CN112000776B

CN112000776B - Topic matching method, device, equipment and storage medium based on voice semantics

Info

Publication number: CN112000776B
Application number: CN202010880800.2A
Authority: CN
Inventors: 李军
Original assignee: Ping An Property and Casualty Insurance Company of China Ltd
Current assignee: Ping An Property and Casualty Insurance Company of China Ltd
Priority date: 2020-08-27
Filing date: 2020-08-27
Publication date: 2024-07-19
Anticipated expiration: 2040-08-27
Also published as: CN112000776A

Abstract

The invention relates to the field of artificial intelligence, and provides a topic matching method, device, equipment and storage medium based on voice semantics, which are used for improving the matching efficiency and matching accuracy of topics. The topic matching method based on the voice semantics comprises the following steps: acquiring client voice information in the information to be processed and the dialogue voice information through a topic retrieval system; extracting first characteristic information of information to be processed and second characteristic information of customer voice information; the first characteristic information and the second characteristic information are fused to obtain target characteristic information; searching a preset topic knowledge graph to obtain candidate topic information; calculating a space vector included angle cosine value between the candidate topic information and the target feature information, and determining the candidate topic information corresponding to the space vector included angle cosine value conforming to a preset rule as the target topic information. In addition, the invention also relates to a blockchain technology, and the information to be processed and the dialogue voice information can be stored in the blockchain node.

Description

Topic matching method, device, equipment and storage medium based on voice semantics

Technical Field

The invention relates to the field of artificial intelligence, in particular to a topic matching method, device, equipment and storage medium based on voice semantics.

Background

Along with the rapid development of computer intellectualization, information interaction between a robot and a person also develops in a diversified manner, and topic information is acquired according to interaction voice to be one information interaction mode. Currently, in order to realize an information interaction mode of obtaining topic information according to an interaction voice, a mode of extracting features of the content of the interaction voice and retrieving and obtaining corresponding topic information according to keywords or labels in the feature information extracted by the features is generally adopted.

The following drawbacks exist for the above approach: the corresponding topic information is obtained through the keyword retrieval or the label extraction mode, so that the obtained topic information has low correspondence with the characteristic information, and multiple retrieval operations are needed to obtain topics with high correspondence, so that the topic matching efficiency is low; the corresponding topic information is obtained only according to the characteristic information retrieval in the content of the interactive voice, so that the content of the characteristic information is monotonous, and when the content information of the knowledge of the content needs to be inferred simply in the content of the interactive voice, the corresponding characteristic information cannot be obtained accurately, so that the topic matching accuracy is low.

Disclosure of Invention

The invention mainly aims to solve the problems of low topic matching efficiency and low topic matching accuracy.

The first aspect of the invention provides a topic matching method based on voice semantics, which comprises the following steps:

acquiring information to be processed and dialogue voice information through a preset topic retrieval system, and extracting client voice information in the dialogue voice information, wherein the information to be processed comprises basic information of clients and historical operation information of the clients on the topic retrieval system;

Extracting first characteristic information of the information to be processed and second characteristic information of the client voice information, wherein the first characteristic information comprises location characteristic information, age characteristic information, sex characteristic information and historical topic characteristic information of the client, and the second characteristic information comprises semantic characteristic information, dialect characteristic information, intonation characteristic information and emotion characteristic information;

According to a preset attention mechanism, the first characteristic information and the second characteristic information are fused to obtain target characteristic information;

creating a target knowledge graph of the information to be processed and the client voice information, and matching the target knowledge graph with a preset topic knowledge graph to obtain candidate topic information, wherein the topic knowledge graph comprises various topic information corresponding to basic information, intonation information, emotion information and dialect information of the client;

calculating a space vector included angle cosine value between the candidate topic information and the target feature information, analyzing the space vector included angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as target topic information.

Optionally, in a first implementation manner of the first aspect of the present invention, the calculating a spatial vector angle cosine value between the candidate topic information and the target feature information, analyzing the spatial vector angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as the target topic information further includes:

obtaining the score of the target topic information, and storing the target topic information in a preset recommended topic information base according to the score and preset alternative keywords;

acquiring initial historical client information stored in the topic retrieval system, classifying topic types of the initial historical client information, and obtaining classified historical client information;

Acquiring recommended topic information in the recommended topic information base, and performing iterative optimization on the topic retrieval system according to the preset candidate keywords, the recommended topic information and the classified historical client information.

Optionally, in a second implementation manner of the first aspect of the present invention, the obtaining the score of the target topic information, and storing the target topic information in a preset recommended topic information base according to the score and a preset candidate keyword includes:

obtaining the score of the target topic information, and judging whether the value of the score is larger than a preset threshold value;

if the value of the score is larger than a preset threshold value, storing the target topic information in a preset recommended topic information base;

If the value of the score is smaller than or equal to the preset threshold value, acquiring a preset alternative keyword and a target topic type of the target topic information;

Creating a corresponding relation between the target topic type and the preset alternative keywords, and storing the preset alternative keywords creating the corresponding relation in the recommended topic information base.

Optionally, in a third implementation manner of the first aspect of the present invention, the obtaining initial historical client information stored in the topic retrieval system, classifying a topic type of the initial historical client information, and obtaining classified historical client information includes:

acquiring initial historical client information stored in the topic retrieval system and historical topic information corresponding to the initial historical client information;

Clustering the historical topic information and the target topic information through a preset clustering algorithm to obtain the type of the interest topic;

And classifying the initial historical client information according to the interest topic type to obtain classified historical client information.

Optionally, in a fourth implementation manner of the first aspect of the present invention, the extracting first feature information of the information to be processed and second feature information of the client voice information includes:

sequentially carrying out feature selection processing and feature vector extraction processing on the information to be processed to obtain first feature information;

respectively carrying out semantic conversion processing, audio signal diagram generation processing, syllable language recognition processing and emotion recognition processing on the customer voice information to obtain semantic information, dialect syllable information, audio signal diagram and first emotion information, and carrying out emotion recognition processing on the semantic information to obtain second emotion information;

matching dialect information corresponding to the dialect syllable information from a preset dialect database, carrying out statistical analysis on the audio signal diagram to obtain intonation information, and carrying out type classification and fusion processing on the first emotion information and the second emotion information to obtain fusion emotion information;

sequentially performing word segmentation processing, part-of-speech splicing filtering processing and keyword extraction on the semantic information to obtain semantic feature information;

Extracting feature vectors of the dialect information, the intonation information and the fusion emotion information respectively to obtain dialect feature information, intonation feature information and emotion feature information;

And determining the semantic feature information, the dialect feature information, the intonation feature information and the emotion feature information as second feature information of the client voice information.

Optionally, in a fifth implementation manner of the first aspect of the present invention, the calculating a spatial vector angle cosine value between the candidate topic information and the target feature information, analyzing the spatial vector angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as the target topic information further includes:

Acquiring good evaluation information of the dialogue voice information, wherein the good evaluation information is used for indicating that the accuracy of the target topic information is higher than a preset accuracy threshold;

And generating an analysis report according to the good evaluation information, the information to be processed, the dialogue voice information and the target topic information.

Optionally, in a sixth implementation manner of the first aspect of the present invention, before the obtaining, by a preset topic retrieval system, information to be processed and dialogue voice information, and extracting customer voice information in the dialogue voice information, the method further includes:

Configuring service information and a service topic type corresponding to the service information;

Matching business topic information corresponding to the business topic type from a preset database, and creating the corresponding relation among the business information, the business topic type, the business topic information and preset information, wherein the preset information comprises basic information, intonation information, emotion information and dialect information of a client;

and generating a topic knowledge map according to the business information, the business topic type, the business topic information and the preset information which are created with the corresponding relation.

The second aspect of the present invention provides a topic matching device based on speech semantics, comprising:

The information acquisition module is used for acquiring information to be processed and dialogue voice information through a preset topic retrieval system, and extracting client voice information in the dialogue voice information, wherein the information to be processed comprises basic information of a client and historical operation information of the client on the topic retrieval system;

the feature extraction module is used for extracting first feature information of the information to be processed and second feature information of the client voice information, wherein the first feature information comprises location feature information, age feature information, gender feature information and historical topic feature information of the client, and the second feature information comprises semantic feature information, dialect feature information, intonation feature information and emotion feature information;

the fusion processing module is used for carrying out fusion processing on the first characteristic information and the second characteristic information according to a preset attention mechanism to obtain target characteristic information;

The creating and retrieving module is used for creating a target knowledge graph of the information to be processed and the client voice information, matching the target knowledge graph with a preset topic knowledge graph to obtain candidate topic information, wherein the topic knowledge graph comprises various topic information corresponding to basic information, intonation information, emotion information and dialect information of the client;

The calculation and determination module is used for calculating a space vector included angle cosine value between the candidate topic information and the target feature information, analyzing the space vector included angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as target topic information.

Optionally, in a first implementation manner of the second aspect of the present invention, the topic matching device based on voice semantics further includes:

The storage module is used for acquiring the score of the target topic information and storing the target topic information in a preset recommended topic information base according to the score and preset alternative keywords;

The classification module is used for acquiring initial historical client information stored in the topic retrieval system, classifying topic types of the initial historical client information and obtaining classified historical client information;

The optimization module is used for acquiring recommended topic information in the recommended topic information base, and carrying out iterative optimization on the topic retrieval system according to the preset alternative keywords, the recommended topic information and the classified historical client information.

Optionally, in a second implementation manner of the second aspect of the present invention, the storage module may be further specifically configured to:

Optionally, in a third implementation manner of the second aspect of the present invention, the classification module may be further specifically configured to:

Optionally, in a fourth implementation manner of the second aspect of the present invention, the feature extraction module may be further specifically configured to:

Optionally, in a fifth implementation manner of the second aspect of the present invention, the topic matching device based on voice semantics further includes:

The acquisition module is used for acquiring good evaluation information of the dialogue voice information, wherein the good evaluation information is used for indicating that the accuracy of the target topic information is higher than a preset accuracy threshold;

The first generation module is used for generating an analysis report according to the good evaluation information, the information to be processed, the dialogue voice information and the target topic information.

Optionally, in a sixth implementation manner of the second aspect of the present invention, the topic matching device based on voice semantics further includes:

the configuration module is used for configuring service information and service topic types corresponding to the service information;

the creating module is used for matching the business topic information corresponding to the business topic type from a preset database, and creating the business information, the business topic type, the corresponding relation between the business topic information and preset information, wherein the preset information comprises basic information, intonation information, emotion information and dialect information of a client;

the second generation module is used for generating a topic knowledge map according to the business information, the business topic type, the business topic information and the preset information which are created with the corresponding relation.

A third aspect of the present invention provides a topic matching device based on speech semantics, comprising: a memory and at least one processor, the memory having instructions stored therein, the memory and the at least one processor being interconnected by a line; the at least one processor invokes the instructions in the memory to cause the speech semantic based topic matching device to perform the speech semantic based topic matching method described above.

A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored therein that, when executed on a computer, cause the computer to perform the above-described topic matching method based on speech semantics.

According to the technical scheme provided by the invention, the information to be processed and the dialogue voice information are obtained through a preset topic retrieval system, and the customer voice information in the dialogue voice information is extracted; extracting first characteristic information of information to be processed and second characteristic information of customer voice information; according to a preset attention mechanism, the first characteristic information and the second characteristic information are fused to obtain target characteristic information; creating a target knowledge graph of the information to be processed and the client voice information, and matching the target knowledge graph with a preset topic knowledge graph to obtain candidate topic information; calculating a space vector included angle cosine value between the candidate topic information and the target feature information, analyzing the space vector included angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as target topic information. According to the invention, the characteristic information is enriched by combining the to-be-processed information and the customer voice information in the dialogue information, the correspondence between the obtained topics and the characteristic information is improved, the first characteristic information and the second characteristic information are fused according to the preset attention mechanism, so that the target characteristic information is focused on the characteristic information on the basis of not losing the original information, when the content information of the interactive voice needs to be inferred simply, the corresponding characteristic information can be accurately obtained, the accuracy of the target characteristic information is improved, candidate topic information is obtained by searching a dialogue topic knowledge map, and the target topic content is matched from the topic content obtained by searching through a space vector included angle cosine minimum value algorithm, the matching efficiency and the matching accuracy of the topics are improved, and the efficiency and the accuracy of obtaining the target topic information are further improved.

Drawings

FIG. 1 is a schematic diagram of one embodiment of a topic matching method based on speech semantics in an embodiment of the present invention;

FIG. 2 is a schematic diagram of another embodiment of a topic matching method based on speech semantics in an embodiment of the present invention;

FIG. 3 is a schematic diagram of another embodiment of a topic matching method based on speech semantics in an embodiment of the present invention;

FIG. 4 is a schematic diagram of an embodiment of a topic matching device based on speech semantics in an embodiment of the present invention;

FIG. 5 is a schematic diagram of another embodiment of a topic matching device based on speech semantics in an embodiment of the present invention;

FIG. 6 is a schematic diagram of another embodiment of a topic matching device based on speech semantics in an embodiment of the present invention;

FIG. 7 is a schematic diagram of an embodiment of a topic matching device based on speech semantics in an embodiment of the present invention.

Detailed Description

The embodiment of the invention provides a topic matching method, device, equipment and storage medium based on voice semantics, which improves the efficiency and accuracy of acquiring target topic information.

The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.

For easy understanding, the following describes a specific flow of an embodiment of the present invention, referring to fig. 1, and an embodiment of a topic matching method based on speech semantics in an embodiment of the present invention includes:

101. The method comprises the steps of acquiring information to be processed and dialogue voice information through a preset topic retrieval system, and extracting client voice information in the dialogue voice information, wherein the information to be processed comprises basic information of clients and historical operation information of the clients on the topic retrieval system.

It can be understood that the execution subject of the present invention may be a topic matching device based on speech semantics, and may also be a terminal or a server, which is not limited herein. The embodiment of the invention is described by taking a server as an execution main body as an example.

For example: when a user calls, the server reads the dialed telephone number through a preset topic retrieval system, retrieves a preset database according to the telephone number to obtain user information of the dialed telephone, extracts historical operation information corresponding to the user on the topic retrieval system from the preset database, and determines the user information and the historical operation information as information to be processed; when the opposite party (namely the client) is connected with the telephone, the topic retrieval system generates dialogue voice information chatting with the client, the server carries out real-time artificial intelligent voiceprint recognition on the dialogue voice information through a voice recognition model in the topic retrieval system, and carries out classification processing on the dialogue voice information according to voiceprints obtained by the voiceprint recognition to obtain the client voice information so as to realize the real-time recognition of the client voice information, reduce the storage space and efficiency when retrieving information to be processed, and facilitate the processing of the client voice information in the dialogue voice information.

102. Extracting first characteristic information of information to be processed and second characteristic information of customer voice information, wherein the first characteristic information comprises location characteristic information, age characteristic information, sex characteristic information and historical topic characteristic information of customers, and the second characteristic information comprises semantic characteristic information, dialect characteristic information, intonation characteristic information and emotion characteristic information.

And the server calls a preset vocabulary model to perform feature extraction on the information to be processed to obtain first feature information. The server calls a preset Gaussian mixture model-hidden Markov model (gaussian mixture model-hidden markov model, GMM-HMM), converts customer voice information into text information to obtain voice text information, performs word segmentation processing on the voice text information to obtain word segmentation, performs feature extraction on the word segmentation to obtain voice content feature information, samples the customer voice information for a preset time to obtain voice fragments, performs tone recognition, emotion recognition and voiceprint recognition of dialects on each voice fragment respectively, performs feature extraction processing to obtain initial tone feature, initial emotion feature and initial dialect feature, performs feature selection processing on the initial tone feature, the initial emotion feature and the initial dialect feature respectively through a preset correlation-based feature selection and principal component analysis method to obtain candidate tone feature, candidate emotion feature and candidate dialect feature, and matches the voice content feature information, the emotion feature information and the dialect feature information from a preset tone database, an emotion database and a dialect feature database respectively, and takes the voice content feature information, the emotion feature information and the feature information as second feature information. The voice content characteristic information, the intonation characteristic information, the emotion characteristic information and the dialect characteristic information are determined to be the second characteristic information, so that the corresponding topic information can be matched in multiple directions according to the voice information of the client, and the accuracy of matching the target topic information is improved.

103. And according to a preset attention mechanism, carrying out fusion processing on the first characteristic information and the second characteristic information to obtain target characteristic information.

The server calculates probability of selecting feature information of preset words according to the first feature information and the second feature information by using a predefined attention variable according to an attention scoring function of a preset attention mechanism to obtain a first probability value and a second probability value; acquiring a first correlation degree of the first characteristic information and the preset word and a second correlation degree of the second characteristic information and the preset word; carrying out weighted average calculation on the first probability value according to the first correlation degree to obtain a first attention value, and carrying out weighted average calculation on the second probability value according to the second correlation degree to obtain a second attention value; multiplying the first characteristic information with the first attention value to obtain a first characteristic matrix vector, and multiplying the second characteristic information with the second attention value to obtain a second characteristic matrix vector; and carrying out matrix addition on the first characteristic matrix vector and the second characteristic matrix vector to obtain target characteristic information.

Through the operation, the characteristic information with strong relative criticality is focused in the plurality of first characteristic information and second characteristic information, the attention to other characteristic information with weak relative criticality is reduced, the problem of overload of the characteristic information is solved, and the efficiency and the accuracy of searching and matching according to the first characteristic information and the second characteristic information are improved.

104. Creating a target knowledge graph of the information to be processed and the client voice information, and matching the target knowledge graph with a preset topic knowledge graph to obtain candidate topic information, wherein the topic knowledge graph comprises various topic information corresponding to basic information, intonation information, emotion information and dialect information of the client.

Before the topic retrieval system acquires the information to be processed and the dialogue voice information, a topic knowledge map is constructed according to various topic information corresponding to user basic information, various intonation information, various emotion information and various dialect information which are crawled from various network platforms and various historical topic information read from the topic retrieval system, and the constructed topic knowledge map is continuously trained through a preset knowledge map construction model, so that the preset topic knowledge map is obtained.

Creating a target knowledge graph of the information to be processed and the client voice information through a preset knowledge graph construction model, creating an index of the target knowledge graph, and calling a preset search engine to search the preset topic knowledge graph according to the index to obtain candidate topic information matched with the target knowledge graph. And searching by adopting the target knowledge graph to acquire candidate topic information so as to quickly and accurately combine the information to be processed and the client voice information to match the corresponding candidate topic information.

105. Calculating a space vector included angle cosine value between the candidate topic information and the target feature information, analyzing the space vector included angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as target topic information.

The server respectively carries out vector conversion on the candidate topic information and the target feature information to obtain a topic vector and a feature vector, and calculates a cosine value of a space vector included angle between the topic vector and the feature vector through the following formula: wherein a represents a candidate topic information vector, B represents a target feature information vector, n represents the number of candidate topic information, and i represents an i-th target feature information vector. The preset conditions are as follows: and ordering the space vector angle cosine values in order from small to large to obtain ordering information, wherein the first space vector angle cosine value in the ordering information can be used as a target value, and the space vector angle cosine value at a preset position (for example, three first positions in the ordering information) in the ordering information can also be used as the target value. By taking the candidate topic information corresponding to the target value as target topic information, the efficiency and accuracy of the matched target topic information are improved.

In the embodiment of the invention, the characteristic information is enriched by combining the to-be-processed information and the customer voice information in the dialogue information, the correspondence between the obtained topics and the characteristic information is improved, the first characteristic information and the second characteristic information are fused according to the preset attention mechanism, so that the target characteristic information is focused on the characteristic information on the basis of not losing the original information, the corresponding characteristic information can be accurately obtained when the content information of the insight of the interactive voice needs to be simply inferred, the accuracy of the target characteristic information is improved, the candidate topic information is obtained by searching a dialogue topic knowledge map, the target topic content is matched from the searched topic content through a space vector included angle cosine minimum algorithm, the matching efficiency and the matching accuracy of topics are improved, and the efficiency and the accuracy of obtaining the target topic information are further improved.

Referring to fig. 2, another embodiment of a topic matching method based on speech semantics in an embodiment of the present invention includes:

201. The method comprises the steps of acquiring information to be processed and dialogue voice information through a preset topic retrieval system, and extracting client voice information in the dialogue voice information, wherein the information to be processed comprises basic information of clients and historical operation information of the clients on the topic retrieval system.

202. Extracting first characteristic information of information to be processed and second characteristic information of customer voice information, wherein the first characteristic information comprises location characteristic information, age characteristic information, sex characteristic information and historical topic characteristic information of customers, and the second characteristic information comprises semantic characteristic information, dialect characteristic information, intonation characteristic information and emotion characteristic information.

203. And according to a preset attention mechanism, carrying out fusion processing on the first characteristic information and the second characteristic information to obtain target characteristic information.

204. Creating a target knowledge graph of the information to be processed and the client voice information, and matching the target knowledge graph with a preset topic knowledge graph to obtain candidate topic information, wherein the topic knowledge graph comprises various topic information corresponding to basic information, intonation information, emotion information and dialect information of the client.

205. Calculating a space vector included angle cosine value between the candidate topic information and the target feature information, analyzing the space vector included angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as target topic information.

The implementation method of steps 201-205 is similar to the implementation method of steps 101-105 described above, and will not be described again here.

206. And acquiring good evaluation information of the dialogue voice information, wherein the good evaluation information is used for indicating that the accuracy of the target topic information is higher than a preset accuracy threshold.

The good evaluation information is information which evaluates the accuracy of the target topic information when the user performs voice interaction on the acquired target topic information smoothly, and the accuracy of the target topic information is higher than a preset accuracy threshold. The server extracts the good evaluation information from the database in the preset user terminal, and can also directly receive the good evaluation information input by the user on the preset display page.

207. And generating an analysis report according to the well evaluation information, the information to be processed, the dialogue voice information and the target topic information.

When the user performs voice interaction smoothly according to the acquired target topic information, the server performs statistics and analysis on the information generated in the process (namely, the good evaluation information, the client information, the dialogue voice information and the target topic information), and performs weight recognition and marking on the generated information through a preset hidden Markov model based on the role weight so as to display the contribution rate of the good evaluation information, the client information, the dialogue voice information and the target topic information to the voice interaction smoothly, and further generate a corresponding analysis report so as to be convenient to use the analysis report as reference and assistance of subsequent topic matching.

According to the embodiment of the invention, on the basis of improving the efficiency and accuracy of acquiring the target topic information, the analysis report is generated according to the good evaluation information, the information to be processed, the dialogue voice information and the target topic information, so that the target topic information can be sorted and classified according to the analysis report, and the availability and the multidirectional performance of the acquired target topic information are improved.

Referring to fig. 3, another embodiment of a topic matching method based on speech semantics in an embodiment of the present invention includes:

301. The method comprises the steps of acquiring information to be processed and dialogue voice information through a preset topic retrieval system, and extracting client voice information in the dialogue voice information, wherein the information to be processed comprises basic information of clients and historical operation information of the clients on the topic retrieval system.

Optionally, the server acquires information to be processed and dialogue voice information through a preset topic retrieval system, and extracts customer voice information in the dialogue voice information, wherein the information to be processed comprises basic information of a customer and service information and service topic types corresponding to the service information before historical operation information of the customer on the topic retrieval system; matching business topic information corresponding to the business topic type from a preset database, and creating corresponding relations among the business information, the business topic type, the business topic information and preset information, wherein the preset information comprises basic information, intonation information, emotion information and dialect information of a client; and generating a topic knowledge map according to the business information, the business topic type, the business topic information and the preset information which are created with the corresponding relation.

For example: setting the business information as a drama ticket of an second theater, wherein the business topic type corresponding to the business information is drama, acquiring each business drama information corresponding to the drama in a preset database, marking the drama ticket of the business information on each business drama information corresponding to the acquired business drama, creating a corresponding relation between the business information and the business topic information corresponding to the business topic type, creating the corresponding relation between the marked business drama information and basic information, tone information, emotion information and dialect information of a customer, and generating a topic knowledge map according to the business information, the business topic type, the business topic information and the preset information which are created with the corresponding relation. By creating the corresponding relation among the business information, the business topic type, the business topic information and the preset information, the content richness of the target topic information can be enhanced in a multi-angle manner on the basis of effectively acquiring the target topic information, and the topic is subjected to data structuring by generating a topic knowledge graph, so that the efficiency and the accuracy of the subsequent candidate topic information retrieval are improved.

302. Extracting first characteristic information of information to be processed and second characteristic information of customer voice information, wherein the first characteristic information comprises location characteristic information, age characteristic information, sex characteristic information and historical topic characteristic information of customers, and the second characteristic information comprises semantic characteristic information, dialect characteristic information, intonation characteristic information and emotion characteristic information.

Optionally, the server sequentially performs feature selection processing and feature vector extraction processing on the information to be processed to obtain first feature information; respectively carrying out semantic conversion processing, audio signal diagram generation processing, syllable recognition processing and emotion recognition processing on the customer voice information to obtain semantic information, dialect syllable information, audio signal diagram and first emotion information, and carrying out emotion recognition processing on the semantic information to obtain second emotion information; matching dialect information corresponding to dialect syllable information from a preset dialect database, carrying out statistical analysis on an audio signal diagram to obtain intonation information, and carrying out type classification and fusion processing on the first emotion information and the second emotion information to obtain fusion emotion information; sequentially performing word segmentation processing, part-of-speech splicing filtering processing and keyword extraction on the semantic information to obtain semantic feature information; extracting feature vectors of the dialect information, the intonation information and the fusion emotion information respectively to obtain dialect feature information, intonation feature information and emotion feature information; and determining the semantic feature information, the dialect feature information, the intonation feature information and the emotion feature information as second feature information of the client voice information.

The dialect information comprises dialect content, a dialect location and features of the dialect location corresponding to the client voice information. When the server performs statistics and analysis on the audio signal diagram to obtain intonation information, the statistics and analysis are performed according to the difference value and the time interval between the highest point and the lowest point in the audio signal diagram, for example: when the difference between the highest point and the lowest point is smaller than the first threshold value and the time interval is larger than the second threshold value, the client is in the smooth tone. When the server performs type classification and fusion processing on the first emotion information and the second emotion information to obtain fusion emotion information, attention value calculation can be performed on the first emotion information and the second emotion information through a preset attention mechanism, so that bias processing (namely weight) is performed on the first emotion information and the second emotion information, emotion categories of the first emotion information and the second emotion information after bias processing are classified, matrix addition processing is performed on the classified first emotion information and second emotion information, fusion emotion information is obtained, emotion recognition is performed through combination of text information (first emotion information) and voice information (second emotion information), and accuracy of emotion characteristic information is improved.

303. And according to a preset attention mechanism, carrying out fusion processing on the first characteristic information and the second characteristic information to obtain target characteristic information.

304. Creating a target knowledge graph of the information to be processed and the client voice information, and matching the target knowledge graph with a preset topic knowledge graph to obtain candidate topic information, wherein the topic knowledge graph comprises various topic information corresponding to basic information, intonation information, emotion information and dialect information of the client.

305. Calculating a space vector included angle cosine value between the candidate topic information and the target feature information, analyzing the space vector included angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as target topic information.

The implementation method of steps 303-305 is similar to the implementation method of steps 103-105 described above, and will not be described again here.

306. The score of the target topic information is obtained, and the target topic information is stored in a preset recommended topic information base according to the score and preset alternative keywords.

When the server acquires target topic information and renders the target topic information on a display page of the terminal, a user judges whether the target topic information is effective in dialogue topic communication according to the current voice interaction content so as to score the target topic information to obtain a score, after the server receives the score through a topic retrieval system, the server judges whether the value of the score is in a preset range value, if the value of the score is not in the preset range value, a preset alternative keyword is acquired from a preset alternative keyword library, the preset alternative keyword is conveyed to the display page, and the user clicks and selects the corresponding preset alternative keyword to obtain a selection result. When the server detects that preset alternative keywords selected by a user exist in the selection result, the preset alternative keywords are stored in a preset recommended topic information base according to the topic type of the target topic information; when the server detects that the corresponding preset alternative keywords selected by the user do not exist in the selection result, the target topic information is directly stored in a preset recommended topic information base. The target topic information and the preset alternative keywords are stored in the preset recommended topic information base according to the grading of the target topic information, so that the target topic information can be quickly and accurately acquired next time.

Optionally, the server acquires the score of the target topic information and judges whether the value of the score is larger than a preset threshold value; if the scoring value is greater than a preset threshold value, storing the target topic information in a preset recommended topic information base; if the value of the score is smaller than or equal to a preset threshold value, obtaining a preset alternative keyword and a target topic type of target topic information; creating a corresponding relation between the target topic type and the preset alternative keywords, and storing the preset alternative keywords creating the corresponding relation in a recommended topic information base.

For example: if the score of the target topic information is 85 and the score is less than 90, the server acquires a preset candidate keyword with the association degree with the target feature information being more than 80 from a preset candidate keyword library, the preset candidate keyword is conveyed to a display page, a user clicks the corresponding preset candidate keyword, meanwhile, the server classifies the type of the target topic information through a topic retrieval system to obtain the topic type of the target topic information, and the topic type is marked on the preset candidate keyword. The preset recommended topic information base is divided into storage spaces corresponding to the topic types, and the storage spaces corresponding to the topic types store alternative topic knowledge maps constructed according to preset alternative keywords of the marked topic types and corresponding topic information; if the score of the target topic information is 95 points and the score of the target topic information is more than 90 points, the target topic information is used as a recommended topic for the next operation of the current voice interaction content and is stored in a recommended topic information base, and when the server analyzes the characteristic information corresponding to the topic types of the same or similar recommended topics, the recommended topic is directly called from the recommended topic information base without carrying out searching and matching operations of a knowledge graph, so that the searching efficiency and accuracy are improved.

307. The method comprises the steps of obtaining initial historical client information stored in a topic retrieval system, classifying topic types of the initial historical client information, and obtaining classified historical client information.

The method comprises the steps that a server obtains historical topic information in a topic retrieval system and the bias topic type of the historical topic information (the bias topic type is the topic type of the historical topic information, the corresponding weight is relatively high), initial historical client information stored in the topic retrieval system is obtained, all clients are classified according to the initial historical client information stored in the bias topic type in the topic retrieval system, so that the interest of each client is obtained, topic information matching according to the interest is achieved, when the clients are subjected to voice interaction in the follow-up process, the topic information can be directly and effectively called, the corresponding topic information can be mutually referred to the clients of the same interest type, and therefore accuracy of target topic information obtaining is improved.

Optionally, the server acquires initial historical client information stored in the topic retrieval system and historical topic information corresponding to the initial historical client information; clustering historical topic information and target topic information by a preset clustering algorithm to obtain the type of the interest topic; and classifying the initial historical client information according to the interest topic type to obtain classified historical client information.

Topic retrieval systems store more and more topic information and customer information as usage time increases. The topic retrieval system is used for extracting the historical topic information corresponding to the initial historical client information and the initial historical client information stored in a preset time period, and clustering the historical topic information and the target topic information through a preset clustering algorithm K-MEANS clustering algorithm and a Gaussian mixture model-based expectation maximization clustering algorithm so as to acquire the interest and hobby content of each client. And classifying the initial historical client information by taking the interest topic type as a space coordinate axis and taking the weight of the interest topic type as the length of the coordinate axis. The initial historical client information is classified according to the interest topic types, so that the client information is managed, statistically analyzed and the classification status of the initial historical client information is intuitively displayed, the corresponding target topic information is conveniently acquired according to the initial historical client information next time, and the efficiency and the accuracy of acquiring the corresponding target topic information by the initial historical client information are improved.

308. Acquiring recommended topic information in a recommended topic information base, and performing iterative optimization on a topic retrieval system according to preset candidate keywords, the recommended topic information and the classified historical client information.

The topic retrieval system is continuously optimized and updated according to the recommended topic information in the recommended topic information base and the classified historical client information by taking preset alternative keywords as optimization parameters through a preset multi-objective optimization algorithm or a boundary cross aggregation algorithm, so that the topic retrieval system can acquire corresponding recommended topic information rapidly and effectively according to the preset alternative keywords when acquiring target topic information intelligently, the intelligence of the topic retrieval system is improved, and the accuracy and the speed of acquiring the target topic information and the recommended topic information are improved.

According to the embodiment of the invention, on the basis of improving the efficiency and accuracy of acquiring the target topic information, the target topic information and the preset alternative keywords are stored in the preset recommended topic information base according to the score of the target topic information, so that the target topic information can be acquired quickly and accurately next time, the topic type classification is carried out on the initial historical client information, the matching of the topic information according to the interest and hobbies is realized, so that the topic information can be directly and effectively called when the client is subjected to voice interaction later, and the corresponding topic information can be mutually referred between clients with the same interest and hobbies, so that the accuracy of acquiring the target topic information is improved, and the intelligent of a topic retrieval system is improved and the accuracy and speed of acquiring the target topic information and the recommended topic information are improved through iterative optimization according to the preset alternative keywords, the recommended topic information and the classified historical client information.

The topic matching method based on the voice semantics in the embodiment of the present invention is described above, and the topic matching device based on the voice semantics in the embodiment of the present invention is described below, referring to fig. 4, one embodiment of the topic matching device based on the voice semantics in the embodiment of the present invention includes:

The information acquisition module 401 is configured to acquire information to be processed and dialogue voice information through a preset topic retrieval system, and extract customer voice information in the dialogue voice information, where the information to be processed includes basic information of a customer and historical operation information of the customer on the topic retrieval system;

the feature extraction module 402 is configured to extract first feature information of information to be processed, including location feature information, age feature information, gender feature information, and historical topic feature information of the client, and second feature information of client voice information, including semantic feature information, dialect feature information, intonation feature information, and emotion feature information;

the fusion processing module 403 is configured to perform fusion processing on the first feature information and the second feature information according to a preset attention mechanism, so as to obtain target feature information;

The creating and retrieving module 404 is configured to create a target knowledge graph of the information to be processed and the voice information of the client, and match the target knowledge graph with a preset topic knowledge graph to obtain candidate topic information, where the topic knowledge graph includes various topic information corresponding to basic information, intonation information, emotion information and dialect information of the client;

The calculation determining module 405 is configured to calculate a spatial vector angle cosine value between the candidate topic information and the target feature information, analyze the spatial vector angle cosine value according to a preset condition to obtain a target value, and determine the candidate topic information corresponding to the target value as the target topic information.

The function implementation of each module in the topic matching device based on the voice semantics corresponds to each step in the topic matching method embodiment based on the voice semantics, and the function and the implementation process of the function implementation are not described in detail herein.

The topic matching method based on the voice semantics in the embodiment of the present invention is described above, and the topic matching device based on the voice semantics in the embodiment of the present invention is described below, referring to fig. 5, one embodiment of the topic matching device based on the voice semantics in the embodiment of the present invention includes:

The calculation determining module 405 is configured to calculate a spatial vector angle cosine value between the candidate topic information and the target feature information, analyze the spatial vector angle cosine value according to a preset condition to obtain a target value, and determine candidate topic information corresponding to the target value as target topic information;

The obtaining module 406 is configured to obtain good evaluation information of the dialogue speech information, where the good evaluation information is used to indicate that accuracy of the target topic information is higher than a preset accuracy threshold;

The first generation module 407 is configured to generate an analysis report according to the well-assessed information, the information to be processed, the dialogue voice information and the target topic information.

Referring to fig. 6, another embodiment of a topic matching device based on speech semantics in an embodiment of the present invention includes:

the storage module 408 is configured to obtain a score of the target topic information, and store the target topic information in a preset recommended topic information base according to the score and a preset candidate keyword;

the classification module 409 is configured to obtain initial historical client information stored in the topic retrieval system, and classify topic types of the initial historical client information to obtain classified historical client information;

The optimizing module 410 is configured to obtain recommended topic information in the recommended topic information base, and perform iterative optimization on the topic retrieval system according to the preset candidate keywords, the recommended topic information and the classified historical client information.

Optionally, the storage module 408 may be further specifically configured to:

if the scoring value is greater than a preset threshold value, storing the target topic information in a preset recommended topic information base;

If the value of the score is smaller than or equal to a preset threshold value, obtaining a preset alternative keyword and a target topic type of target topic information;

Creating a corresponding relation between the target topic type and the preset alternative keywords, and storing the preset alternative keywords creating the corresponding relation in a recommended topic information base.

Optionally, the classification module 409 may be further specifically configured to:

acquiring initial historical client information stored in a topic retrieval system and historical topic information corresponding to the initial historical client information;

Clustering historical topic information and target topic information by a preset clustering algorithm to obtain the type of the interest topic;

Optionally, the feature extraction module 402 may be further specifically configured to:

respectively carrying out semantic conversion processing, audio signal diagram generation processing, syllable recognition processing and emotion recognition processing on the customer voice information to obtain semantic information, dialect syllable information, audio signal diagram and first emotion information, and carrying out emotion recognition processing on the semantic information to obtain second emotion information;

Matching dialect information corresponding to dialect syllable information from a preset dialect database, carrying out statistical analysis on an audio signal diagram to obtain intonation information, and carrying out type classification and fusion processing on the first emotion information and the second emotion information to obtain fusion emotion information;

Optionally, the topic matching device based on voice semantics further includes:

The configuration module 411 is configured to configure service information and a service topic type corresponding to the service information;

the creating module 412 is configured to match the service topic information corresponding to the service topic type from the preset database, and create a corresponding relationship between the service information, the service topic type, the service topic information and the preset information, where the preset information includes basic information, intonation information, emotion information and dialect information of the client;

the second generation module 413 is configured to generate a topic knowledge map according to the service information, the service topic type, the service topic information and the preset information, where the service topic information, the service topic type, the service topic information and the preset information are created in a corresponding relationship.

The function implementation of each module and each unit in the topic matching device based on the voice semantics corresponds to each step in the topic matching method embodiment based on the voice semantics, and the function and the implementation process of the function implementation are not described in detail herein.

The topic matching device based on voice semantics in the embodiment of the present invention is described in detail from the point of view of the modularized functional entity in fig. 4 and fig. 6, and the topic matching device based on voice semantics in the embodiment of the present invention is described in detail from the point of view of hardware processing.

Fig. 7 is a schematic structural diagram of a topic matching device based on speech semantics, where the topic matching device 700 based on speech semantics may be relatively different due to different configurations or performances, and may include one or more processors (central processing units, CPU) 710 (e.g., one or more processors) and a memory 720, one or more storage mediums 730 (e.g., one or more mass storage devices) storing applications 733 or data 732. Wherein memory 720 and storage medium 730 may be transitory or persistent. The program stored in the storage medium 730 may include one or more modules (not shown), each of which may include a series of instruction operations on the speech semantic based topic matching device 700. Still further, the processor 710 may be configured to communicate with the storage medium 730 to execute a series of instruction operations in the storage medium 730 on the speech semantic based topic matching device 700.

The speech semantic based topic matching device 700 can also include one or more power sources 740, one or more wired or wireless network interfaces 750, one or more input output interfaces 760, and/or one or more operating systems 731, such as Windows Serve, mac OS X, unix, linux, freeBSD, and the like. It will be appreciated by those skilled in the art that the speech semantic based topic matching device structure shown in FIG. 7 does not constitute a limitation of a speech semantic based topic matching device and may include more or fewer components than illustrated, or some components in combination, or a different arrangement of components.

The present invention also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, and may also be a volatile computer readable storage medium, where instructions are stored in the computer readable storage medium, when the instructions are executed on a computer, cause the computer to perform the steps of the topic matching method based on speech semantics.

Further, the computer-readable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function, and the like; the storage data area may store data created from the use of blockchain nodes, and the like.

The blockchain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm and the like. The blockchain (Blockchain), essentially a de-centralized database, is a string of data blocks that are generated in association using cryptographic methods, each of which contains information from a batch of network transactions for verifying the validity (anti-counterfeit) of its information and generating the next block. The blockchain may include a blockchain underlying platform, a platform product services layer, an application services layer, and the like.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. The topic matching method based on the voice semantics is characterized by comprising the following steps of:

Determining the semantic feature information, the dialect feature information, the intonation feature information and the emotion feature information as second feature information of the client voice information;

calculating a space vector included angle cosine value between the candidate topic information and the target feature information, analyzing the space vector included angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as target topic information;

2. The speech semantic based topic matching method of claim 1, wherein the obtaining the score of the target topic information, and storing the target topic information in a preset recommended topic information base according to the score and a preset candidate keyword, comprises:

3. The topic matching method based on speech semantics according to claim 1, wherein the obtaining initial historical client information stored in the topic retrieval system, classifying the topic type of the initial historical client information, and obtaining classified historical client information, includes:

4. The topic matching method based on voice semantics according to claim 1, wherein the calculating the spatial vector angle cosine value between the candidate topic information and the target feature information, analyzing the spatial vector angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as the target topic information, further comprises:

5. The topic matching method based on voice semantics according to any one of claims 1-4, wherein before the obtaining of the information to be processed and the dialogue voice information, and the extracting of the customer voice information in the dialogue voice information, further comprises:

6. The topic matching device based on the voice semantics is characterized by comprising:

The feature extraction module is used for sequentially carrying out feature selection processing and feature vector extraction processing on the information to be processed to obtain first feature information; respectively carrying out semantic conversion processing, audio signal diagram generation processing, syllable language recognition processing and emotion recognition processing on the customer voice information to obtain semantic information, dialect syllable information, audio signal diagram and first emotion information, and carrying out emotion recognition processing on the semantic information to obtain second emotion information; matching dialect information corresponding to the dialect syllable information from a preset dialect database, carrying out statistical analysis on the audio signal diagram to obtain intonation information, and carrying out type classification and fusion processing on the first emotion information and the second emotion information to obtain fusion emotion information; sequentially performing word segmentation processing, part-of-speech splicing filtering processing and keyword extraction on the semantic information to obtain semantic feature information; extracting feature vectors of the dialect information, the intonation information and the fusion emotion information respectively to obtain dialect feature information, intonation feature information and emotion feature information; determining the semantic feature information, the dialect feature information, the intonation feature information and the emotion feature information as second feature information of the client voice information;

the calculation and determination module is used for calculating a space vector included angle cosine value between the candidate topic information and the target feature information, analyzing the space vector included angle cosine value according to a preset condition to obtain a target value, and determining the candidate topic information corresponding to the target value as target topic information;

7. A speech semantic based topic matching apparatus, characterized in that the speech semantic based topic matching apparatus comprises: a memory and at least one processor, the memory having instructions stored therein, the memory and the at least one processor being interconnected by a line;

the at least one processor invoking the instructions in the memory to cause the speech semantic based topic matching device to perform the speech semantic based topic matching method of any of claims 1-5.

8. A computer readable storage medium comprising a stored data area storing data created from use of blockchain nodes and a stored program area storing a computer program, characterized in that the computer program when executed by a processor implements the speech semantic based topic matching method according to any of claims 1-5.