Man-machine conversation semantic understanding method and system

Abstract

The invention discloses a man-machine dialogue semantic understanding method, which comprises the following steps that: a robot performs man-machine dialogue with a client, and receives dialogue content input by the client; the recognition engine recognizes the dialogue content through the recognition model; the recognition engine performs intention recognition on the dialogue content based on the knowledge base, and judges and obtains the client intention; man-machine conversation knowledge background related data is stored in the knowledge base; and the robot judges branches in the flow chart which needs to be executed by the robot according to the obtained client intention, and determines the node jumping direction of the flow chart according to an intention recognition result, so that the man-machine conversation process is promoted. In addition, the invention also discloses a man-machine conversation semantic understanding system. According to the method, the scene-based recognition model is added into the man-machine dialogue system, configuration is performed in combination with the knowledge base, iteration is performed on the knowledge base through automatic variable extraction, and more comprehensive and more accurate semantic understanding is realized through customer intention recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of call centers, and in particular to a method and system for understanding the semantics of human-computer dialogue. Background Art

[0002] In the existing technology, voice robots often have difficulty understanding customers' questions and inquiries in situations involving ambiguous sentences, complex proper nouns, etc. during human-computer interaction. Traditional methods also have difficulty processing new words and knowledge proposed by customers, and are unable to read new words or knowledge in real time and automatically iterate the system, resulting in a poor customer experience. Summary of the invention

[0003] Based on this, in order to solve the technical problems in the existing technology, a human-computer dialogue semantic understanding system is proposed, including a robot and a knowledge base;

[0004] The robot conducts a human-machine dialogue with the customer and receives the dialogue content input by the customer. The robot has a recognition engine, and the recognition engine uses a recognition model to recognize the dialogue content it receives. The recognition engine recognizes the intention of the dialogue content input by the customer based on the knowledge base, and determines and obtains the customer's intention.

[0005] The knowledge base stores data related to the knowledge background of human-computer dialogue, and the data source of the knowledge base includes system presets or input parameters in the human-computer dialogue process; the knowledge base includes one or more sub-bases, and the sub-bases associated with intent recognition include intent base, entity base, synonym base, and hot word base;

[0006] The robot determines the branch in the flowchart that it needs to execute next based on the acquired customer intention, and determines the node jump direction of the flowchart based on the intention recognition result, thereby advancing the human-computer dialogue process; wherein, the flowchart is composed of one or more nodes and the logical configuration within each node.

[0007] In one embodiment, the recognition engine uses the recognition model to recognize the conversation content it receives, including speech recognition. When the conversation content input by the customer is speech, the recognition engine first performs speech recognition on the speech input by the customer, and then performs intent recognition on the speech recognition result in combination with the knowledge base to determine and obtain the customer's intent;

[0008] Among them, the recognition model is obtained by pre-training of the recognition engine; the recognition model is trained based on different scenarios, and different recognition models are selected for training according to different business needs of different nodes in the human-computer dialogue.

[0009] In one embodiment, the human-computer dialogue semantic understanding system also includes a knowledge base manager, which is used to manage data in the knowledge base; wherein the knowledge base manager includes an intent library template, an entity library template, a synonym library template, and a hot word library template, which are respectively used to configure data attributes in the intent library, entity library, synonym library, and hot word library.

[0010] In one embodiment, the intention library contains the customer intention that can be judged by the robot, and the customer intention is the key meaning of a conversation between the customer and the robot; the customer intention corresponds to one or more identification methods, including keyword identification, similar question identification, and regular verification;

[0011] The intent data in the intent library includes custom intents and system preset intents. The system preset intents are system preset intent data stored by default.

[0012] Use the knowledge base manager to configure one or more intent data attributes in its intent base template. The intent data attributes include intent name, keywords, similar questions, and regular expression verification.

[0013] Among them, keywords are generated through customization or with the assistance of a large model; among them, similar questions are generated through customization or with the assistance of a large model to generate similar questions of customers with the same intent; among them, regular verification is to use regular expressions to express recognizable questions, combined with referencing entities in the entity library to improve the accuracy of semantic recognition.

[0014] In one embodiment, the entity library is a user-defined database for storing entity sets involved in the business; when setting the intent recognition method, entities in the entity library are referenced to extract corresponding information;

[0015] The entity sets in the entity library include user entities generated by the user and system entities preset by the system. System entities are stored in the system preset entity set by default.

[0016] Configure one or more entity attributes of an entity set in its entity library template through the knowledge base manager, where the entity attributes include entity name, entity description, entity value, and regular expression validation;

[0017] Among them, the entity value can be customized, that is, the entity value contained in the entity can be customized; among them, regular verification is to use regular expressions to add self-references or references to other entity sets in the entity library to express the entity values ​​it contains.

[0018] In one embodiment, the synonym database includes one or more pairs of standard words and synonyms with corresponding associations. One standard word corresponds to multiple synonyms. The robot uses the synonym database to deal with the situation where the same intention has multiple expressions.

[0019] One or more synonym attributes of the synonym set are configured in the synonym database template through the knowledge base manager, and the synonym attributes include standard words, synonyms, and synonym conversion;

[0020] The standard word corresponds to one or more synonyms; the synonyms contained in the synonym set are generated by customization; when the synonym conversion property of the synonym set is turned on, the standard word is used as the name of the synonym set.

[0021] In one embodiment, the hot word library includes pre-set hot words and hot words passed in through parameters; the hot words are specific proper nouns in specific businesses; the hot word library is iterated in real time in the human-computer dialogue process, and the variable values ​​and parameter values ​​extracted in the dialogue are fed back to the recognition engine, and the recognition engine learns and updates the recognition model according to the fed-back variable values ​​and parameter values ​​to deal with proper nouns;

[0022] Use the knowledge base manager to configure one or more hot word attributes of the hot word set in its hot word base template. Hot word attributes include hot word category, hot word parameter value, hot word variable value, and hot word text:

[0023] Among them, the hot word category is divided into global hot words or node hot words according to the recognition engine scope of the service; global hot words serve all recognition engines in the human-computer dialogue process, and global hot words are set through the hot word library template page; node hot words serve the recognition engines of one or more specific nodes in the human-computer dialogue process, and node hot words are set through the user input configuration page of the node in the flow chart;

[0024] Among them, the hot word parameter value is passed in from the outside, the hot word variable value is extracted from the current human-computer dialogue, and the hot word text is obtained through customization.

[0025] In addition, in order to solve the technical problems in the prior art, a method for semantic understanding of human-computer dialogue is proposed, including:

[0026] The robot conducts a human-machine dialogue with the customer and receives the dialogue content input by the customer;

[0027] The robot has a recognition engine, which uses a recognition model to recognize the conversation content it receives; the recognition engine recognizes the intention of the conversation content input by the customer based on the knowledge base, and determines and obtains the customer's intention;

[0028] The knowledge base stores data related to the knowledge background of human-computer dialogue. The data sources of the knowledge base include system presets or input parameters in the human-computer dialogue process. The knowledge base includes one or more sub-bases, among which the sub-bases associated with intent recognition include intent base, entity base, synonym base, and hot word base.

[0029] The robot determines the branch in the flowchart that it needs to execute next based on the acquired customer intention, and determines the node jump direction of the flowchart based on the intention recognition result, thereby advancing the human-computer dialogue process; wherein, the flowchart is composed of one or more nodes and the logical configuration within each node.

[0030] In one embodiment, the recognition engine uses the recognition model to recognize the conversation content it receives, including speech recognition. When the conversation content input by the customer is speech, the recognition engine first performs speech recognition on the speech input by the customer, and then performs intent recognition on the speech recognition result in combination with the knowledge base to determine and obtain the customer's intent;

[0031] Among them, the recognition model is obtained by pre-training of the recognition engine; the recognition model is trained based on different scenarios, and different recognition models are selected for training according to different business needs of different nodes in the human-computer dialogue.

[0032] In one embodiment, the recognition engine selects and sets a recognition method based on the intent library, and the recognition method includes keyword recognition, similar question recognition, and regular expression verification;

[0033] When the recognition method is keyword recognition, the recognition engine uses the keywords stored in the intent library to recognize the intent, or calls the synonyms or hot words in the synonym library and hot word library to recognize the intent;

[0034] When the recognition method is similar question recognition, the recognition engine uses similar questions stored in the intent library to perform intent recognition;

[0035] Among them, when the recognition method is regular verification, the recognition engine calls the entity library to perform self-reference regular verification for intent recognition.

[0036] Implementing the embodiments of the present invention will have the following beneficial effects:

[0037] The present invention adds a scenario-based recognition model to the human-computer dialogue system and configures it in combination with a knowledge base, iterates the knowledge base through automatic variable extraction, and uses customer intent recognition to achieve more comprehensive and accurate semantic understanding. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0039] in:

[0040] Figure 1 It is a flowchart of the method for understanding the semantics of human-computer dialogue in the present invention;

[0041] Figure 2 It is a schematic diagram of the human-computer dialogue semantic understanding system in the present invention; DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] The present invention discloses a method for understanding the semantics of human-computer dialogue, comprising:

[0044] The robot conducts a human-machine dialogue with the customer and receives the dialogue content input by the customer;

[0045] The robot has a recognition engine, which uses a recognition model to recognize the conversation content it receives;

[0046] Among them, the recognition engine recognizes the intention of the conversation content input by the customer based on the knowledge base, judges and obtains the customer's intention;

[0047] In particular, the recognition engine uses the recognition model to recognize the conversation content it receives, including speech recognition. When the conversation content input by the customer is speech, the recognition engine first performs speech recognition on the voice input by the customer, and then performs intent recognition on the speech recognition result in combination with the knowledge base to determine and obtain the customer's intention;

[0048] The recognition model is obtained by pre-training the recognition engine. The recognition model is trained based on different scenarios, and different recognition models are selected for training according to different business requirements of different nodes in the human-computer dialogue.

[0049] For example, identification models include banking models and address models;

[0050] The knowledge base stores data related to the knowledge background of human-computer dialogue. The data sources of the knowledge base include system presets or input parameters in the human-computer dialogue process. The robot semantically understands the dialogue content input by the customer, and associates the dialogue content with the data related to the customer's intention in the knowledge base to achieve intention recognition.

[0051] The knowledge base includes one or more sub-bases, wherein the sub-bases associated with intent recognition include an intent base, an entity base, a synonym base, and a hot word base;

[0052] The knowledge base manager is used to manage the data in the knowledge base; the knowledge base manager includes an intent base template, an entity base template, a synonym base template, and a hot word base template, which are used to configure the data attributes in the intent base, entity base, synonym base, and hot word base respectively;

[0053] The intent library contains the customer intent that the robot can judge. The customer intent is the key meaning of a conversation between the customer and the robot. The customer intent corresponds to one or more recognition methods, including keyword recognition, similar question recognition, and regular expression verification.

[0054] In particular, the intent data in the intent library includes custom intents and system preset intents. The system preset intents are system preset intent data stored by default.

[0055] Use the knowledge base manager to configure one or more intent data attributes in its intent base template. The intent data attributes include intent name, keywords, similar questions, and regular expression verification.

[0056] Among them, keywords are generated by self-definition or with the assistance of large models;

[0057] Among them, similar questions are generated by customization, or similar questions of customers with the same intention are generated by using a large model;

[0058] Among them, regular verification is to use regular expressions to express recognizable questions, combined with entities in the reference entity library to improve the accuracy of semantic recognition;

[0059] The entity library is a user-defined database used to store entity sets involved in the business; when setting the intent recognition method, entities in the entity library are referenced to extract corresponding information;

[0060] For example, the entity names in the entity set involved in the epidemic prevention investigation include departments, places where beef and mutton are purchased, symptoms, etc.;

[0061] In particular, the entity sets in the entity library include user entities generated by the user and system entities preset by the system. System entities are stored in the system preset entity set by default;

[0062] Configure one or more entity attributes of an entity set in its entity library template through the knowledge base manager, where the entity attributes include entity name, entity description, entity value, and regular expression validation;

[0063] Among them, the entity value can be customized, that is, the entity value contained in the entity can be customized;

[0064] Among them, regular validation is to use regular expressions to self-reference or reference other entity sets in the entity library to express the entity values ​​it contains;

[0065] The synonym database contains one or more pairs of standard words and synonyms with corresponding relationships. One standard word corresponds to multiple synonyms. The robot uses the synonym database to deal with the situation where the same intention has multiple expressions. The synonym database is used to convert synonyms into standard words, which is convenient for users to read and is also conducive to the weight ranking of customer questions in the later optimization.

[0066] One or more synonym attributes of the synonym set are configured in the synonym database template through the knowledge base manager, and the synonym attributes include standard words, synonyms, and synonym conversion;

[0067] Among them, the standard word corresponds to one or more synonyms;

[0068] Among them, the synonyms included in the synonym set are generated by user definition;

[0069] When the synonym conversion attribute of the synonym set is turned on, the standard word is used as the name of the synonym set;

[0070] Converting synonyms to standard words can make it easier for users to read and to sort by weight later.

[0071] The hot word library includes pre-set hot words and hot words passed in through parameters; hot words are specific proper nouns in specific businesses; the hot word library is iterated in real time during the human-computer dialogue process, and the variable values ​​and parameter values ​​extracted in the dialogue are fed back to the recognition engine. The recognition engine learns and updates the recognition model based on the fed-back variable values ​​and parameter values ​​to deal with proper nouns;

[0072] Use the knowledge base manager to configure one or more hot word attributes of the hot word set in its hot word base template. Hot word attributes include hot word category, parameter value, variable value, and hot word text:

[0073] Among them, the hot word category is divided into global hot words or node hot words according to the recognition engine scope of the service; global hot words serve all recognition engines in the human-computer dialogue process, and global hot words are set through the hot word library template page; node hot words serve the recognition engines of one or more specific nodes in the human-computer dialogue process, and node hot words are set through the user input configuration page of the node in the flow chart;

[0074] Among them, the hot word parameter value is input from the outside, the hot word variable value is extracted from the current human-computer dialogue, and the hot word text is obtained by customization;

[0075] For example, in a medical conversation scenario, the patient's name, gender, hospital name, and patient contact information are input from outside in advance, and these are hot word parameters; and during the conversation, confirmation of whether it is the person himself, the treatment department, the patient's occupation, the patient's city, and physical symptoms are extracted in the current conversation, and these are all hot word variables;

[0076] In the human-computer dialogue process, the robot determines the branch in the flowchart that needs to be executed next based on the customer's intention, and determines the node jump direction of the flowchart based on the intention recognition result, thereby advancing the human-computer dialogue process;

[0077] Among them, the flow chart is composed of one or more nodes and the logical configuration within each node;

[0078] Specifically, in the human-computer dialogue process, the robot receives the dialogue content input by the customer, and the recognition engine performs intent recognition on the received dialogue content based on the knowledge base;

[0079] The recognition engine selects and sets the recognition method based on the intent library, including keyword recognition, similar question recognition, and regular expression verification;

[0080] When the recognition method is keyword recognition, the recognition engine uses the keywords stored in the intent library to recognize the intent, or calls the synonyms or hot words in the synonym library and hot word library to recognize the intent;

[0081] In particular, the hot word parameter values ​​in the hot word library are passed in from the outside, the hot word variable values ​​are extracted from the current conversation, and the hot word text is obtained through customization;

[0082] When the recognition method is similar question recognition, the recognition engine uses similar questions stored in the intent library to perform intent recognition;

[0083] Among them, when the recognition method is regular verification, the recognition engine calls the entity library to perform self-reference regular verification for intent recognition.

[0084] In addition, the present invention discloses a human-machine dialogue semantic understanding system, including a robot and a knowledge base;

[0085] Among them, the robot conducts a human-machine dialogue with the customer and receives the dialogue content input by the customer;

[0086] The robot has a recognition engine, which uses a recognition model to recognize the conversation content it receives;

[0087] Among them, the recognition engine recognizes the intention of the conversation content input by the customer based on the knowledge base, judges and obtains the customer's intention;

[0088] In particular, the recognition engine uses the recognition model to recognize the conversation content it receives, including speech recognition. When the conversation content input by the customer is speech, the recognition engine first performs speech recognition on the voice input by the customer, and then performs intent recognition on the speech recognition result in combination with the knowledge base to determine and obtain the customer's intention;

[0089] The recognition model is obtained by pre-training the recognition engine. The recognition model is trained based on different scenarios, and different recognition models are selected for training according to different business requirements of different nodes in the human-computer dialogue.

[0090] For example, identification models include banking models and address models;

[0091] The knowledge base stores data related to the knowledge background of human-computer dialogue, and the data sources of the knowledge base include system presets or input parameters in the human-computer dialogue process;

[0092] The robot semantically understands the conversation content input by the customer, and associates the conversation content with the data related to the customer's intention in the knowledge base to achieve intention recognition;

[0093] The knowledge base includes one or more sub-bases, wherein the sub-base associated with intent recognition includes an intent base, an entity base, a synonym base, and a hot word base;

[0094] The human-computer dialogue semantic understanding system also includes a knowledge base manager, which is used to manage data in the knowledge base; wherein the knowledge base manager includes an intent base template, an entity base template, a synonym base template, and a hot word base template, which are used to configure data attributes in the intent base, the entity base, the synonym base, and the hot word base respectively;

[0095] The intent library contains the customer intent that the robot can judge. The customer intent is the key meaning of a conversation between the customer and the robot. The customer intent corresponds to one or more recognition methods, including keyword recognition, similar question recognition, and regular expression verification.

[0096] In particular, the intent data in the intent library includes custom intents and system preset intents. The system preset intents are system preset intent data stored by default.

[0097] Use the knowledge base manager to configure one or more intent data attributes in its intent base template. The intent data attributes include intent name, keywords, similar questions, and regular expression verification.

[0098] Among them, keywords are generated by self-definition or with the assistance of large models;

[0099] Among them, similar questions are generated by customization, or similar questions of customers with the same intention are generated by using a large model;

[0100] Among them, regular verification is to use regular expressions to express recognizable questions, combined with entities in the reference entity library to improve the accuracy of semantic recognition;

[0101] The entity library is a user-defined database used to store entity sets involved in the business; when setting the intent recognition method, entities in the entity library are referenced to extract corresponding information;

[0102] For example, the entity names in the entity set involved in the epidemic prevention investigation include departments, places where beef and mutton are purchased, symptoms, etc.;

[0103] In particular, the entity sets in the entity library include user entities generated by the user and system entities preset by the system. System entities are stored in the system preset entity set by default;

[0104] Configure one or more entity attributes of an entity set in its entity library template through the knowledge base manager, where the entity attributes include entity name, entity description, entity value, and regular expression validation;

[0105] Among them, the entity value can be customized, that is, the entity value contained in the entity can be customized;

[0106] Among them, regular validation is to use regular expressions to self-reference or reference other entity sets in the entity library to express the entity values ​​it contains;

[0107] The synonym database contains one or more pairs of standard words and synonyms with corresponding relationships. One standard word corresponds to multiple synonyms. The robot uses the synonym database to deal with the situation where the same intention has multiple expressions. The synonym database is used to convert synonyms into standard words, which is convenient for users to read and is also conducive to the weight ranking of customer questions in the later optimization.

[0108] One or more synonym attributes of the synonym set are configured in the synonym database template through the knowledge base manager, and the synonym attributes include standard words, synonyms, and synonym conversion;

[0109] Among them, the standard word corresponds to one or more synonyms;

[0110] Among them, the synonyms included in the synonym set are generated by user definition;

[0111] When the synonym conversion attribute of the synonym set is turned on, the standard word is used as the name of the synonym set;

[0112] Converting synonyms to standard words can make it easier for users to read and to sort by weight later.

[0113] The hot word library includes pre-set hot words and hot words passed in through parameters; hot words are specific proper nouns in specific businesses; the hot word library is iterated in real time during the human-computer dialogue process, and the variable values ​​and parameter values ​​extracted in the dialogue are fed back to the recognition engine. The recognition engine learns and updates the recognition model based on the fed-back variable values ​​and parameter values ​​to deal with proper nouns;

[0114] Use the knowledge base manager to configure one or more hot word attributes of the hot word set in its hot word base template. The attributes include hot word category, hot word parameter value, hot word variable value, and hot word text:

[0115] Among them, the hot word category is divided into global hot words or node hot words according to the recognition engine scope of the service; global hot words serve all recognition engines in the human-computer dialogue process, and global hot words are set through the hot word library template page; node hot words serve the recognition engines of one or more specific nodes in the human-computer dialogue process, and node hot words are set through the user input configuration page of the node in the flow chart;

[0116] Among them, the hot word parameter value is input from the outside, the hot word variable value is extracted from the current human-computer dialogue, and the hot word text is obtained by customization;

[0117] In the human-computer dialogue process, the robot determines the branch in the flowchart that needs to be executed next based on the customer's intention, and determines the node jump direction of the flowchart based on the intention recognition result, thereby advancing the human-computer dialogue process;

[0118] Among them, the flow chart is composed of one or more nodes and the logical configuration within each node;

[0119] Specifically, in the human-computer dialogue process, the robot receives the dialogue content input by the customer, and the recognition engine performs intent recognition on the received dialogue content based on the knowledge base;

[0120] The recognition engine selects and sets the recognition method based on the intent library, including keyword recognition, similar question recognition, and regular expression verification;

[0121] When the recognition method is keyword recognition, the recognition engine uses the keywords stored in the intent library to recognize the intent, or calls the synonyms or hot words in the synonym library and hot word library to recognize the intent;

[0122] In particular, the hotword parameter values ​​in the hotword library are passed in from the outside, and the hotword variable values ​​are extracted from the current conversation;

[0123] When the recognition method is similar question recognition, the recognition engine uses similar questions stored in the intent library to perform intent recognition;

[0124] Among them, when the recognition method is regular verification, the recognition engine calls the entity library to perform self-reference regular verification for intent recognition.

[0125] Implementing the embodiments of the present invention will have the following beneficial effects:

[0126] The present invention can realize the setting of a scene-based recognition model in the nodes of the human-computer dialogue process to perform speech recognition, and further combines a real-time self-iterative knowledge base and supports semantic understanding optimization and intent recognition in the scene through entities, synonyms and hot words based on the contextual dialogue scenario, thereby greatly improving the semantic understanding performance.

[0127] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A human-computer dialogue semantic understanding system, characterized in that: Including robots, knowledge base; The robot conducts a human-machine dialogue with the customer and receives the dialogue content input by the customer. The robot has a recognition engine, and the recognition engine uses a recognition model to recognize the dialogue content it receives. The recognition engine recognizes the intention of the dialogue content input by the customer based on the knowledge base, and determines and obtains the customer's intention. The knowledge base stores data related to the knowledge background of human-computer dialogue, and the data source of the knowledge base includes system presets or input parameters in the human-computer dialogue process; the knowledge base includes one or more sub-bases, and the sub-bases associated with intent recognition include intent base, entity base, synonym base, and hot word base; The robot determines the branch in the flowchart that it needs to execute next based on the acquired customer intention, and determines the node jump direction of the flowchart based on the intention recognition result, thereby advancing the human-computer dialogue process; wherein, the flowchart is composed of one or more nodes and the logical configuration within each node.

2. The human-computer dialogue semantic understanding system according to claim 1, characterized in that: in, The recognition engine uses the recognition model to recognize the conversation content it receives, including voice recognition. When the conversation content input by the customer is voice, the recognition engine first performs voice recognition on the voice input by the customer, and then combines the knowledge base to perform intent recognition on the voice recognition result to determine and obtain the customer's intention; Among them, the recognition model is obtained by pre-training of the recognition engine; the recognition model is trained based on different scenarios, and different recognition models are selected for training according to different business needs of different nodes in the human-computer dialogue.

3. The human-computer dialogue semantic understanding system according to claim 1, characterized in that: in, The human-computer dialogue semantic understanding system also includes a knowledge base manager, which is used to manage the data in the knowledge base; wherein the knowledge base manager includes an intent library template, an entity library template, a synonym library template, and a hot word library template, which are respectively used to configure the data attributes in the intent library, entity library, synonym library, and hot word library.

4. The human-computer dialogue semantic understanding system according to claim 3, It is characterized in that in, The intent library contains the customer intent that the robot can judge. The customer intent is the key meaning of a conversation between the customer and the robot. The customer intent corresponds to one or more recognition methods, including keyword recognition, similar question recognition, and regular expression verification. The intent data in the intent library includes custom intents and system preset intents. The system preset intents are system preset intent data stored by default. Use the knowledge base manager to configure one or more intent data attributes in its intent base template. The intent data attributes include intent name, keywords, similar questions, and regular expression verification. Among them, keywords are generated through customization or with the assistance of a large model; among them, similar questions are generated through customization or with the assistance of a large model to generate similar questions of customers with the same intent; among them, regular verification is to use regular expressions to express recognizable questions, combined with referencing entities in the entity library to improve the accuracy of semantic recognition.

5. The human-computer dialogue semantic understanding system according to claim 3, It is characterized in that in, The entity library is a user-defined database used to store entity sets involved in the business. When setting the intent recognition method, entities in the entity library are referenced to extract corresponding information. The entity sets in the entity library include user entities generated by the user and system entities preset by the system. System entities are stored in the system preset entity set by default. Configure one or more entity attributes of an entity set in its entity library template through the knowledge base manager, where the entity attributes include entity name, entity description, entity value, and regular expression validation; Among them, the entity value can be customized, that is, the entity value contained in the entity can be customized; among them, regular verification is to use regular expressions to add self-references or references to other entity sets in the entity library to express the entity values ​​it contains.

6. The human-computer dialogue semantic understanding system according to claim 3, It is characterized in that in, The synonym database contains one or more pairs of standard words and synonyms with corresponding associations. One standard word corresponds to multiple synonyms. The robot uses the synonym database to deal with the situation where the same intention has multiple expressions. The synonym database is used to convert synonyms into standard words, which is convenient for users to read and is also conducive to the weight ranking of customer questions in later optimization. One or more synonym attributes of the synonym set are configured in the synonym database template through the knowledge base manager, and the synonym attributes include standard words, synonyms, and synonym conversion; The standard word corresponds to one or more synonyms; the synonyms contained in the synonym set are generated by customization; when the synonym conversion property of the synonym set is turned on, the standard word is used as the name of the synonym set.

7. The human-computer dialogue semantic understanding system according to claim 3, It is characterized in that The hot word library includes pre-set hot words and hot words passed in through parameters; hot words are specific proper nouns in specific businesses; the hot word library is iterated in real time during the human-computer dialogue process, and the variable values ​​and parameter values ​​extracted in the dialogue are fed back to the recognition engine. The recognition engine learns and updates the recognition model based on the fed-back variable values ​​and parameter values ​​to deal with proper nouns; Use the knowledge base manager to configure one or more hot word attributes of the hot word set in its hot word base template. Hot word attributes include hot word category, hot word parameter value, hot word variable value, and hot word text: Among them, the hot word category is divided into global hot words or node hot words according to the recognition engine scope of the service; global hot words serve all recognition engines in the human-computer dialogue process, and global hot words are set through the hot word library template page; node hot words serve the recognition engines of one or more specific nodes in the human-computer dialogue process, and node hot words are set through the user input configuration page of the node in the flow chart; Among them, the hot word parameter value is passed in from the outside, the hot word variable value is extracted from the current human-computer dialogue, and the hot word text is obtained through customization.

8. A method for understanding the semantics of human-computer dialogue, characterized in that: include: The robot conducts a human-machine dialogue with the customer and receives the dialogue content input by the customer; The robot has a recognition engine, which uses a recognition model to recognize the conversation content it receives; the recognition engine recognizes the intention of the conversation content input by the customer based on the knowledge base, and determines and obtains the customer's intention; The knowledge base stores data related to the knowledge background of human-computer dialogue. The data sources of the knowledge base include system presets or input parameters in the human-computer dialogue process. The knowledge base includes one or more sub-bases, among which the sub-bases associated with intent recognition include intent base, entity base, synonym base, and hot word base. The robot determines the branch in the flowchart that it needs to execute next based on the acquired customer intention, and determines the node jump direction of the flowchart based on the intention recognition result, thereby advancing the human-computer dialogue process; wherein, the flowchart is composed of one or more nodes and the logical configuration within each node.

9. The method for understanding semantics of human-computer dialogue according to claim 8, characterized in that: in, The recognition engine uses the recognition model to recognize the conversation content it receives, including voice recognition. When the conversation content input by the customer is voice, the recognition engine first performs voice recognition on the voice input by the customer, and then combines the knowledge base to perform intent recognition on the voice recognition result to determine and obtain the customer's intention; Among them, the recognition model is obtained by pre-training of the recognition engine; the recognition model is trained based on different scenarios, and different recognition models are selected for training according to different business needs of different nodes in the human-computer dialogue.

10. The method for understanding semantics of human-computer dialogue according to claim 8, characterized in that: in, The recognition engine selects and sets the recognition method based on the intent library, including keyword recognition, similar question recognition, and regular expression verification; When the recognition method is keyword recognition, the recognition engine uses the keywords stored in the intent library to recognize the intent, or calls the synonyms or hot words in the synonym library and hot word library to recognize the intent; When the recognition method is similar question recognition, the recognition engine uses similar questions stored in the intent library to perform intent recognition; Among them, when the recognition method is regular verification, the recognition engine calls the entity library to perform self-reference regular verification for intent recognition.

Images