WO2021139243A1 - 基于人机交互的数据处理方法、装置、设备及存储介质 - Google Patents
基于人机交互的数据处理方法、装置、设备及存储介质 Download PDFInfo
- Publication number
- WO2021139243A1 WO2021139243A1 PCT/CN2020/118431 CN2020118431W WO2021139243A1 WO 2021139243 A1 WO2021139243 A1 WO 2021139243A1 CN 2020118431 W CN2020118431 W CN 2020118431W WO 2021139243 A1 WO2021139243 A1 WO 2021139243A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- question
- intent
- intention
- standard
- preset
- Prior art date
Links
- 230000003993 interaction Effects 0.000 title claims abstract description 36
- 238000003672 processing method Methods 0.000 title claims abstract description 32
- 238000012545 processing Methods 0.000 claims abstract description 42
- 238000004458 analytical method Methods 0.000 claims abstract description 19
- 230000011218 segmentation Effects 0.000 claims description 51
- 230000014509 gene expression Effects 0.000 claims description 30
- 230000004044 response Effects 0.000 claims description 29
- 239000012634 fragment Substances 0.000 claims description 15
- 238000000034 method Methods 0.000 claims description 10
- 238000005516 engineering process Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 102100034761 Cilia- and flagella-associated protein 418 Human genes 0.000 description 1
- 101100439214 Homo sapiens CFAP418 gene Proteins 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3343—Query execution using phonetics
Definitions
- This application relates to the field of big data, and in particular to a data processing method, device, equipment and storage medium based on human-computer interaction.
- the main purpose of this application is to provide a data processing method based on human-computer interaction, which aims to solve the technical problem of how to improve the efficiency of the intelligent question answering system for querying and replying content.
- the present application provides a data processing method based on human-computer interaction, and the data processing method includes the following steps:
- connection relationship corresponding to the current user's intention from the connection relationship corresponding to the node, wherein one connection relationship corresponds to a standard answer to the question;
- this application also provides a data processing device based on human-computer interaction, the data processing device includes:
- the parsing module is used to receive the voice data of the user's response to the current question and perform intent analysis to determine the current user's intent;
- the positioning module is used to locate the node corresponding to the current question in the preset standard Q&A knowledge base;
- the matching module is configured to match the target connection relationship corresponding to the current user's intention from the connection relationship corresponding to the node, wherein one of the connection relationships corresponds to a standard answer to the question;
- the first output module is used to locate the target node connected to the target connection relationship, and output the corresponding question of the target node in the preset standard Q&A knowledge base.
- the data processing device includes:
- the obtaining module is used to obtain the questions associated with the preset application scenarios and the standard answers corresponding to the questions, wherein one question corresponds to at least two standard answers, and the standard answer is the standard text expression of the main intent;
- the establishment module is used to connect each node with each question as the node and the standard answer corresponding to each question as the connection relationship between each node, and connect each question through the standard answer according to the questioning sequence of each question in the preset application scenario A directed graph is formed, and the directed graph is used as the preset standard Q&A knowledge base.
- the present application also provides a data processing device based on human-computer interaction.
- the data processing device includes a memory, a processor, and a device that is stored in the memory and can run on the processor.
- a data processing program when the data processing program is executed by the processor, implements the steps of the data processing method as described below:
- connection relationship corresponding to the current user's intention from the connection relationship corresponding to the node, wherein one connection relationship corresponds to a standard answer to the question;
- the present application also provides a computer-readable storage medium having a data processing program stored on the computer-readable storage medium, and when the data processing program is executed by a processor, the following data is realized Steps of the processing method:
- connection relationship corresponding to the current user's intention from the connection relationship corresponding to the node, wherein one connection relationship corresponds to a standard answer to the question;
- This application is based on pre-establishing a question-and-answer knowledge base with a picture library, receiving the voice data of the user’s response to the current question and performing intent analysis to determine the current user’s intention, and locate the node of the current question in the preset standard question-and-answer knowledge base.
- the connection relationship corresponding to the node the connection relationship corresponding to the current user's intention is matched, the next node connected to the connection relationship is located, and the text description information of the node is output as the reply content of the reply to the user.
- the query response content does not require a complicated relationship table, only the current problem node is located, and the text description of the next node corresponding to the user's intention is matched. Information, output reply content more quickly, and improve user experience.
- FIG. 1 is a schematic structural diagram of an operating environment of a data processing device based on human-computer interaction related to a solution of an embodiment of the application;
- FIG. 2 is a schematic flowchart of a first embodiment of a data processing method based on human-computer interaction according to this application;
- FIG. 3 is a schematic flowchart of a second embodiment of a data processing method based on human-computer interaction according to this application;
- FIG. 4 is a schematic diagram of a detailed flow chart of an embodiment of step S20 in FIG. 2;
- FIG. 5 is a detailed flowchart of an embodiment of step S205 in FIG. 4;
- FIG. 6 is a schematic flowchart of a third embodiment of a data processing method based on human-computer interaction according to this application;
- FIG. 7 is a schematic diagram of functional modules of an embodiment of a data processing apparatus based on human-computer interaction according to the present application.
- This application provides a data processing device.
- FIG. 1 is a schematic structural diagram of an operating environment of a data processing device based on human-computer interaction involved in a solution in an embodiment of the application.
- the data processing device includes: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005.
- the communication bus 1002 is used to implement connection and communication between these components.
- the user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).
- the memory 1005 may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as a magnetic disk memory.
- the memory 1005 may also be a storage device independent of the aforementioned processor 1001.
- FIG. 1 does not constitute a limitation on the data processing device, and may include more or less components than shown in the figure, or a combination of certain components, or different components.
- the layout of the components does not constitute a limitation on the data processing device, and may include more or less components than shown in the figure, or a combination of certain components, or different components. The layout of the components.
- the memory 1005 which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a computer program.
- the operating system is a program that manages and controls data processing equipment and software resources, and supports the operation of data processing programs and other software and/or programs.
- the network interface 1004 is mainly used to access the network; the user interface 1003 is mainly used to detect and confirm instructions and edit instructions.
- the processor 1001 may be used to call a data processing program stored in the memory 1005, and execute the operations of the following data processing method embodiments.
- FIG. 2 is a schematic flowchart of a first embodiment of a data processing method based on human-computer interaction in this application.
- the data processing method includes the following steps:
- Step S10 Receive the voice data of the user's response to the current question and perform intent analysis to determine the current user's intent
- the user's reply voice for the first question is received, the reply voice is analyzed intent, and the current user's intent is determined from the intent analysis data. Furthermore, since there are many forms of expression of the same intention, for example, the content of the user's answer has main intentions such as positive intention to answer the question positively, “Yes” and negative intention “No", and there are also non-main intents to answer the question on the user side. Intentions, such as “no time to understand insurance products", “I don't know if I need it", etc.
- chat bots are now able to recognize the user’s intention in the user’s question.
- the robot receives the voice data input by the user and transfers the voice data through natural language technology. It is converted into text information, and then based on the text information, it is compared in a predetermined standard Q&A knowledge base to query whether there is an answer corresponding to the intention expressed by the user, and then output the query result.
- Step S20 locate the node corresponding to the current question in the preset standard Q&A knowledge base
- the preset standard question and answer knowledge base first determine the current process link, then determine the current output problem, and then locate the node of the current problem.
- the business side the business side’s long-term work experience is counted, and after the first version is obtained, the image library is optimized and supplemented according to the actual robot work scenario, and questions and answers of different process links are obtained.
- Step S30 Match the target connection relationship corresponding to the current user's intention from the connection relationship corresponding to the node, wherein one connection relationship corresponds to a standard answer to the question;
- the user has different intentions, and different intentions correspond to different answers to the question.
- the answer to a question is represented by the connection relationship connecting the upper and lower questions in the preset Q&A knowledge base, so after locating the node of the current question, in the subordinate connection relationship of the node, the target connection corresponding to the current user's intention is matched relationship.
- the standard answer to the preset question has two intentions: “willing” or “unwilling".
- the corresponding standard answer is “need” Expressions such as “buy” and “want to buy”
- standard answers for "unwilling” include expressions such as “no need” and “don't want”
- the specific and two common answers will also lead to different next questions, such as users
- To answer the expression of "willing” intent get the questions of the next node of the connection relationship corresponding to the "willing” intent, such as asking the user "Do you have time now, introduce you to our latest auto insurance products", etc.; If the user expresses "unwilling”, then get the question of the next node of the target connection relationship corresponding to the "unwilling” intention, such as asking the user "Is there any other insurance needs” or “that will not bother you”.
- Step S40 Locate the target node connected to the target connection relationship, and output the corresponding question of the target node in the preset standard Q&A knowledge base.
- the node is positioned at the next node connected by the connection relationship, and the text description information of the node has a causal relationship with the previous question and the current user's intention.
- different questions will be raised in the scene dialogue, just like asking the questioner "have you eaten?"
- the answering party will have "about to eat” and “have already eaten”.
- the questioning party receives the first intention "I am going to eat”
- it will cause the questioning party to ask the answering party "what do you plan to eat or plan to eat", “who is going to eat with”, “where to eat” and other questions.
- the questioning party receives the second intention “have eaten”, it will cause the questioning party to ask questions such as “what did you eat”, “who did you eat with”, “where to eat”, etc., so different intentions, Will trigger a different next question, and the resulting question will become the content of the next question after the questioner receives the answer from the answerer.
- the user is obtained from the preset standard Q&A knowledge base The answer content or operation set by the next node associated with the first answer of the answer.
- the intelligent robot asks “have you eaten", and receives the user's intent expression of "have eaten”.
- the intent expression of "have eaten” is not limited, then query the node of the question of "have you eaten”. If the node is queried in the standard question and answer knowledge base, the connection relationship in the node that matches the intention "eat” is identified, and the text description content of the next node corresponding to the connection relationship is obtained and output.
- This embodiment is based on pre-establishing a question-and-answer knowledge base in the way of constructing a gallery, receives the voice data of the user’s response to the current question and performs intent analysis to determine the current user’s intention, and locates the node of the current question in the preset standard question-and-answer knowledge base.
- the connection relationship corresponding to the node the connection relationship corresponding to the current user's intention is matched, the next node connected to the connection relationship is located, and the text description information of the node is output as the reply content of the reply to the user.
- the query response content does not require a complicated relationship table, only the current problem node is located, and the text description of the next node corresponding to the user's intention is matched. Information, output reply content more quickly, and improve user experience.
- the current user intent includes a main intent and a non-main intent
- the main intent indicates that the user positively answers the current question
- the non-main intent indicates the user's side response The current issue.
- the content of the user's answer contains positive intentions of positively answering the question "Yes” and negative intentions "No"
- non-main intentions of the user to answer the question such as "no time to understand insurance products", "don't know whether you need it or not”, etc. Therefore, it is necessary to categorize the user's possible intentions for each question according to specific business scenarios in advance, and establish the standard Q & A knowledge base and non-standard Q & A knowledge base corresponding to the main intent library and the non-main intent, and then query according to the user's intent. Reply to the user's response content.
- Fig. 3 is a schematic flowchart of a second embodiment of a data processing method based on human-computer interaction according to this application.
- the method before step S10, the method further includes:
- Step S001 Obtain a question associated with a preset application scenario and a standard answer corresponding to the question, wherein one question corresponds to at least two standard answers, and the standard answer is a standard text expression of the main intent;
- Step S002 Use each question as a node, and use the standard answer corresponding to each question as the connection relationship to connect each node.
- the standard answers are used to connect each question into a A directed graph, using the directed graph as the preset standard Q&A knowledge base.
- different question and answer data exist in different application scenarios.
- the content of the question and answer between the salesperson and the user will be related to the real estate, such as the type, price, location of the house, etc.
- the content of the question and answer between the salesperson and the user It will be related to insurance, such as the scope of insurance, claim rate, premium insurance period, etc.
- intelligent question answering robots intelligent robots are used to replace sales staff to provide users with consulting services. Among them, frequently used questions and related answers can be collected from the staff’s daily work, and then optimized by experts to obtain standard commonly used Question and answer data.
- each common question is used as a node, and the answer corresponding to each common question is used as the relationship connecting each question node.
- each question is connected with only one answer to indicate that the next question is triggered by one of the answers to the previous question, and there are multiple answers to one question, and each answer Only one question can be raised, and the user's intention corresponds to one answer.
- FIG. 4 is a detailed flowchart of step S20 in FIG.
- the above step S30 further includes:
- Step S201 Receive voice data and input the voice data into a preset language recognition model for voice recognition, and output text information corresponding to the voice data;
- the voice training data of the staff's daily work conversations with the user is obtained in advance, and a voice recognition model is trained based on the voice training data. Based on the trained voice recognition model, after receiving the voice data of the user's answer to the current question, the voice data is input into the voice recognition model for voice recognition, and the text information of the voice data is obtained.
- Step S202 Perform word segmentation processing on the text information to obtain multiple word segmentation fragments
- any word segmentation algorithm in the prior art can be used to perform word segmentation to obtain multiple word segmentation fragments.
- the natural language text "Mr. Zhang needs auto insurance” is segmented through the n-gram word segmentation algorithm, and the segmented segments are “Mr. Zhang”, “Needs or Not", and "Auto Insurance”.
- Step S203 sequentially match each word segmentation segment with the preset knowledge vocabulary to obtain at least one concept label, where the concept label is used to map the abstract concept of the word segmentation segment;
- the preset knowledge vocabulary is queried, and the concept of each segmentation segment is matched. If the concept corresponding to the individual segmentation segment cannot be identified, the segmentation segment is skipped and the next segmentation continues to be identified.
- the word segmentation segment at least one concept corresponding to the word segmentation segment is identified, and the word segmentation segment is conceptually labeled with the matched concept to obtain the concept label of the word segmentation segment. It should be noted that a word segmentation corresponds to at least one concept in a question, so at least one concept label can be finally obtained, where the concept label is used to map the abstract concept of the word segmentation segment.
- Step S204 Arrange and combine the concept tags according to the expression order of their corresponding word segmentation fragments in the text information to obtain a concept tag sequence;
- the concept tags obtained in the above steps are arranged and combined in order according to the expression order of the corresponding word segmentation fragments in the text information to obtain multiple non-conflicting concept tag sequences, that is, in the same text , Only one segmentation covered by the concept tag can appear in the same sequence.
- the concept tags corresponding to the word segmentation "need or not” are "needed” and “unnecessary”, so the two concept tags need to be divided into different concept tags.
- the concept label sequences corresponding to the word segmentation "Mr. Zhang”, “Needs or Not”, and “Auto Insurance” are two concepts of "User, Need, Insurance” and “User, Don't Need, Insurance” respectively.
- Step S205 Perform intent matching on the concept tag sequence based on a preset intent knowledge network to determine the current user intent, where the preset intent knowledge network consists of each concept tag and each concept tag The association relationship structure of the corresponding intent tag.
- the pre-established intention knowledge network is an intention knowledge network generated by mapping real world intentions through human daily description language.
- the intention knowledge network includes the intention expression sets corresponding to different questions in the current application scenario, each intention expression is stored in the intention knowledge network in the form of tags, and the intention label sets corresponding to different questions are related to each other.
- each intent tag set will record a set of intent tags that are related to the problem and may exist, and those concept tags will be associated with these intent tags. For example, with regard to insurance, there is an intention to ask about the insurance period, type of insurance, and premiums.
- FIG. 5 is a schematic diagram of the detailed flow of step S205 in FIG.
- the above step S205 includes:
- Step S2051 Respectively match a single concept tag in the concept tag sequence or a combined tag composed of at least two adjacent concept tags with the intention knowledge network;
- a single concept label in the sequence of concept labels is matched with the intention knowledge network, or multiple adjacent concept labels are matched with the intention knowledge network.
- multiple adjacent concept tags are combined to match the intention knowledge network.
- the corresponding concept label sequence after word cut and concept label are "user, need, insurance” and “user, need, insurance”.
- “user, no need, insurance” may also be “user, no, need, insurance”
- the adjacent concept tags of "no, need” need to be combined in order to match the "unnecessary” intent tag.
- the purpose is to correctly match the user's intention, because if the order of the concept tags is disrupted, it does not conform to the expression habits of human natural language. Naturally, the concept tags are out of order. Matching will not match the correct intent result.
- Step S2052 If there is a concept tag that matches the intent tag, obtain the intent tag and merge the intent tag and the concept tag that has not yet been matched into a new concept tag sequence, and continue to perform intent tagging on the new concept tag sequence Matching, until all the concept tags in the concept tag sequence are matched or the intent tags are not matched, the final matching result is output, where the final matching result is an intent tag set formed by a combination of one or more intent tags;
- Step S2053 Determine the current user's intention based on the final matching result.
- FIG. 6 is a schematic flowchart of a third embodiment of a data processing method based on human-computer interaction according to this application.
- Step S60 If the target connection relationship corresponding to the current user's intent cannot be matched, it is determined that the current user's intent is a non-main intent, and the preset general question and answer knowledge base is queried whether there is a non-standard answer corresponding to the current user's intent, where:
- the preset general question and answer knowledge base is a database that stores non-standard answers corresponding to non-main intents associated with preset application scenarios and response content associated with the non-standard answers, and the non-standard answers represent standard text of users' non-main intents. expression;
- non-main intention refers to an intention other than the main intention, such as the question of "whether you want to buy auto insurance"
- users may also answer non-main intents corresponding to non-main intents such as "no money”, “no time to understand", and "purchased”.
- the standard answer in response to this situation, extract this kind of non-main intent data from the commonly used question and answer data, and establish a relational table as a general question and answer knowledge base based on the relationship between each non-main intent and the corresponding question.
- Step S70 If there is a non-standard answer corresponding to the current user's intention in the preset general question and answer knowledge base, output the response content associated with the non-standard answer;
- Step S80 If there is no non-standard answer corresponding to the current user's intention in the preset general question and answer knowledge base, output the preset response content.
- connection relationship corresponding to the current user's intention cannot be queried in the standard question and answer knowledge base, it means that the intention expressed by the user's current reply content is non-main intention, then the general question and answer knowledge base is queried to determine the Whether there is a non-standard answer corresponding to the current user's intention in the general question and answer knowledge base, and if the general question and answer knowledge base has a non-standard answer corresponding to the current user's intention, the reply content associated with the non-standard answer is output.
- a certain node asks the user "Is there a willingness to buy auto insurance”.
- the default standard answer corresponding to the intent of this node is "willing" and "unwilling”. If the user currently receives If the intent is "no money”, then the current question node in the standard Q&A knowledge base does not have an answer corresponding to the current user’s intent, and then query whether there is a non-standard answer corresponding to "no money” in the general Q&A knowledge base, and query
- the nodes in the general Q&A knowledge base asking the user "will you want to buy auto insurance” are preset with non-standard answers corresponding to the intent of "no money”, “no time to understand", “purchased”, etc., so they are in the general Q&A knowledge base
- the application also provides a data processing device.
- FIG. 7 is a schematic diagram of functional modules of an embodiment of a data processing apparatus based on human-computer interaction according to the present application.
- the data processing device includes:
- the parsing module 10 is used to receive the voice data of the user's reply to the current question and perform intent analysis to determine the current user's intention;
- the positioning module 20 is used to locate the node corresponding to the current question in the preset standard Q&A knowledge base;
- the matching module 30 is configured to match the target connection relationship corresponding to the current user's intention from the connection relationship corresponding to the node, where one connection relationship corresponds to a standard answer to the question;
- the first output module 40 is configured to locate the target node connected to the target connection relationship, and output the corresponding question of the target node in the preset standard Q&A knowledge base.
- the parsing module 10 includes:
- a recognition unit configured to receive voice data and input the voice data into a preset language recognition model for voice recognition, and output text information corresponding to the voice data;
- the word segmentation unit is used to perform word segmentation processing on the text information to obtain multiple word segmentation fragments;
- a labeling unit configured to sequentially match each word segmentation segment with a preset knowledge vocabulary to obtain at least one concept label, where the concept label is used to map the abstract concept of the word segmentation segment;
- An arranging unit for arranging and combining the concept tags according to the expression order of their corresponding word segmentation fragments in the text information to obtain a concept tag sequence
- the matching unit is configured to perform intent matching on the concept tag sequence based on a preset intent knowledge network to determine the current user intent, wherein the preset intent knowledge network consists of each concept tag and each concept tag The association relationship structure of the corresponding intent tag.
- the matching unit includes:
- the matching subunit is used to respectively match a single concept tag in the concept tag sequence or a combined tag composed of at least two adjacent concept tags with the intention knowledge network;
- the output subunit is used to obtain the intent label if there is a concept label that matches the intent label and merge the intent label and the unmatched concept label into a new concept label sequence, and continue with the new concept label sequence Perform intent tag matching until all the concept tags in the current concept tag sequence are matched or the intent tag cannot be matched, and the final matching result is output, where the final matching result is an intent tag formed by combining one or more intent tags set;
- the determining subunit is used to determine the current user's intention based on the final matching result.
- the data processing device includes:
- the query module is used to determine if the target connection relationship corresponding to the current user's intent cannot be matched, determine that the current user's intent is a non-main intent, and query whether there is a non-standard answer corresponding to the current user's intent in the preset general Q&A knowledge base,
- the preset general question and answer knowledge base is a database storing non-standard answers corresponding to non-main intents associated with preset application scenarios and response content associated with the non-standard answers, and the non-standard answers represent user non-main intents Standard text expression;
- the second output module is configured to output the response content associated with the non-standard answer if there is a non-standard answer corresponding to the current user's intention in the preset general question and answer knowledge base;
- the third output module is configured to output the preset response content if there is no non-standard answer corresponding to the current user's intention in the preset general question and answer knowledge base.
- the parsing module 10 receives the voice data of the user’s response to the current question and performs intent analysis to determine the current user’s intent.
- the positioning module 20 locates the node corresponding to the current question in a preset standard Q&A knowledge base.
- the first matching module 30 Match the target connection relationship corresponding to the current user's intention from the connection relationship corresponding to the node.
- the first output module 40 locates the target node connected to the target connection relationship, and outputs the target node in the preset standard question and answer Corresponding questions in the knowledge base.
- the application also provides a computer-readable storage medium.
- a data processing program is stored on the computer-readable storage medium.
- the computer-readable storage medium may be nonvolatile or volatile.
- the data processing program is implemented when executed by a processor. The steps of the data processing method as described in any of the above embodiments.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Acoustics & Sound (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
一种基于人机交互的数据处理方法,涉及大数据领域,包括以下步骤:接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图(S10);在预置标准问答知识库中定位当前问题对应的节点(S20);从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系(S30);定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题(S40)。同时还公开了一种基于人机交互的数据处理装置、设备及存储介质,解决了针对用于人机交互的数据处理繁杂,耗费时间大的技术问题。
Description
本申请要求于2020年01月08日提交中国专利局、申请号为202010017461.5、发明名称为“基于人机交互的数据处理方法、装置、设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
本申请涉及大数据领域,尤其涉及一种基于人机交互的数据处理方法、装置、设备及存储介质。
目前,人工智能作为新兴行业,近年来发展迅猛。随着人工智能技术的发展,诞生了越来越多的智能机器人。有服务于闲聊的娱乐型机器人、有类似于导购的前台机器人,也有服务于特定行业的业务型机器人。然而,发明人意识到,传统智能问答机器人的问答数据库都是以关系型数据库建立,随着问答数据的增多,问题与答案之间的关系越复杂,查询效率越来越低。
因此,如何提高智能问答系统查询回复内容的效率是亟待解决的问题。
发明内容
本申请的主要目的在于提供一种基于人机交互的数据处理方法,旨在解决如何提高智能问答系统查询回复内容的效率的技术问题。
为实现上述目的,本申请提供的一种基于人机交互的数据处理方法,所述数据处理方法包括以下步骤:
接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图;
在预置标准问答知识库中定位当前问题对应的节点;
从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;
定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
进一步地,为实现上述目的,本申请还提供一种基于人机交互的数据处理装置,所述数据处理装置包括:
解析模块,用于接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图;
定位模块,用于在预置标准问答知识库中定位当前问题对应的节点;
匹配模块,用于从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;
第一输出模块,用于定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
可选地,所述数据处理装置包括:
获取模块,用于获取预设应用场景关联的问题以及所述问题对应的标准答案,其中,一个问题至少对应两个标准答案,所述标准答案为主干意图的标准文字表达;
建立模块,用于分别以各问题为节点、以各问题对应的标准答案作为连接各节点的连接关系,根据每个问题在预设应用场景中的提问顺序,通过所述标准答案将各问题连接成有向图,将所述有向图作为所述预置标准问答知识库。
进一步地,为实现上述目的,本申请还提供一种基于人机交互的数据处理设备,所述数据处理设备包括存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的数据处理程序,所述数据处理程序被所述处理器执行时实现如下所述的数据处理方法的步骤:
接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图;
在预置标准问答知识库中定位当前问题对应的节点;
从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;
定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
进一步地,为实现上述目的,本申请还提供一种计算机可读存储介质,所述计算机可读存储介质上存储有数据处理程序,所述数据处理程序被处理器执行时实现如下所述的数据处理方法的步骤:
接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图;
在预置标准问答知识库中定位当前问题对应的节点;
从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;
定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
本申请基于预先以图库的构建方式建立问答知识库,接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图,在预置标准问答知识库中定位当前问题的节点,从所述节点对应的连接关系中,匹配与当前用户意图对应的连接关系,定位与所述连接关系相连的下一节点,并输出该节点的文字描述信息作为回复用户的答复内容。本申请相比传统技术中以关系表的形式构建问答知识库来说,查询回复内容不需要繁杂的关系表,只需定位当前问题节点,匹配与用户意图对应连接关系的下一节点的文字描述信息,更快速的输出回复内容,提升用户体验。
图1为本申请实施例方案涉及的基于人机交互的数据处理设备运行环境的结构示意图;
图2为本申请基于人机交互的数据处理方法第一实施例的流程示意图;
图3为本申请基于人机交互的数据处理方法第二实施例的流程示意图;
图4为图2中步骤S20一实施例的细化流程示意图;
图5为图4中步骤S205一实施例的细化流程示意图;
图6为本申请基于人机交互的数据处理方法第三实施例的流程示意图;
图7为本申请基于人机交互的数据处理装置一实施例的功能模块示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
应当理解,此处所描述的具体实施例仅用以解释本申请,并不用于限定本申请。
本申请提供一种数据处理设备。
参照图1,图1为本申请实施例方案涉及的基于人机交互的数据处理设备运行环境的 结构示意图。
如图1所示,该数据处理设备包括:处理器1001,例如CPU,通信总线1002、用户接口1003,网络接口1004,存储器1005。其中,通信总线1002用于实现这些组件之间的连接通信。用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard),网络接口1004可选的可以包括标准的有线接口、无线接口(如WI-FI接口)。存储器1005可以是高速RAM存储器,也可以是稳定的存储器(non-volatile memory),例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
本领域技术人员可以理解,图1中示出的数据处理设备的硬件结构并不构成对数据处理设备的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
如图1所示,作为一种计算机可读存储介质的存储器1005中可以包括操作系统、网络通信模块、用户接口模块以及计算机程序。其中,操作系统是管理和控制数据处理设备和软件资源的程序,支持数据处理程序以及其它软件和/或程序的运行。
在图1所示的数据处理设备的硬件结构中,网络接口1004主要用于接入网络;用户接口1003主要用于侦测确认指令和编辑指令等。而处理器1001可以用于调用存储器1005中存储的数据处理程序,并执行以下数据处理方法的各实施例的操作。
基于上述数据处理设备硬件结构,提出本申请数据处理方法的各个实施例。
参照图2,图2为本申请基于人机交互的数据处理方法第一实施例的流程示意图。本实施例中,所述数据处理方法包括以下步骤:
步骤S10:接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图;
人工智能作为新兴行业,进来发展迅猛。伴随着人工智能技术的发展与进步,各大科技企业相继推出智能机器人产品,其中,用于问答的智能机器人的核心技术是根据预置的问答知识库,输出对应的答复内容,而传统的问答知识库是基于问答数据中问题与答案的关系按关系型数据库建立问答知识库,这种方法建立的问答知识库随着问答数据的增加和问答关系的复杂,终会变得不堪重负,导致需要花费较长时间才能做出应答。
本实施例中,接收用户针对所述第一问题的答复语音,对所述答复语音进行意图解析,从意图解析数据中确定当前用户意图。进一步地,由于同一个意图的表述形式有很多,例如,用户回答的内容存在正面回答该问题的肯定意图“有”和否定意图“没有”等主干意图,也存在用户侧面答复该问题的非主干意图,如“没时间了解保险产品”、“不知道自己是否需要”等。所以需要预先根据具体业务场景,对用户针对每个问题可能出现的意图进行分类,分别建立主干意图库和非主干意图所对应的标准问答知识库和非标准问答知识库,然后根据用户意图,查询答复用户的应答内容。需要注意的是,意图是指系统为了明确用户意愿而抽象提炼出来的,是在一个对话任务中要理解的用户想法。
此外,在用户的话语中明确出现了表达意图的一类词汇,比如:“希望”、“想要”、“需要”等等,因而只需要识别出这些固定的意图特征词汇,再与句子中的其他成分作联系即可。比如用户输入“我想预定到北京的机票”,机器人可以识别出“我想”和“北京的机票”,然后在预置的知识数据库进行相关匹配处理。
例如,现在有些聊天机器人能够在用户的问题中识别出用户的意图,如订机票或酒店的场景中,机器人与用户聊天时,机器人接收用户输入的语音数据,将所述语音数据通过自然语言技术转化成文本信息,然后基于所述文本信息,在预设标准问答知识库中进行比对,查询是否存在用户所表达的意图对应的答案,然后输出查询结果。
步骤S20:在预置标准问答知识库中定位当前问题对应的节点;
本实施例中,在预置标准问答知识库中,首先确定当前所处流程环节,然后确定当前 输出的问题,进而定位当前问题的节点。在所述标准问答知识库中,根据具体业务场景,由业务方长期的工作经验中统计并在得出初版后根据实际机器人工作场景的情形对意图库做优化和补充,得到不同流程环节的问答知识,其中,可能在不同流程环节中存在相同的问题题目,或存在相同用户意图的连接关系,因而在查询回复当前用户意图对应关系的下一节点内容时需先确定当前问题的节点。
步骤S30:从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;
本实施例中,针对每个问题,用户会有不同的意图,不同的意图对应该问题不同的答案。一个问题的答案在预置的问答知识库以连接上下问题的连接关系表示,所以在定位出当前问题所处节点后,在该节点的下属连接关系中,匹配出与当前用户意图对应的目标连接关系。
例如,在咨询用户是否有购买车险的意向的场景中,预设问题的标准答案有“有意愿”或“无意愿”两种意图,针对“有意愿”的意图,对应的标准答案有“需要购买”、“想购买”等表达;针对“无意愿”对应的标准答案有“不需要”、“不想”等表达,而具体的而两种常用答案也会引发不同的下一个问题,如用户回答“有意愿”的意图表达,则获取与“有意愿”意图对应的连接关系的下一节点的问题,如询问用户户“目前有时间吗,为您介绍我司的最新车险产品”等;若用户表达“无意愿”,则获取与“无意愿”意图对应的目标连接关系的下一节点的问题,如询问用户“有无其他保险需要”或“那不打扰您了”。
步骤S40:定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
本实施例中,定位于所述连接关系相连的下一节点,该节点的文字描述信息与上一个问题和当前用户意图存在因果关系。根据用户意图的不同,在场景对话中会引发不同的问题,就像关于提问方提问“吃饭了吗”,对于标准答案中,回答方会有“正要去吃”和“已经吃了”,那么当提问方收到第一个意图“正要去吃”时,会引发提问方问回答方“打算或准备去吃什么”、“跟谁去吃”、“去哪吃呢”等问题,而如果提问方收到第二个意图“已经吃了”时,那么会引发提问方提问“吃什么了”、“跟谁吃的”、“在哪吃的”等问题,所以不同的意图,会引发不同的下一个问题,而引发的问题就会成为提问方在收到回答方所回答的内容之后,接下来提问的内容,基于这种关系,从预设的标准问答知识库中获取用户回答的所述第一答案关联的下一个节点设置的应答内容或操作。
例如,关于智能机器人提问“吃饭了没”,收到用户回答“吃过了”的意图表达,其中,“吃过了”的意图表达不限,则查询“吃饭了没”的问题所在节点,若在标准问答知识库中查询到所述节点,则识别所述节点中与意图“吃过了”相符的连接关系,获取所述连接关系对应的下一节点的文字描述内容并输出。
本实施例基于预先以图库的构建方式建立问答知识库,接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图,在预置标准问答知识库中定位当前问题的节点,从所述节点对应的连接关系中,匹配与当前用户意图对应的连接关系,定位与所述连接关系相连的下一节点,并输出该节点的文字描述信息作为回复用户的答复内容。本申请相比传统技术中以关系表的形式构建问答知识库来说,查询回复内容不需要繁杂的关系表,只需定位当前问题节点,匹配与用户意图对应连接关系的下一节点的文字描述信息,更快速的输出回复内容,提升用户体验。
进一步地,在本申请基于人际交互的数据处理方法中,所述当前用户意图包括主干意图和非主干意图,所述主干意图表示用户正面答复所述当前问题,所述非主干意图表示用户侧面答复所述当前问题。
本实施例中,由于同一个意图的表述形式有很多,例如用户针对“有没购买保险的需 要”的问题,用户回答的内容存在正面回答该问题的肯定意图“有”和否定意图“没有”等主干意图,也存在用户侧面答复该问题的非主干意图,如“没时间了解保险产品”、“不知道自己是否需要”等。所以需要预先根据具体业务场景,对用户针对每个问题可能出现的意图进行分类,分别建立主干意图库和非主干意图所对应的标准问答知识库和非标准问答知识库,然后根据用户意图,查询答复用户的应答内容。
参照图3,图3为本申请基于人机交互的数据处理方法第二实施例的流程示意图。本实施例中,在上述步骤S10之前,还包括:
步骤S001:获取预设应用场景关联的问题以及所述问题对应的标准答案,其中,一个问题至少对应两个标准答案,所述标准答案为主干意图的标准文字表达;
步骤S002:分别以各问题为节点、以各问题对应的标准答案作为连接各节点的连接关系,根据每个问题在预设应用场景中的提问顺序,通过所述标准答案将各问题连接成有向图,将所述有向图作为所述预置标准问答知识库。
本实施例中,不同的应用场景存在不同的问答数据。例如,在房产销售的应用场景中,销售人员与用户之间的问答内容会跟房产相关,如房子的类型、价格、位置等;在保险销售的应用场景中销售人员与用户之间的问答内容会跟保险相关,如保险的范围、理赔费率、保费保期等。随着智能问答机器人的推出,以智能机器人代替销售人员,为用户提供咨询服务,其中,常用问题及相关答复内容可从工作人员的日常工作中收集,然后由专家进行优化,进而得到标准的常用问答数据。
需要注意的是,工作人员在为用户提供咨询服务或销售服务的时候,根据用户答复内容所表达的意图,会引发销售人员回答或提出不一样的内容或问题,所以基于应用场景的问题与对应用户意图的答案之间存在一定的逻辑关系,整理每个常用问题与对应用户意图的答案之间的关系后,以各常用问题作为节点,各常用问题对应的答案作为连接各问题节点的关系,构建标准问答知识库。其中,所述标准问答知识库中,每个问题之间只有一个答案进行连接,以表示下一问题是由上一个问题其中一个答案所引发出的,而一个问题存在多个答案,每个答案只能引发一个问题,用户意图对应一个答案。
参照图4,图4为图2中步骤S20的细化流程示意图。本实施例中,上述步骤S30进一步包括:
步骤S201:接收语音数据并将所述语音数据输入预置语言识别模型进行语音识别,输出所述语音数据对应的文本信息;
本实施例中,为方便通过识别用户的语音答复,预先获取工作人员日常工作中与用户对话的语音训练数据,基于所述语音训练数据,训练一个语音识别模型。基于训练好的语音识别模型,接收用户针对当前问题作出答复的语音数据之后,将所语音数据输入语音识别模型进行语音识别,得到所述语音数据的文本信息。
步骤S202:对所述文本信息进行切词处理,得到多个切词片段;
本实施例中,对所述文本信息进行切词,可以利用现有技术中的任一种切词算法进行切词,得到多个切词片段。例如,通过n-gram切词算法对待自然语言文本“张先生需不需要车险”进行切词,得到切词片段分别为“张先生”、“需不需要”、“车险”。
步骤S203:依次将每个切词片段与预置知识词表进行匹配,得到至少一个概念标签,其中,所述概念标签用于映射切词片段的抽象概念;
本实施例中,基于切词片段,查询预置知识词库,匹配每个切词片段的概念,如果无法识别其中个别切词片段对应的概念,则跳过该切词片段,继续识别下一个切词片段,至少识别出一个切词片段对应的概念,并对匹配得到的概念对切词片段进行概念标注,得到分词片段的概念标签。需要注意的是,一个分词在一个问题中对应的概念至少一个,因此 最终可得到至少一个概念标签,其中,所述概念标签用于映射切词片段的抽象概念。此外,由于不可能将所有可能出现的切词片段对应的概念标签都存入预置的知识词表,所以存在个别切词片段未能匹配到概念标签,因而会出现个别切词片段无法识别。例如,切词片段“张先生”对应抽象的概念是用户,因此切词片段“张先生”对应的概念标签即为“用户”,同理,“需不要要”对应的概念标签是“需要”、“不需要”,“车险”对应的概念标签为“保险”。
步骤S204:对所述概念标签按照其对应的切词片段在所述文本信息中的表达顺序,按序排列并组合得到概念标签序列;
本实施例中,对上述步骤得到的概念标签按照其对应的切词片段在所述文本信息中的表达顺序,按序进行排列组合,得到多个不冲突的概念标签序列,即为同一文本中,同一个序列中只能出现一个概念标签所覆盖的切词。例如,切词片段“需不需要”对应的概念标签为“需要”、“不需要”,因此需要将这两个概念标签分到不同的概念标签中。例如,切词片段“张先生”、“需不需要”、“车险”对应的概念标签序列分别为“用户、需要、保险”和“用户、不需要、保险”两种概念。
步骤S205:基于预置意图知识网络,对所述概念标签序列进行意图匹配,确定当前用户意图,其中,所述所述预置意图知识网络由每个所述概念标签与每个所述概念标签所对应的意图标签的关联关系构成。
本实施例中,所述预先建立的意图知识网络是将现实世界的意图通过人的日常描述语言,映射生成的意图知识网络。具体的,意图知识网络中包括当前应用场景不同问题对应的意图表达集合,每种意图表达以标签的形式存储在意图知识网络中,且不同问题对应的意图标签集合之间相互关联。其中,每个意图标签集合中会记载与问题相关的、可能存在的意图标签集合,以及那些概念标签会与这些意图标签有关联。例如,对于保险,有询问保险保期、险种、保费等意图。
因而,只需将每个概念标签序列中的概念标签分别与意图知识网络进行匹配即可确定用户的真实意图。
参照图5,图5为图4中步骤S205的细化流程示意图。本实施例中,上述步骤S205包括:
骤S2051:分别将所述概念标签序列中的单个概念标签,或相邻的至少两个概念标签构成的组合标签,与所述意图知识网络进行匹配;
本实施例中,分别将概念标签序列中的单个概念标签,与意图知识网络进行匹配,或将相邻多个概念标签与所述意图知识网络进行匹配。其中,考虑到可能存在部分意图标签需要多个概念标签组合才能匹配得到,因而将相邻多个概念标签组合,进而与所述意图知识网络进行匹配。例如,对于文本“张先生需不需要车险”经过切词和概念标签后对应的概念标签序列分别为“用户、需要、保险”和“用户、不需要、保险”。而“用户、不需要、保险”还可能是“用户、不、需要、保险”,那么其中的“不、需要”的相邻概念标签就需要组合,才能匹配出“不需要”的意图标签。将“用户、需要、保险”与意图知识网络进行匹配,以便推导该概念标签序列对应的用户意图;再将“用户、不需要、保险”与意图知识网络进行匹配,以便推导该概念标签序列对应的用户意图。
需要注意的是,按照与所述文本相同的表达顺序,目的是正确匹配用户意图,因为若将概念标签顺序打乱,则不符合人类自然语言的表达习惯,自然地,对乱序的概念标签进行匹配也不会匹配出正确的意图结果。
步骤S2052:若存在概念标签匹配到意图标签,则获取所述意图标签并将所述意图标签与尚未进行匹配的概念标签合并为新的概念标签序列,对该新的概念标签序列继续进行意图标签匹配,直到所述概念标签序列中的全部概念标签匹配完毕或匹配不到意图标签为 止,输出最终匹配结果,其中,所述最终匹配结果为一个或多个意图标签组合而成的意图标签集;
步骤S2053:以所述最终匹配结果确定当前用户意图。
本实施例中,对于第一个概念标签序列“用户、需要、保险”,根据概念标签“用户”和“需要”与意图知识网络的匹配结果,确定有“需要”的意图标签,则获取该意图标签,即“需要”意图标签,并返回用户对应的概念标签“用户”;将概念标签“用户”、意图标签“需要”和未进行匹配的概念标签“保险”合并为新的概念标签序列,并与意图知识网络进行匹配,得出“用户需要保险”的意图标签。此时概念标签全部使用,无法继续匹配,则确定最终意图为“需要保险”意图。
参照图6,图6为本申请基于人机交互的数据处理方法第三实施例的流程示意图。
步骤S60:若匹配不出与当前用户意图对应的目标连接关系,则判定所述当前用户意图为非主干意图,查询预置通用问答知识库是否存在与当前用户意图对应的非标准答案,其中,所述预置通用问答知识库为存储预设应用场景关联的非主干意图对应的非标准答案以及所述非标准答案关联的答复内容的数据库,所述非标准答案表示用户非主干意图的标准文字表达;
本实施例中,由于现实应用场景中,用户的意图表达非常多,不可能将用户的所有意图都配置到标准问答知识库中,因此存在标准问答知识库中,没有当前用户意图对应的目标连接关系,即表示用户所答复的内容所表达的意图不是主干意图,则调用接口查询预置的通用问答知识库,所述通用问答知识库中存储有用户对于标准问答知识库中每个问题可能出现的非主干意图所对应的非标准答案,所述非标准答案即为非主干意图的标准文字表达。
进一步地,在现实应用场景中,用户针对一个问题除了主干意图之外,还存在非主干意图,此处的非主干意图指主干意图之外的意图,如在“有无意愿购买车险”的问题中,用户除了“有意愿”和“无意愿”两种主干意图对应的标准答案之外,还可能会回答“没钱”、“没空了解”、“已购买”等非主干意图对应的非标准答案,针对这种情况,将这类非主干意图的数据从常用问答数据中抽离,并基于各非主干意图与对应问题之间的关系,建立关系表作为通用问答知识库。
例如,在“有无意愿购买车险”的问题中,用户除了“有意愿”和“无意愿”两种标准答案之外,还可能会回答“没钱”、“没空了解”、“已购买”等非标准答案,筛选出“没钱”、“没空了解”、“已购买”等非标准答案,建立非标准答案“没钱”、“没空了解”、“已购买”与问题“有无意愿购买车险”的关联表,当用户答案在问答知识库中为匹配到相应答案时,则在所述通用问答知识库。
步骤S70:若所述预置通用问答知识库存在与当前用户意图对应的非标准答案,则输出与所述非标准答案关联的答复内容;
步骤S80:若所述预置通用问答知识库不存在与当前用户意图对应的非标准答案,则输出预置应答内容。
本实施例中,若在标准问答知识库中查询不到与当前用户意图对应的连接关系,表示用户当前答复内容所表达的意图为非主干意图,则查询所述通用问答知识库,确定所述通用问答知识库中是否存在与当前用户意图对应的非标准答案,若所述通用问答知识库存在与当前用户意图对应的非标准答案,则输出与所述非标准答案关联的答复内容。若所述通用问答知识库中不存在当前用户意图对应的非标准答案,则表明通用问答知识库中未配置与当前用户意图对应的非标准答案,也即未配置有关联的回复内容,则输出预先设置的统一应答内容。
例如,某一节点为询问用户“有无意愿购买车险”的问题,在标准问答知识库中该节 点的预设标准答案对应意图为“有意愿”和“无意愿”,若当前接收到的用户意图为“没钱”,则所述标准问答知识库中当前问题节点不存在与当前用户意图对应的答案,则查询通用问答知识库中是否存在与“没钱”对应的非标准答案,而查询到通用问答知识库中询问用户“有无意愿购买车险”问题的节点预设有“没钱”、“没空了解”、“已购买”等意图对应的非标准答案,因而在通用问答知识库中匹配用户意图“没钱”的关联的应答内容,其中应答内容可能是“这份保险不需要多少钱,这边有个优惠...”等,此外,还可能是内容“那就不打扰您了”,结束本次服务。
本申请还提供一种数据处理装置。
参照图7,图7为本申请基于人机交互的数据处理装置一实施例的功能模块示意图。本实施例中,所述数据处理装置包括:
解析模块10,用于接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图;
定位模块20,用于在预置标准问答知识库中定位当前问题对应的节点;
匹配模块30,用于从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;
第一输出模块40,用于定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
可选地,所述解析模块10包括:
识别单元,用于接收语音数据并将所述语音数据输入预置语言识别模型进行语音识别,输出所述语音数据对应的文本信息;
切词单元,用于对所述文本信息进行切词处理,得到多个切词片段;
标注单元,用于依次将每个切词片段与预置知识词表进行匹配,得到至少一个概念标签,其中,所述概念标签用于映射切词片段的抽象概念;
排列单元,用于对所述概念标签按照其对应的切词片段在所述文本信息中的表达顺序,按序排列并组合得到概念标签序列;
匹配单元,用于基于预置意图知识网络,对所述概念标签序列进行意图匹配,确定当前用户意图,其中,所述预置意图知识网络由每个所述概念标签与每个所述概念标签所对应的意图标签的关联关系构成。
可选地,所述匹配单元包括:
匹配子单元,用于分别将所述概念标签序列中的单个概念标签,或相邻的至少两个概念标签构成的组合标签,与所述意图知识网络进行匹配;
输出子单元,用于若存在概念标签匹配到意图标签,则获取所述意图标签并将所述意图标签与尚未进行匹配的概念标签合并为新的概念标签序列,对该新的概念标签序列继续进行意图标签匹配,直到当前概念标签序列中的全部概念标签匹配完毕或匹配不到意图标签为止,输出最终匹配结果,其中,所述最终匹配结果为一个或多个意图标签组合而成的意图标签集;
确定子单元,用于以所述最终匹配结果确定当前用户意图。
可选地,所述数据处理装置包括:
查询模块,用于若匹配不出与当前用户意图对应的目标连接关系,则判定所述当前用户意图为非主干意图,查询预置通用问答知识库是否存在与当前用户意图对应的非标准答案,其中,所述预置通用问答知识库为存储预设应用场景关联的非主干意图对应的非标准答案以及所述非标准答案关联的答复内容的数据库,所述非标准答案表示用户非主干意图的标准文字表达;
第二输出模块,用于若所述预置通用问答知识库存在与当前用户意图对应的非标准答 案,则输出与所述非标准答案关联的答复内容;
第三输出模块,用于若所述预置通用问答知识库不存在与当前用户意图对应的非标准答案,则输出预置应答内容。
本实施例中,解析模块10接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图,定位模块20在预置标准问答知识库中定位当前问题对应的节点,第一匹配模块30从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,第一输出模块40定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
本申请还提供一种计算机可读存储介质。
本实施例中,所述计算机可读存储介质上存储有数据处理程序,所述计算机可读存储介质可以是非易失性,也可以是易失性,所述数据处理程序被处理器执行时实现如上述任一项实施例中所述的数据处理方法的步骤。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM)中,包括若干指令用以使得一台终端(可以是手机,计算机,服务器或者网络设备等)执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述,但是本申请并不局限于上述的具体实施方式,上述的具体实施方式仅仅是示意性的,而不是限制性的,本领域的普通技术人员在本申请的启示下,在不脱离本申请宗旨和权利要求所保护的范围情况下,还可做出很多形式,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,这些均属于本申请的保护之内。
Claims (20)
- 一种基于人机交互的数据处理方法,其中,所述数据处理方法包括:接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图;在预置标准问答知识库中定位当前问题对应的节点;从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
- 如权利要求1所述的数据处理方法,其中,所述当前用户意图包括主干意图和非主干意图,所述主干意图表示用户正面答复所述当前问题,所述非主干意图表示用户侧面答复所述当前问题。
- 如权利要求2所述的数据处理方法,其中,在所述接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图的步骤之前,还包括:获取预设应用场景关联的问题以及所述问题对应的标准答案,其中,一个问题至少对应两个标准答案,所述标准答案为主干意图的标准文字表达;分别以各问题为节点、以各问题对应的标准答案作为连接各节点的连接关系,根据每个问题在预设应用场景中的提问顺序,通过所述标准答案将各问题连接成有向图,将所述有向图作为所述预置标准问答知识库。
- 如权利要求1所述的数据处理方法,其中,所述接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图包括:接收语音数据并将所述语音数据输入预置语言识别模型进行语音识别,输出所述语音数据对应的文本信息;对所述文本信息进行切词处理,得到多个切词片段;依次将每个切词片段与预置知识词表进行匹配,得到至少一个概念标签,其中,所述概念标签用于映射切词片段的抽象概念;对所述概念标签按照其对应的切词片段在所述文本信息中的表达顺序,按序排列并组合得到概念标签序列;基于预置意图知识网络,对所述概念标签序列进行意图匹配,确定当前用户意图,其中,所述预置意图知识网络由每个所述概念标签与每个所述概念标签所对应的意图标签的关联关系构成。
- 如权利要求4所述的数据处理方法,其中,所述基于预置意图知识网络,对所述概念标签序列进行意图匹配,确定当前用户意图包括:分别将所述概念标签序列中的单个概念标签,或相邻的至少两个概念标签构成的组合标签,与所述预置意图知识网络进行匹配;若存在概念标签匹配到意图标签,则获取所述意图标签并将所述意图标签与尚未进行匹配的概念标签合并为新的概念标签序列,对该新的概念标签序列继续进行意图标签匹配,直到所述概念标签序列中的全部概念标签匹配完毕或匹配不到意图标签为止,输出最终匹配结果,其中,所述最终匹配结果为一个或多个意图标签组合而成的意图标签集;以所述最终匹配结果确定当前用户意图。
- 如权利要求2所述的数据处理方法,其中,所述数据处理方法还包括:若匹配不出与当前用户意图对应的目标连接关系,则判定所述当前用户意图为非主干意图,查询预置通用问答知识库是否存在与当前用户意图对应的非标准答案,其中,所述预置通用问答知识库为存储预设应用场景关联的非主干意图对应的非标准答案以及所述非标准答案关联的答复内容的数据库,所述非标准答案为非主干意图的文字表达;若所述预置通用问答知识库存在与当前用户意图对应的非标准答案,则输出与所述非标准答案关联的答复内容;若所述预置通用问答知识库不存在与当前用户意图对应的非标准答案,则输出预置应答内容。
- 一种基于人机交互的数据处理设备,其中,所述数据处理设备包括存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的数据处理程序,所述数据处理程序被所述处理器执行时实现如下的基于人机交互的数据处理方法的步骤:接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图;在预置标准问答知识库中定位当前问题对应的节点;从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
- 如权利要求7所述的基于人机交互的数据处理设备,其中,所述当前用户意图包括主干意图和非主干意图,所述主干意图表示用户正面答复所述当前问题,所述非主干意图表示用户侧面答复所述当前问题。
- 如权利要求8所述的基于人机交互的数据处理设备,其中,所述基于人机交互的数据处理的程序被所述处理器执行在所述接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图的步骤之前,还执行以下步骤:获取预设应用场景关联的问题以及所述问题对应的标准答案,其中,一个问题至少对应两个标准答案,所述标准答案为主干意图的标准文字表达;分别以各问题为节点、以各问题对应的标准答案作为连接各节点的连接关系,根据每个问题在预设应用场景中的提问顺序,通过所述标准答案将各问题连接成有向图,将所述有向图作为所述预置标准问答知识库。
- 如权利要求7所述的基于人机交互的数据处理设备,其中,所述基于人机交互的数据处理的程序被所述处理器执行所述接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图的步骤时,包括以下步骤:接收语音数据并将所述语音数据输入预置语言识别模型进行语音识别,输出所述语音数据对应的文本信息;对所述文本信息进行切词处理,得到多个切词片段;依次将每个切词片段与预置知识词表进行匹配,得到至少一个概念标签,其中,所述概念标签用于映射切词片段的抽象概念;对所述概念标签按照其对应的切词片段在所述文本信息中的表达顺序,按序排列并组合得到概念标签序列;基于预置意图知识网络,对所述概念标签序列进行意图匹配,确定当前用户意图,其中,所述预置意图知识网络由每个所述概念标签与每个所述概念标签所对应的意图标签的关联关系构成。
- 如权利要求10所述的基于人机交互的数据处理设备,其中,所述基于人机交互的数据处理的程序被所述处理器执行所述基于预置意图知识网络,对所述概念标签序列进行意图匹配,确定当前用户意图的步骤时,包括以下步骤:分别将所述概念标签序列中的单个概念标签,或相邻的至少两个概念标签构成的组合标签,与所述预置意图知识网络进行匹配;若存在概念标签匹配到意图标签,则获取所述意图标签并将所述意图标签与尚未进行匹配的概念标签合并为新的概念标签序列,对该新的概念标签序列继续进行意图标签匹配,直到所述概念标签序列中的全部概念标签匹配完毕或匹配不到意图标签为止,输出最 终匹配结果,其中,所述最终匹配结果为一个或多个意图标签组合而成的意图标签集;以所述最终匹配结果确定当前用户意图。
- 如权利要求8所述的基于人机交互的数据处理设备,其中,所述基于人机交互的数据处理的程序被所述处理器执行所述的基于人机交互的数据处理方法的步骤时,包括以下步骤:若匹配不出与当前用户意图对应的目标连接关系,则判定所述当前用户意图为非主干意图,查询预置通用问答知识库是否存在与当前用户意图对应的非标准答案,其中,所述预置通用问答知识库为存储预设应用场景关联的非主干意图对应的非标准答案以及所述非标准答案关联的答复内容的数据库,所述非标准答案为非主干意图的文字表达;若所述预置通用问答知识库存在与当前用户意图对应的非标准答案,则输出与所述非标准答案关联的答复内容;若所述预置通用问答知识库不存在与当前用户意图对应的非标准答案,则输出预置应答内容。
- 一种计算机可读存储介质,其中,所述计算机可读存储介质上存储有数据处理程序,所述数据处理程序被处理器执行时实现如下基于人机交互的数据处理方法的步骤:接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图;在预置标准问答知识库中定位当前问题对应的节点;从所述节点对应的连接关系中,匹配与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
- 根据权利要求13所述的计算机可读存储介质,其中,所述当前用户意图包括主干意图和非主干意图,所述主干意图表示用户正面答复所述当前问题,所述非主干意图表示用户侧面答复所述当前问题。
- 根据权利要求14所述的计算机可读存储介质,其中,所述数据处理程序被处理器执行所述接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图的步骤之前,还执行以下步骤:获取预设应用场景关联的问题以及所述问题对应的标准答案,其中,一个问题至少对应两个标准答案,所述标准答案为主干意图的标准文字表达;分别以各问题为节点、以各问题对应的标准答案作为连接各节点的连接关系,根据每个问题在预设应用场景中的提问顺序,通过所述标准答案将各问题连接成有向图,将所述有向图作为所述预置标准问答知识库。
- 根据权利要求15所述的计算机可读存储介质,其中,所述数据处理程序被处理器执行所述接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图的步骤时,包括以下步骤:接收语音数据并将所述语音数据输入预置语言识别模型进行语音识别,输出所述语音数据对应的文本信息;对所述文本信息进行切词处理,得到多个切词片段;依次将每个切词片段与预置知识词表进行匹配,得到至少一个概念标签,其中,所述概念标签用于映射切词片段的抽象概念;对所述概念标签按照其对应的切词片段在所述文本信息中的表达顺序,按序排列并组合得到概念标签序列;基于预置意图知识网络,对所述概念标签序列进行意图匹配,确定当前用户意图,其中,所述预置意图知识网络由每个所述概念标签与每个所述概念标签所对应的意图标签的关联关系构成。
- 根据权利要求16所述的计算机可读存储介质,其中,所述数据处理程序被处理器执行所述基于预置意图知识网络,对所述概念标签序列进行意图匹配,确定当前用户意图的步骤时,包括以下步骤:分别将所述概念标签序列中的单个概念标签,或相邻的至少两个概念标签构成的组合标签,与所述预置意图知识网络进行匹配;若存在概念标签匹配到意图标签,则获取所述意图标签并将所述意图标签与尚未进行匹配的概念标签合并为新的概念标签序列,对该新的概念标签序列继续进行意图标签匹配,直到所述概念标签序列中的全部概念标签匹配完毕或匹配不到意图标签为止,输出最终匹配结果,其中,所述最终匹配结果为一个或多个意图标签组合而成的意图标签集;以所述最终匹配结果确定当前用户意图。
- 根据权利要求14所述的计算机可读存储介质,其中,所述数据处理程序被处理器执行如下所述的数据处理方法的步骤:若匹配不出与当前用户意图对应的目标连接关系,则判定所述当前用户意图为非主干意图,查询预置通用问答知识库是否存在与当前用户意图对应的非标准答案,其中,所述预置通用问答知识库为存储预设应用场景关联的非主干意图对应的非标准答案以及所述非标准答案关联的答复内容的数据库,所述非标准答案为非主干意图的文字表达;若所述预置通用问答知识库存在与当前用户意图对应的非标准答案,则输出与所述非标准答案关联的答复内容;若所述预置通用问答知识库不存在与当前用户意图对应的非标准答案,则输出预置应答内容。
- 一种基于人机交互的数据处理装置,其中,所述数据处理装置包括:解析模块,用于接收用户针对当前问题答复的语音数据并进行意图解析,以确定当前用户意图,所述用户意图包括主干意图和非主干意图,所述主干意图包括肯定意图和否定意图;定位模块,用于在预置标准问答知识库中定位当前问题对应的节点;匹配模块,用于从所述节点对应的连接关系中,匹配出与当前用户意图对应的目标连接关系,其中,一条所述连接关系对应问题的一个标准答案;第一输出模块,用于定位与所述目标连接关系相连的目标节点,并输出所述目标节点在预置标准问答知识库中对应的问题。
- 如权利要求7所述的数据处理装置,其中,所述数据处理装置还包括:获取模块,用于获取预设应用场景关联的问题以及所述问题对应的标准答案,其中,一个问题至少对应两个标准答案,所述标准答案为表示用户主干意图的标准文字表达;建立模块,用于分别以各问题为节点、以各问题对应的标准答案作为连接各节点的连接关系,根据每个问题在预设应用场景中的提问顺序,通过所述标准答案将各问题连接成有向图,将所述有向图作为所述预置标准问答知识库。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010017461.5 | 2020-01-08 | ||
CN202010017461.5A CN111241260A (zh) | 2020-01-08 | 2020-01-08 | 基于人机交互的数据处理方法、装置、设备及存储介质 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021139243A1 true WO2021139243A1 (zh) | 2021-07-15 |
Family
ID=70880313
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/118431 WO2021139243A1 (zh) | 2020-01-08 | 2020-09-28 | 基于人机交互的数据处理方法、装置、设备及存储介质 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN111241260A (zh) |
WO (1) | WO2021139243A1 (zh) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111241260A (zh) * | 2020-01-08 | 2020-06-05 | 平安科技(深圳)有限公司 | 基于人机交互的数据处理方法、装置、设备及存储介质 |
CN113742480A (zh) * | 2020-06-18 | 2021-12-03 | 北京汇钧科技有限公司 | 客服应答方法和装置 |
CN111737437B (zh) * | 2020-06-30 | 2024-06-28 | 北京捷通华声科技股份有限公司 | 问答知识的提取方法、提取装置与计算机可读存储介质 |
CN111914073A (zh) * | 2020-07-15 | 2020-11-10 | 中国联合网络通信集团有限公司 | 客服应答方法、装置、设备及存储介质 |
CN112035623B (zh) * | 2020-09-11 | 2023-08-04 | 杭州海康威视数字技术股份有限公司 | 智能问答方法、装置、电子设备及存储介质 |
CN112417110A (zh) * | 2020-10-27 | 2021-02-26 | 联想(北京)有限公司 | 一种信息处理方法及装置 |
CN112667790B (zh) * | 2020-12-18 | 2024-05-10 | 平安科技(深圳)有限公司 | 智能问答方法、装置、设备及存储介质 |
CN115186674B (zh) * | 2022-06-20 | 2024-07-12 | 成都飞机工业(集团)有限责任公司 | 一种航空失效案例管理方法、装置、设备及存储介质 |
CN115860823B (zh) * | 2023-03-03 | 2023-05-16 | 深圳市人马互动科技有限公司 | 人机互动问卷答题场景中的数据处理方法及相关产品 |
CN118535715B (zh) * | 2024-07-26 | 2024-10-29 | 四川声通玄武信息科技有限公司 | 一种基于树状结构知识库的自动回复方法、设备及存储介质 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108090177A (zh) * | 2017-12-15 | 2018-05-29 | 上海智臻智能网络科技股份有限公司 | 多轮问答系统的生成方法、设备、介质及多轮问答系统 |
CN108108340A (zh) * | 2017-11-28 | 2018-06-01 | 北京光年无限科技有限公司 | 用于智能机器人的对话交互方法及系统 |
CN109543014A (zh) * | 2018-11-05 | 2019-03-29 | 腾讯科技(深圳)有限公司 | 人机对话方法、装置、终端及服务器 |
US20190386938A1 (en) * | 2018-05-23 | 2019-12-19 | Capital One Services, Llc | Method and System of Converting Email Message to AI Chat |
CN111241260A (zh) * | 2020-01-08 | 2020-06-05 | 平安科技(深圳)有限公司 | 基于人机交互的数据处理方法、装置、设备及存储介质 |
-
2020
- 2020-01-08 CN CN202010017461.5A patent/CN111241260A/zh active Pending
- 2020-09-28 WO PCT/CN2020/118431 patent/WO2021139243A1/zh active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108108340A (zh) * | 2017-11-28 | 2018-06-01 | 北京光年无限科技有限公司 | 用于智能机器人的对话交互方法及系统 |
CN108090177A (zh) * | 2017-12-15 | 2018-05-29 | 上海智臻智能网络科技股份有限公司 | 多轮问答系统的生成方法、设备、介质及多轮问答系统 |
US20190386938A1 (en) * | 2018-05-23 | 2019-12-19 | Capital One Services, Llc | Method and System of Converting Email Message to AI Chat |
CN109543014A (zh) * | 2018-11-05 | 2019-03-29 | 腾讯科技(深圳)有限公司 | 人机对话方法、装置、终端及服务器 |
CN111241260A (zh) * | 2020-01-08 | 2020-06-05 | 平安科技(深圳)有限公司 | 基于人机交互的数据处理方法、装置、设备及存储介质 |
Also Published As
Publication number | Publication date |
---|---|
CN111241260A (zh) | 2020-06-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021139243A1 (zh) | 基于人机交互的数据处理方法、装置、设备及存储介质 | |
CN110008322B (zh) | 多轮对话场景下的话术推荐方法和装置 | |
US11315560B2 (en) | Method for conducting dialog between human and computer | |
US20200301954A1 (en) | Reply information obtaining method and apparatus | |
JP6604836B2 (ja) | 対話テキスト要約装置及び方法 | |
CN107506372A (zh) | 一种机器人客服在混合类型会话下的自动会话切换方法 | |
CN104598445B (zh) | 自动问答系统和方法 | |
WO2018006727A1 (zh) | 机器人客服转人工客服的方法和装置 | |
CN107153965A (zh) | 一种多终端的智能客服解决方法 | |
US10395641B2 (en) | Modifying a language conversation model | |
WO2020253064A1 (zh) | 语音的识别方法及装置、计算机设备、存储介质 | |
CN106407178A (zh) | 一种会话摘要生成方法及装置 | |
JP2020521210A (ja) | 情報処理方法及び端末、コンピュータ記憶媒体 | |
CN110580516B (zh) | 一种基于智能机器人的交互方法及装置 | |
CN110266900A (zh) | 客户意图的识别方法、装置及客服系统 | |
WO2015196987A1 (zh) | 支持自然语言的数据查询方法、开放平台及用户终端 | |
CN115470338B (zh) | 一种基于多路召回的多场景智能问答方法和系统 | |
US20230043260A1 (en) | Persisting an AI-supported conversation across multiple channels | |
CN110457454A (zh) | 一种对话方法、服务器、对话系统及存储介质 | |
CN113407677B (zh) | 评估咨询对话质量的方法、装置、设备和存储介质 | |
CN114490975A (zh) | 用户问题的标注方法及装置 | |
WO2021135103A1 (zh) | 一种语义分析方法、装置、计算机设备及存储介质 | |
CN117874163A (zh) | 推理提示方法、答案推理方法、装置、设备与存储介质 | |
US12045780B2 (en) | Online interview method and system | |
CN111556096B (zh) | 信息推送方法、装置、介质及电子设备 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20912931 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20912931 Country of ref document: EP Kind code of ref document: A1 |