is obtained by adopting a deep learning technology to perform model optimization on a neural network model constructed based on an artificial neural network. By way of example, the intelligent question-answering model may be a natural language processing model with a large number of model parameters and/or complex network structures, and such a model may be referred to as a large language model (LLM, large Language Model). Experiments made by persons of ordinary skill in the art to which the present application pertains have demonstrated that large language models generally have good accuracy and processing efficiency when used to handle large-scale text data and/or complex semantic problems. In the embodiment of the application, the large language model can comprise various general or special large models in the field of natural language processing, such as chatgpt-3.5, chatgpt-4, chatglm2-6b, senschat, MOSS and the like.

Because the same neural network model can obtain different answer results when answering the same question information for a plurality of times, the application adopts a plurality of intelligent question-answering models mainly for enriching candidates of the answer information so as to promote the matching degree between the finally determined answer information and the question information as much as possible. Therefore, on the basis, it is easy to understand that the plurality of intelligent question-answering models in the embodiment of the application can specifically include the same intelligent question-answering model or different intelligent question-answering models. The different intelligent question-answering models may specifically include an intelligent question-answering model obtained by performing model optimization on the same neural network model by using different training data, an intelligent question-answering model obtained by performing model optimization on different neural network models based on the same training data, and an intelligent question-answering model obtained by performing model optimization on different neural network models by using different training data.

In practical application, one intelligent question-answer model can be used for intelligently answering question information to obtain one or more candidate answer information, so that the computer equipment can finally obtain a plurality of candidate answer information by utilizing a plurality of intelligent question-answer models. Wherein each candidate answer information comprises one or more text segments, each segment being composed of at least one text character. Optionally, the text segment included in each candidate answer information may be obtained by performing text cutting on the text sequence included in the candidate answer information by using one or more of the following text cutting modes (1) - (3) by using a computer device (or an intelligent question-answering model):

(1) Text cuts are made according to the target character length (e.g., 10 characters). The target character length may be determined based on the maximum feature length that the computer device or the corresponding intelligent question-answering model can process when features are embedded, or may be set according to the related service requirements of the intelligent question-answering service, which is not limited in the embodiment of the present application.

As an exemplary processing manner, if the character length of the remaining text sequence is detected to be less than the target character length in the text cutting process, the remaining text sequence may be directly used as a text segment. Alternatively, the remaining text sequences may be character padded to obtain a character length as the target character length. In this case, the character length of each text segment may be smaller, and the device resources (such as computing resources and storage resources) that need to be adopted by the computer device when processing each text segment with smaller character length are generally different, so as to reduce the occurrence of frequent calling and releasing of related device resources by the computer device, thereby being beneficial to maintaining the functional stability of the computer device, and further improving the data processing efficiency of the computer device to a certain extent.

(2) And cutting the text according to punctuation marks adopted by the sentence breaking. Punctuation marks may include, but are not limited to, commas, periods, semicolons, dashes, and the like. Because each character sequence separated by punctuation marks in a text sequence usually has relatively complete semantic information, the method can cut candidate answer information so that the semantic information of each text segment is relatively complete, computer equipment can refer to the semantic information of each text segment, candidate text segments with relatively high matching degree with question information are selected, and finally the matching degree between answer information generated by the computer equipment and the question information is also ensured to a certain extent.

(3) Text cutting is performed according to text phrase and/or phrase pairs. Wherein, the text phrase and/or phrase refers to a text character sequence consisting of N text characters which can be used for expressing certain semantics, N is usually an integer greater than 1, and in individual cases, N can also be 1. Because the text phrase or phrase can express certain semantic information, and the character length of the text phrase or phrase is generally shorter, the method is used for cutting the candidate answer information, so that the computer equipment is not only beneficial to selecting the candidate text fragments with higher matching degree with the question information, but also can process each text fragment when the computer equipment has fewer equipment resources, the performance requirement of the embodiment of the application on the computer equipment is reduced, and the intelligent question-answering efficiency of the computer equipment is improved to a certain extent.

S403, determining the matching degree between each text segment and the question information, and taking the text segments with the matching degree larger than or equal to the matching degree threshold as candidate text segments to obtain one or more candidate text segments.

In one embodiment, the one or more candidate text snippets are selected by the computer device from among the text snippets contained in a plurality of candidate answer information, each candidate answer information including at least one text snippet. It should be noted that, in the embodiments of the present application, a plurality refers to at least two. In addition, the plurality of candidate answer information may be obtained after the computer device uses the plurality of intelligent question-answer models to intelligently answer the question information respectively. Specifically, the computer device may be obtained by selecting one or more candidate answer information from the candidate answer information obtained by any intelligent question-answer model after intelligently answering the question information.

Since each candidate answer information includes at least one text segment, the plurality of candidate answer information may include a plurality of text segments, and the computer device may calculate a degree of matching between each text segment of the plurality of text segments and the question information, respectively, and select one or more candidate text segments from the plurality of text segments based on a magnitude relationship between the degree of matching and a degree of matching threshold. It should be noted that, the matching degree mentioned herein may include, but is not limited to, one or two of a semantic matching degree and a text matching degree, and may also be a segment matching degree determined based on the semantic matching degree and the text matching degree together, which is not limited in this embodiment of the present application.

In one embodiment, the match threshold may be a fixed value or a non-fixed value that is dynamically determined based on the respective determined match. When the matching degree threshold is a fixed value, the matching degree threshold may be preset in the computer device, or may be obtained from a question request or other received processing requests by the computer device. When the matching degree threshold is a non-fixed value, the computer device may alternatively use, as the matching degree threshold, the matching degree located at the X-th bit after being arranged in order from large to small (or from small to large) among the determined respective matching degrees, and X is illustratively a positive integer. Of course, the computer device may also perform some mathematical operation (e.g., arithmetic averaging, weighted averaging, etc.) on each degree of matching to determine the degree of matching threshold based on the value indicated by the result of the operation. The matching degree threshold may be in positive correlation with a numerical value indicated by the operation result.

In one implementation of this embodiment, the number of matching degree thresholds may be one. At this time, the criteria for screening the candidate text fragments from the candidate answer information obtained by each intelligent question-answering model by the computer device is consistent, so that the matching degree between each candidate text fragment selected by the computer device and the question information is higher, and finally, the matching degree between the answer information containing at least one candidate text fragment and the question information is also higher.

In yet another implementation of this embodiment, the number of matching degree thresholds may also be plural. In this case, each intelligent question-answering model may correspond to a matching degree threshold, so that the computer device finally screens one or more candidate text fragments, and there are text fragments in the candidate answer information of each intelligent question-answering model. The candidate answer information obtained by different intelligent question and answer models can be obtained by carrying out semantic understanding on the question information at different angles (or knowledge fields) and then carrying out answer. Then, setting corresponding matching degree thresholds for different intelligent question-answering models can enable one or more candidate text fragments obtained by the computer equipment to indicate semantic understanding results of multiple angles as far as possible. Based on the above, when the generated answer information contains at least one candidate text segment, the answer information can be generated by comprehensively referencing semantic understanding of a plurality of angles by the computer equipment, so that the matching degree between the semantic information of the answer information and the question intention of the question information is improved to a certain extent, the answer information is more in line with the question requirement of the question object, and the question and answer experience of the question object can be effectively improved.

S404, generating answer information of the question information, wherein the answer information comprises at least one candidate text segment.

In one embodiment, the answer information may be obtained after the computer device text-completions at least one of the one or more candidate text segments. The at least one candidate text segment may be selected randomly, or may be selected according to the matching degree between each candidate text segment and the question information, for example, a candidate text segment with the matching degree located in the first Y (Y is a positive integer) bits is selected. Further, text completion may be implemented in accordance with context information associated with the candidate text snippet.

In one implementation of the present embodiment, the context information may include spatial context information. For ease of understanding, text completion of a candidate text segment by a computer device will be described. In this case, the text completion may be performed in the following manner: and inquiring one or more text fragments belonging to one candidate answer information together with the candidate text fragments from a plurality of text fragments corresponding to each candidate answer information, and splicing the inquired text fragments with the candidate text fragments according to the arrangement sequence of the text fragments in the candidate answer information. Based on this, it is easy to see that, in practical application of the embodiment of the present application, the computer device may also first select a target candidate text segment from one or more candidate text segments, determine candidate answer information to which the target candidate text segment belongs, and further use the candidate answer information as answer information of the question information. Wherein the number of target candidate text segments may be one or more, but is typically one. In addition, the selection manner of the target candidate text segment may refer to the selection manner of at least one candidate text segment in the above embodiment, which is not described herein again.

In yet another implementation of the present embodiment, the context information may also include semantic context information. In this case, the computer device may specifically perform intent recognition on the question information to obtain a question intent of the question information, and perform semantic recognition on each text segment corresponding to each candidate answer information to obtain semantic information of each text segment, so as to generate an answer sentence based on the semantic information of each text segment, and the at least one candidate text segment. Wherein the computer device may generate an answer sentence with reference to one candidate text segment, and in practice, the answer sentence generated by the computer device with reference to a different candidate text segment may be the same.

As an exemplary implementation manner, when the computer device generates an answer sentence with reference to any candidate text segment, the semantic information of the candidate text segment may be determined first, then a text segment whose semantic information and the semantic information of the candidate text segment can form logical fair semantic context information is selected from a plurality of text segments, and the candidate text segment and the text segment are spliced according to the semantic context information, so as to obtain the answer information. The computer device may also limit the length of the answer information in order to avoid that the answer information is too lengthy to cause the questioning object to be understood quickly. For example, in the process of splicing, if the length of the text content obtained by splicing is detected to be greater than or equal to the preset length, the text content obtained by splicing is taken as an answer sentence.

In this case, if the degree of correlation between the semantic information of the answer sentence and the question intention is greater than or equal to the degree of correlation threshold, the answer sentence is taken as the answer information. If the correlation degree between the semantic information of the answer sentence and the questioning intention is smaller than the correlation degree threshold value, a new answer sentence is regenerated, or the relevant processing logic for text completion according to the space context information is triggered and executed, so that the answer information of the questioning information is obtained. The method for setting the correlation threshold may refer to the principle of the method for determining the matching threshold, which is not described herein. In addition, it is easy to understand that, since the answer information in the embodiment is generated by referring to the semantic information between the candidate text segment and each text segment, the semantic information expressed by the answer information can have a higher degree of matching with the semantic of the question information to a certain extent, so that the answer information can be more accurate.

It should be further noted that, in the two foregoing embodiments, when the computer device queries a text segment for text completion, in addition to the requirement of the context information, an accessory condition that needs to be satisfied by the text segment may be added. Optionally, the accessory conditions are as follows: the relevance between the text segment and the knowledge field related to the question information is required to be greater than or equal to a relevance threshold value, so that the finally generated answer information can be presented in a simpler and more concise mode while having higher matching degree with the question information. The method for setting the correlation threshold may refer to the principle of the method for determining the matching degree threshold, which is not described herein.

To facilitate a clearer understanding of the intelligent question-answering method shown in fig. 2 described above. The following describes an exemplary manner of responding to a question-answer request received by a computer device in connection with a specific example and a flow as shown in fig. 5. In this example, the number of intelligent question-answering models is three.

Based on fig. 5, after receiving the question request, the computer device may obtain the question information from the question request, and further answer the question information by using LLM1, LLM2, and LLM3, so as to obtain answer 1 corresponding to LLM1, answer 2 corresponding to LLM2, and answer 3 corresponding to LLM 3. Further, the computer device performs text segmentation on each answer to obtain a plurality of text blocks, and performs Embedding on each text block to obtain a feature vector corresponding to each text block. In addition, the computer device performs Embedding on the question information to obtain a feature vector (such as the question vector in fig. 5) corresponding to the question information, and then determines a relevant text block from the plurality of text blocks by performing vector matching on the question vector and the feature vector corresponding to each text block (see, for details, the query matching flow shown in fig. 5). The relevant text block herein refers to: text blocks having a degree of match between the feature vector and the question vector greater than or equal to a degree of match threshold. Finally, the computer device may perform text completion on the relevant text block to obtain a final answer, and may further feed the answer back to the sender of the question request. Since the number of text blocks is usually multiple, in order to facilitate subsequent vector matching with a higher rate, optionally, the computer device may perform Embedding on each text block, and temporarily store the feature vector corresponding to each text block in the vector library (see, in particular, the answer in-library flow shown in fig. 5).

In the method, after the computer equipment acquires the questioning information, a plurality of intelligent questioning and answering models are adopted to intelligently answer the questioning information respectively, answers obtained by each intelligent questioning and answering model aiming at the questioning information are used as candidate answer information of the questioning information, and then the answer information is generated based on text fragments with the matching degree with the questioning information being greater than or equal to a matching degree threshold value in text fragments contained in each candidate answer information, so that the answer information in the method is determined by comprehensively referencing the answers of the plurality of intelligent questioning and answering models by the computer equipment. Because in the reality application of artificial intelligence field, the question-answering ability of each intelligent question-answering model to the same knowledge field is uneven, and the knowledge field that different question information relates to may be various, then, utilize a plurality of intelligent question-answering models to answer intelligently, just also used the question-answering ability of different ability levels and/or different knowledge fields, therefore confirm answer information based on the answer of each intelligent question-answering model, just can guarantee the matching degree between this answer information and the question information to a certain extent for answer information can accord with the question intention of the question object more, and then promote the question-answering experience of the question object.

Based on the principle of the above-mentioned intelligent question-answering scheme and the intelligent question-answering method shown in fig. 4, the present application also proposes yet another intelligent question-answering method, which can still be performed by the above-mentioned computer device. Specifically, referring to fig. 6, fig. 6 is a schematic flow chart of the intelligent question-answering method. As shown in fig. 6, the method at least includes steps S601 to S606:

s601, acquiring questioning information from a questioning and answering request in response to the questioning and answering request.

In an embodiment, the specific implementation manner of step S601 may refer to the related embodiment of step S401, which is not described herein.

S602, respectively calling a plurality of intelligent question-answering models to intelligently answer question information, and obtaining candidate answer information obtained by each intelligent question-answering model aiming at the question information, wherein the candidate answer information comprises one or more text fragments.

In one embodiment, the computer device may pre-select the intelligent question-answering model from a plurality of different knowledge domains to obtain a plurality of intelligent question-answering models, and then use the plurality of intelligent question-answering models to perform question-answering processing when receiving the question information.

In yet another embodiment, the computer device may also first determine the intelligent question-answering model to call based on the knowledge domain to which the question information pertains. Specifically, the computer device may first obtain one or more question keywords from the question information, determine a knowledge domain associated with each question keyword, obtain one or more knowledge domains related to the question information, and further select at least one intelligent question-answer model in each knowledge domain related to the question information, so as to obtain a plurality of intelligent question-answer models. In a specific implementation, since knowledge domains which are good for processing by different intelligent question-answering models may be different, and question-answering capacities in the same knowledge domain may also be different, in order to enable the application to achieve higher-quality answer to question information in each knowledge domain, the computer device will select at least one intelligent question-answering model in each knowledge domain to which the question information relates, so as to obtain a plurality of intelligent question-answering models. If the knowledge domain related to the questioning information is one, the computer device may select a plurality of intelligent questioning and answering models in the knowledge domain. If the knowledge domain related to the question information is more than one, the computer equipment can select an intelligent question-answering model in each knowledge domain respectively.

The questioning information may include one or more text characters, and the questioning keywords are formed by at least one text character, so that the computer device may extract one or more questioning keywords from any questioning information. The knowledge domains may be partitioned by the computer device according to the usage scenario of the intelligent question-answering method, and the knowledge domains may specifically include a plurality of levels, such as a primary knowledge domain, a secondary knowledge domain, and the like. The division granularity of the second-level knowledge domain is finer than that of the first-level knowledge domain, and the knowledge domain related to the question information can be precisely determined by fine division, so that answer information of the question information is promoted to have higher accuracy. Among them, the primary knowledge domain may be, for example, a medicine domain, a finance domain, a digital domain, an art domain, a life domain, and the like. The second knowledge domain under the first knowledge domain of the medicine domain may include, but is not limited to, a medicine purchasing domain, a medicine consultation domain, a medical knowledge science popularization domain, a diagnosis and treatment reservation domain, and the like.

In one implementation, to quickly determine the knowledge domain to which the questioning information pertains, the computer device may establish an access connection with one or more keyword libraries, each keyword library may contain one or more keywords, and each keyword library is associated with at least one knowledge domain. Then, when the computer device detects that the question keyword exists in a certain keyword library, the knowledge domain associated with the keyword library is used as the knowledge domain associated with the question keyword. Wherein, the question keywords exist in the keyword library means that: the keyword library contains the same keywords as the question keyword.

In yet another implementation, the computer device may also employ a text classification model to identify knowledge domains to which each question keyword relates. The recognition capability of the text classification model may be obtained by performing supervised training or unsupervised training with related training data, which is not described in detail in the embodiments of the present application.

S603, performing text word recognition processing on the candidate answer information to obtain target text words included in the candidate answer information, wherein the target text words refer to text words associated with knowledge fields related to the candidate answer information.

In a specific embodiment, the knowledge domain to which the candidate answer information relates may include: and generating knowledge fields associated with the intelligent question-answering model of the candidate answer information. The text word recognition processing may include word segmentation processing of candidate answer information, and word meaning recognition processing and/or word feature extraction processing may be performed on each text word obtained by the word segmentation. The word meaning of the text word can be obtained by carrying out word meaning identification processing on the text word, and whether the text word is used as a target text word can be further determined based on the similarity between the word meaning and the meaning of the knowledge field related to the candidate answer information. Similarly, the feature vector of the text word can be obtained by extracting the word feature of the text word, and whether the text word is used as the target text word can be determined based on the similarity between the feature expression and the feature vector of the knowledge field related to the candidate answer information.

It should be noted that, in the embodiment of the present application, one or more target text words may be determined, and when the determined target text word is one, the calculation burden of the computer device in the subsequent step may be reduced, so that the overall efficiency of the intelligent question-answering is improved to a certain extent. When the target text words are multiple, the computer device is helped to determine the text segment capable of accurately expressing the subject semantics of the candidate answer information with a greater probability in step S605, and further, the accuracy of the answer information finally generated by the computer device can be improved.

S604, based on the text words contained in the candidate answer information, carrying out semantic filling processing on the target text words in the knowledge field related to the target text words to obtain a filled text.

In a specific embodiment, the text words adopted by the computer device when performing semantic filling on the target text words are part of text words in the candidate answer information, so that the filled text can have smaller character length compared with the candidate answer information, the calculation load of the computer device when executing subsequent steps is reduced, and the performance stability of the computer device is maintained.

In a specific implementation manner, the computer device may determine one or more semantic related words of the target text word and semantic dependency relationship between each semantic related word and the target text word from the text words contained in the candidate answer information by executing semantic dependency analysis of the target text word in the candidate answer information, so that the computer device may perform text splicing on each semantic related word and the target text word according to the semantic dependency relationship of each semantic related word, and finally obtain the filled text.

If the computer equipment detects that the semantics expressed by the two text words are mutually influenced, the two text words are considered to have semantic dependency relationship. In practical application, the ambiguous expression of the ambiguous word extracted by the computer equipment can be effectively eliminated by analyzing the semantic dependency relationship, so that the semantic information expressed by the text word in the candidate answer information can be accurately obtained. In addition, context association exists among text words with semantic dependency relationship, so that the computer equipment can splice each text word with semantic dependency relationship according to the text words, and a text segment with smoother semantic meaning is obtained. The splicing can be realized according to the arrangement sequence of each text word in the candidate answer information.

In yet another specific implementation manner, the computer device may also determine, from the candidate answer information, a text word adjacent to the target text word, so as to obtain an adjacent text word of the target text word, and then sequentially splice the adjacent text words to the target text word in an adjacent order, so as to obtain the filled text.

And S605, if the semantic integrity of the filled text is greater than or equal to a preset threshold value, using the filled text as a text segment of the candidate answer information.

In one embodiment, in order to enable the selected text segment to fully express semantic information related to the knowledge domain associated with the target text word in the candidate answer information, the computer device may use the filled text as a text segment of the candidate answer information when the semantic integrity of the obtained filled text is greater than or equal to a preset threshold. And when the semantic integrity of the filled text is smaller than a preset threshold, continuing filling (or called secondary filling) based on the filled text, or reselecting other target text words to perform text filling until the filled text with the semantic integrity larger than or equal to the preset threshold is obtained.

It is worth mentioning that the filling mode adopted in the secondary filling can be different from the filling mode adopted in the first filling so as to enrich the obtaining mode of the filling text, thereby increasing the probability and the rate of obtaining the filling text with the semantic integrity larger than the preset threshold value by the computer equipment. Furthermore, it will be appreciated that in practical applications, the computer device may select one or more text segments in each candidate answer information based on the above-described filling manner according to the service requirements.

Therefore, by determining the text segment in the above manner, on the basis of reducing the calculation load of the computer equipment, the semantic integrity of the text segment selected from the candidate answer information is ensured, so that the text segment is determined to be enough to express the gist meaning of the candidate answer information, and the accuracy of the answer information generated by the computer equipment is further improved.

S606, determining the matching degree between each text segment and the question information, and taking the text segments with the matching degree larger than or equal to the matching degree threshold as candidate text segments to obtain one or more candidate text segments.

In one embodiment, when the computer device determines the matching degree between any text segment and the question information, the computer device can comprehensively determine the matching degree between the question information and the text segment based on the knowledge domain association degree between the question information and the text segment and the text similarity between the question information and the text segment, so that the computer device can comprehensively determine the matching degree from multiple dimensions, and the referential of the matching degree is ensured to a certain extent.

Specifically, when calculating the degree of association of knowledge domains, the computer device may first perform knowledge domain analysis on the question information to obtain one or more knowledge domains related to the question information, and further calculate the degree of association (i.e., the degree of association of knowledge domains) between each knowledge domain and the knowledge domain associated with the target text word in the text segment, so as to obtain the degree of association of each corresponding knowledge domain. When knowledge domain analysis is performed on the questioning information, one or more questioning keywords in the questioning information may be obtained first, where the questioning keywords may include text words with a frequency (i.e., word frequency) greater than or equal to a preset word frequency in the questioning information, and may further include text words obtained by random interception and/or text words of a preset type (which may be carried by a questioning request), which is not limited in this embodiment of the present application.

After obtaining the questioning keywords, the computer device may send a tracing request of each questioning keyword to the text word tracing device, where the text word tracing device may query the database for source information of each questioning keyword after receiving the request, and generate tracing results of each questioning keyword after determining a knowledge domain to which the questioning keyword belongs based on the source information of each questioning keyword. Wherein the source information includes one or more of article information and naming person information. The computer device may then receive the traceability results of the respective question keywords, to take the knowledge domain indicated by the traceability results of the respective question keywords as the knowledge domain to which the question information relates. The knowledge field of the question keywords is determined by adopting the text word tracing equipment, the effect that a plurality of equipment work division cooperation is realized to realize the intelligent question-answering function can be achieved, the work division cooperation can effectively reduce the complexity of processing logic executed by each equipment in the intelligent question-answering process, and further reduce the performance requirement on the equipment, so that the intelligent question-answering method can be suitable for equipment with lower performance, and the application field of the intelligent question-answering method is expanded to a certain extent.

In addition, when the computer equipment calculates the text similarity, the vocabulary similarity and/or the semantic similarity between the text segment and the question information can be calculated first, so that the text similarity is obtained. For example, text similarity may be positively correlated with both lexical similarity and semantic similarity.

In yet another embodiment, when determining the matching degree between any text segment and the question information, the computer device may further perform feature extraction processing under at least one feature dimension on the text segment and the question information, so as to obtain a feature matrix of the text segment under each feature dimension, and a feature matrix of the question information under each feature dimension, and finally determine the matching degree between the text segment and the question information according to each feature matrix corresponding to the question information and each feature matrix corresponding to the text segment. Wherein, alternatively, the feature extraction processing of the text segment and the question information may be implemented using the same feature extraction network, and the matching degree between the text segment and the question information may be exemplarily determined with reference to the principle as shown in fig. 7.

Based on fig. 7, after obtaining the feature matrix of the text segment in each feature dimension (i.e. feature dimension 1 to feature dimension P shown in fig. 7, where P is a positive integer), and the feature matrix of the question information in each feature dimension, the computer device may calculate the feature similarity between the question information and the feature matrix of the text segment in the same feature dimension, so as to obtain at least one feature similarity (e.g. feature similarity 1 to feature similarity P, p=p shown in fig. 7). It can be seen that one feature similarity corresponds to one feature dimension. Finally, the computer device may determine a degree of matching between the text segment and the question information based on the similarity of each feature and the similarity weight of each feature dimension.

The computer device may perform a weighted operation (e.g., weighted average) on each feature similarity to obtain a target feature similarity, and further determine a matching degree between the text segment and the question information according to a principle that the matching degree and the target feature similarity are positively correlated. In addition, in a specific implementation, the similarity weight may be related to the feature dimension, and each similarity weight may be a preset fixed value, or may be a variable value dynamically adjusted according to a service requirement, which is not limited in the embodiment of the present application.

In yet another embodiment, when determining the matching degree between any text segment and the question information, the computer device may further perform intent recognition on the question information to obtain an intent feature vector of the question information, and perform semantic recognition on the text segment to obtain a semantic feature vector of the text segment, so as to use the matching degree between the intent feature vector and the semantic feature vector as the matching degree between the question information and the text segment.

In yet another embodiment, when determining the matching degree between any text segment and the question information, the computer device may further perform at least two types of matching processing on the question information and the text segment to obtain at least two matching features, so as to finally predict and obtain the matching degree between the text segment and the question information according to the at least two matching features. Wherein, one matching feature is obtained by adopting one type of matching processing, and at least two types comprise at least two types of interactive matching type, semantic matching type and text matching type. Also, for example, the computer device may predict the degree of matching between the text segment and the question information with reference to the principle shown in fig. 8.

Wherein, the matching feature obtained based on the matching processing of the interactive matching type (or called as interactive matching) can be called as interactive feature, and the interactive feature can be specifically used for indicating the relevance between the question information and the text segment. The so-called interactive matching can be understood as: when the computer equipment extracts the characteristic of the first character of the text segment, the word vector corresponding to the first character can be compared with the word vector of each word in the question information, so that the correlation between the first character of the text segment and the question information is extracted, and the computer equipment can generate the interaction characteristic based on the correlation corresponding to each character.

Further, the matching features resulting from the matching process based on the semantic matching type may be referred to as semantic difference features. The semantic difference feature may be specifically used to indicate a degree of difference between the semantics of the text segment and the semantics of the question information. The semantic difference feature may be obtained by performing a vector difference process on the word vector of the question information and the word vector of the text segment. The matching features resulting from the matching process based on the text matching type may be referred to as text similarity features. Text similarity features may be specifically used to indicate the similarity of text content between a text segment and question information, and may be obtained by cosine similarity calculation of word vectors of question information and word vectors of the text segment, for example.

In this case, the matching degree between the question information and the text segment is determined by comprehensively referring to the correlation, the semantic difference and the similarity of the text content between the question information and the text segment, so that the finally determined matching degree has higher reference value, and further, the computer equipment determines the candidate text segment based on the matching degree and the matching degree threshold value and has higher matching degree with the question intention of the question information.

S607, generating answer information of the question information, wherein the answer information comprises at least one candidate text segment.

In an embodiment, the specific implementation manner of step S607 may refer to the related embodiment of step S404, which is not described herein.

In the method, after acquiring the questioning information, the computer equipment adopts a plurality of intelligent question-answering models to intelligently answer the questioning information respectively, answers obtained by each intelligent question-answering model aiming at the questioning information are used as candidate answer information of the questioning information, and further, based on text fragments contained in each candidate answer information, the matching degree between the candidate answer information and the questioning information is larger than or equal to a matching degree threshold value, and the answer information is generated. The matching degree between the questioning information and the text fragments is determined by comprehensively referring to the correlation, semantic difference and similarity of text contents between the questioning information and the text fragments, so that the finally determined matching degree has higher reference value, and further, the candidate text fragments determined by the computer equipment and the questioning intention of the questioning information have higher matching degree. In addition, after the questioning information is acquired, the computer equipment automatically calls a plurality of intelligent questioning and answering models to answer the questioning information, so that when a related questioning object wants to comprehensively reference answers of different intelligent questioning and answering models, the related questioning object does not need to manually select different intelligent questioning and answering models to repeatedly ask, and further the questioning object does not need to manually screen final answer information from each answer, thereby effectively improving the questioning and answering experience of the questioning object. That is, the questioning object has no perception to the internal logic of the intelligent questioning and answering, so that when the questioning object uses the related intelligent questioning and answering product, no additional use teaching is needed, and the manpower input and decision effort are effectively reduced.

Based on the above related embodiments of the intelligent question-answering method of fig. 4 and fig. 6, the present application also discloses an intelligent question-answering device for executing the intelligent question-answering method shown in fig. 4 and fig. 6. Wherein the apparatus may be a computer program (comprising program code) running in the above mentioned computer device. Referring to fig. 9, the intelligent question answering apparatus may at least include: a response unit 901, a call unit 902, a determination unit 903, and a generation unit 904. Wherein:

a response unit 901, configured to obtain question information from a question-answer request in response to the question-answer request;

the calling unit 902 is configured to call a plurality of intelligent question-answering models respectively to intelligently answer the question information, so as to obtain candidate answer information obtained by each intelligent question-answering model for the question information, where the candidate answer information includes one or more text segments;

a determining unit 903, configured to determine a matching degree between each text segment and the question information, and use the text segment with the matching degree greater than or equal to a matching degree threshold as a candidate text segment, so as to obtain one or more candidate text segments;

a generating unit 904, configured to generate answer information of the question information, where the answer information includes at least one candidate text segment.

In one embodiment, the calling unit 902 may also be configured to perform:

acquiring one or more question keywords from the question information, determining the knowledge domain associated with each question keyword, and obtaining one or more knowledge domains related to the question information;

and respectively selecting at least one intelligent question-answering model in each knowledge field related to the question information to obtain a plurality of intelligent question-answering models.

In yet another embodiment, the calling unit 902 may be further configured to perform:

performing text word recognition processing on any candidate answer information to obtain target text words included in the any candidate answer information, wherein the target text words refer to text words associated with knowledge fields related to the any candidate answer information;

based on text words contained in any candidate answer information, carrying out semantic filling processing on the target text words in the knowledge domain related to the target text words to obtain filled texts;

and if the semantic integrity of the filled text is greater than or equal to a preset threshold value, taking the filled text as a text segment of any candidate answer information.

In yet another embodiment, the invoking unit 902 is further configured to, when performing semantic filling processing on the target text word in the knowledge domain related to the target text word based on the text word included in the arbitrary candidate answer information to obtain a filled text, specifically perform:

Determining one or more semantic association words of the target text word and semantic dependency relationship between each semantic association word and the target text word from the text words contained in any candidate answer information by executing semantic dependency analysis of the target text word in the any candidate answer information;

and performing text splicing on each semantic association word and the target text word according to the semantic dependency relationship of each semantic association word to obtain the filling text.

In yet another embodiment, the determining unit 903 may be specifically configured to perform:

carrying out knowledge domain analysis on the questioning information to obtain one or more knowledge domains related to the questioning information;

for any text segment, calculating the association degree between the knowledge domain associated with the target text word in the any text segment and each knowledge domain related to the question information;

calculating text similarity between any text segment and the question information, wherein the text similarity comprises one or two of vocabulary similarity and semantic similarity;

And determining the matching degree between any text segment and the question information according to the relevance and the text similarity.

In yet another embodiment, the determining unit 903, when configured to perform knowledge domain analysis on the question information to obtain one or more knowledge domains related to the question information, may be specifically configured to perform:

acquiring one or more question keywords in the question information, wherein the question keywords comprise text words with word frequency larger than a preset word frequency in the question information;

sending a tracing request of each questioning keyword to text word tracing equipment so that the text word tracing equipment inquires source information of each questioning keyword from a database, and generating tracing results of each questioning keyword after determining a knowledge field to which the questioning keyword belongs based on the source information of each questioning keyword; wherein the source information includes one or more of article information and naming person information;

and receiving the tracing results of the questioning keywords, and taking the knowledge domain indicated by the tracing results of the questioning keywords as the knowledge domain related to the questioning information.

performing feature extraction processing on any text segment under at least one feature dimension to obtain at least one feature matrix of the any text segment, wherein one feature matrix corresponds to one feature dimension;

acquiring a feature matrix of the questioning information under the at least one feature dimension;

and determining the matching degree between any text segment and the question information according to the feature matrix of the question information in each feature dimension and the feature matrix corresponding to any text segment in each feature dimension.

In yet another embodiment, the determining unit 903 may be specifically configured to perform, when determining the matching degree between the arbitrary text segment and the question information according to the feature matrix of the question information in each feature dimension and the feature matrix corresponding to the arbitrary text segment in each feature dimension:

calculating feature similarity between the questioning information and the feature matrix of any text segment under the same feature dimension to obtain at least one feature similarity; wherein, one feature similarity corresponds to one feature dimension;

And determining the matching degree between any text segment and the question information based on the at least one feature similarity and the similarity weight of each feature dimension.

In yet another embodiment, the determining unit 903, when configured to determine a degree of matching between each text segment and the question information, may be further specifically configured to perform:

carrying out at least two types of matching processing on the question information and any text segment to obtain at least two matching features; wherein, one matching feature is obtained by adopting one type of matching processing, and the at least two types comprise at least two of interactive matching type, semantic matching type and text matching type;

and predicting the matching degree between any text segment and the question information according to the at least two matching features.

In yet another embodiment, the generating unit 904 may be specifically configured to perform:

selecting a target candidate text segment from the one or more candidate text segments;

and taking the candidate answer information of the target candidate text segment as the answer information of the question information.

In yet another embodiment, the generating unit 904 may be further specifically configured to perform:

carrying out intention recognition on the questioning information to obtain the questioning intention of the questioning information;

carrying out semantic recognition on the text fragments corresponding to each candidate answer information to obtain semantic information of each text fragment;

generating an answer sentence based on the semantic information of each text segment, each text segment and at least one candidate text segment;

and if the correlation degree between the semantic information of the answer sentence and the questioning intention is greater than or equal to a correlation degree threshold value, taking the answer sentence as the answer information.

According to one embodiment of the present application, the steps in the intelligent question-answering method shown in fig. 4 and 6 may be performed by the respective units in the intelligent question-answering apparatus shown in fig. 9. For example, step S401 in fig. 4 may be performed by the response unit 901 in the intelligent question-answering apparatus, step S402 may be performed by the call unit 902 in the intelligent question-answering apparatus, step S403 may be performed by the determination unit 903 in the intelligent question-answering apparatus, and step S404 may be performed by the generation unit 904 in the intelligent question-answering apparatus. As another example, step S601 in fig. 6 may be performed by the response unit 901 in the intelligent question-answering apparatus, steps S602 to S604 may be performed by the calling unit 902 in the intelligent question-answering apparatus, step S605 may be performed by the determining unit 903 in the intelligent question-answering apparatus, and step S606 may be performed by the generating unit 904 in the intelligent question-answering apparatus.

According to another embodiment of the present application, each unit in the intelligent question-answering device shown in fig. 9 is divided based on a logic function, and each unit may be respectively or completely combined into one or several other units to form the intelligent question-answering device, or some unit(s) thereof may be further split into a plurality of units with smaller functions to form the intelligent question-answering device, which can achieve the same operation without affecting the implementation of the technical effects of the embodiments of the present application. In other embodiments of the present application, the intelligent question and answer-based device may also include other units, and in practical applications, these functions may also be implemented with assistance of other units, and may be implemented by cooperation of multiple units.

According to another embodiment of the present application, the intelligent question and answer apparatus shown in fig. 9 may be constructed by running a computer program (including program code) capable of executing the steps involved in the method shown in fig. 4 or 6 on a general-purpose computing device such as a computer including a Central Processing Unit (CPU), a random access storage medium (RAM), a read only storage medium (ROM), etc., processing elements and storage elements, and implementing the intelligent question and answer method of the embodiments of the present application. The computer program may be recorded on, for example, a computer storage medium, and loaded into and run in the above-described computer apparatus through the computer storage medium.

In this embodiment of the present application, the calling unit 902 in the intelligent question-answering device calls a plurality of intelligent question-answering models to respectively answer question information intelligently, and adopts the determining unit 903 to select, from text segments included in answer results obtained by each intelligent question-answering model for the question information, text segments with matching degree with the question information being greater than or equal to a matching degree threshold value, so as to finally generate answer information based on the selected text segments through the generating unit 904. Since the question-answering capability of each intelligent question-answering model in the real application may be irregular for the same knowledge domain, and the knowledge domains related to different question information may be various, the calling unit 902 performs intelligent answer by using a plurality of intelligent question-answering models, so that the question-answering capability of different capability levels and/or different knowledge domains may be used. On this basis, the generating unit 904 determines answer information based on answers of the intelligent question-answer models, so that the generated answer information is obtained after comprehensively referring to answers of a plurality of knowledge fields, and accordingly matching degree between the answer information and the question information is effectively guaranteed, the answer information can be more in line with the question intention of a question object, and the question-answer experience of the question object is further improved.

Based on the above description of the method embodiment and the apparatus embodiment, the embodiment of the present application further provides a computer device, please refer to fig. 10. The computer device includes at least a processor 1001 and a storage medium 1002, and the processor 1001 and the storage medium 1002 of the computer device may be connected by a bus or other means. Among them, the storage medium 1002 mentioned above is a memory device in a computer device for storing programs and data. It is understood that the storage medium 1002 herein may include a built-in storage medium in a computer device, and of course may include an extended storage medium supported by a computer device. The storage medium 1002 provides storage space that stores the operating system of the computer device. Also stored in this memory space are one or more computer programs, which may be one or more program codes, adapted to be loaded and executed by the processor 1001.

The storage medium herein may be a high-speed RAM memory or a non-volatile memory (non-volatile memory), such as at least one disk memory; optionally, at least one storage medium located remotely from the processor. The processor 1001, or CPU (Central Processing Unit ), is a computing core as well as a control core of a computer device, which is adapted to implement one or more computer programs, in particular to load and execute one or more computer programs for implementing the respective method flows or the respective functions.

In one embodiment, one or more computer programs stored in storage medium 1002 may be loaded and executed by processor 1001 to implement the corresponding method steps described above in connection with the method embodiments illustrated in fig. 4 and 6; in particular implementations, one or more computer programs in storage medium 1002 are loaded by processor 1001 and perform the steps of:

In one embodiment, the processor 1001 may also be configured to load and execute:

In yet another embodiment, the processor 1001 may also be configured to load and execute:

In yet another embodiment, when the processor 1001 is configured to load a text word included based on the arbitrary candidate answer information, and perform semantic filling processing on the target text word in a knowledge domain related to the target text word, to obtain a filled text, the processor may be further specifically configured to load and execute:

In yet another embodiment, the processor 1001, when configured to load and execute determining a matching degree between each text segment and the question information, may be specifically configured to load and execute:

In yet another embodiment, the processor 1001, when configured to load and execute a knowledge domain analysis on the question information, obtains one or more knowledge domains related to the question information, may be specifically configured to load and execute:

In yet another embodiment, when the processor 1001 is configured to load and execute the feature matrix under each feature dimension according to the question information and the feature matrix corresponding to the each feature dimension of the any text segment, the processor may be specifically configured to load and execute:

In yet another embodiment, the processor 1001, when configured to load and execute determining a matching degree between each text segment and the question information, may be further specifically configured to load and execute:

In yet another embodiment, the processor 1001, when configured to load and execute the answer information that generates the question information, may be specifically configured to load and execute:

In yet another embodiment, the processor 1001, when configured to load and execute the answer information for generating the question information, may be further specifically configured to load and execute:

The embodiment of the application further provides a storage medium (or called a computer storage medium), in which one or more computer programs corresponding to the intelligent question-answering method are stored, and when one or more processors load and execute the one or more computer programs, the description of the intelligent question-answering method in the embodiment can be realized, and will not be repeated here. Further, embodiments of the present application provide a program product comprising a computer program adapted to be loaded by a processor and to perform the intelligent question-answering method as shown in fig. 4 and 6. It will be appreciated that a computer program may be deployed to be executed on one or more devices that are capable of communication with one another. The description of the advantageous effects of the same method as described above in the storage medium and the program product is omitted here.

Those skilled in the art will appreciate that implementing all or part of the above-described embodiments of the method may be accomplished by way of a computer program to instruct related hardware, and the computer program may be stored in a computer storage medium, and the computer program may include the steps of the above-described embodiments of the intelligent question-answering method when executed. The computer storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a random-access Memory (Random Access Memory, RAM), or the like.

It is again particularly emphasized that when the above embodiments of the present application are applied to specific products or technologies, the various embodiments of the present application relate to the acquisition of data related to question and answer requests, question information, etc., and that user permissions or consents need to be obtained, and that the collection, use and processing of the related data all need to comply with the relevant laws and regulations and standards of the relevant country and region.

The foregoing disclosure is merely a partial embodiment of the present application, and it is not intended to limit the scope of the claims of the present application.

Claims

1. An intelligent question-answering method is characterized by comprising the following steps:

2. The method according to claim 1, wherein the method further comprises:

3. The method according to claim 2, wherein the semantic filling processing is performed on the target text word in the knowledge domain related to the target text word based on the text word included in the arbitrary candidate answer information to obtain a filled text, including:

4. A method according to any one of claims 1-3, wherein said determining a degree of matching between each text segment and the question information comprises:

5. The method of claim 4, wherein the performing knowledge domain analysis on the questioning information to obtain one or more knowledge domains to which the questioning information relates comprises:

6. A method according to any one of claims 1-3, wherein said determining a degree of matching between each text segment and the question information comprises:

7. The method of claim 6, wherein determining the matching degree between the arbitrary text segment and the question information according to the feature matrix of the question information in the respective feature dimension and the feature matrix corresponding to the arbitrary text segment in the respective feature dimension includes:

8. A method according to any one of claims 1-3, wherein said determining a degree of matching between each text segment and the question information comprises:

9. A method according to any one of claims 1-3, wherein said generating answer information to said question information comprises:

10. An intelligent question-answering device, comprising:

11. A computer device, comprising:

a processor adapted to implement one or more computer programs;

a storage medium storing one or more computer programs adapted to be loaded by the processor and to perform the intelligent question-answering method according to any one of claims 1-9.

12. A storage medium storing one or more computer programs adapted to be loaded by the processor and to perform the intelligent question-answering method according to any one of claims 1 to 9.