A dialogue interaction method and device

Abstract

The application provides a dialogue interaction method and device, the method comprises: performing intent recognition on a target question input by a user to obtain entity results and multiple intent results, each intent result comprising intent content and a confidence value corresponding to the intent result, and the entity results comprising an intent entity for representing a user intent and entity content corresponding to the intent entity; determining the intent content contained in the intent result with the highest confidence value as an initial recognized intent corresponding to the target question; if the initial recognized intent belongs to a coherent intent, determining a target intent of the target question based on a target historical conversation and the initial recognized intent; and feeding back response information for the target intent to the user. The application can more accurately recognize the real intent of the user and provide effective information, thereby improving work efficiency, by associating an unknown intent with a theme or business in a historical conversation.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a dialogue interaction method and apparatus. Background Technology

[0002] With the popularization of artificial intelligence technology, more and more chatbots are being put into use. Current AI chatbots can meet users' needs to a certain extent, but as chatbots become more widespread, users are placing higher demands on the human-like nature of the chatbot dialogue. Traditional AI chatbot design uses machine learning algorithms to semantically analyze the questions in the dialogue and provide answers based on the context of certain items.

[0003] However, while traditional solutions have no limitations on application scenarios, the algorithms used in machine learning determine the intelligence of intelligent voice dialogue. Due to the limitations of the algorithms, the dialogue often provides words that have no relation to the dialogue context, resulting in low accuracy in recognizing user intent and causing irrelevant answers in the dialogue. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide at least one dialogue interaction method and apparatus, which obtains the target intent corresponding to the target question by associating the unclear intent of the topic or business with the topic or business in the historical dialogue, thereby more accurately identifying the user's true intent and improving the accuracy of identification.

[0005] This application mainly includes the following aspects:

[0006] In a first aspect, embodiments of this application provide a dialogue interaction method, the dialogue interaction method comprising:

[0007] The system performs intent recognition on the user-input target question, obtaining entity results and multiple intent results. Each intent result includes intent content and a corresponding confidence value. Entity results include intent entities that represent the user's intent and the corresponding entity content. The intent content contained in the intent result with the highest confidence value is determined as the initial identified intent corresponding to the target question. If the initial identified intent is a connecting intent, the target intent of the target question is determined based on the target's historical conversation and the initial identified intent. Response information for the target intent is then fed back to the user.

[0008] In one possible implementation, if the initial identification intent belongs to the business intent, then the initial identification intent is determined as the target intent of the target problem.

[0009] In one possible implementation, the intent content includes an intent name. After determining the intent content contained in the intent result with the highest confidence value as the initial identification intent corresponding to the target question, the method further includes: determining whether the intent name of the initial identification intent includes a business topic and a business type; if the intent name of the initial identification intent includes a business topic and a business type, then the intent type of the initial identification intent is determined to be a business intent; if the intent name of the initial identification intent does not include a business topic, then the intent type of the initial identification intent is determined to be a connection intent.

[0010] In one possible implementation, if the initial identification intent is a connection intent, the step of determining the target intent of the target question based on the target historical session and the initial identification intent includes: determining multiple related business intents corresponding to the initial identification intent; determining multiple candidate related business intents from the multiple related business intents according to a pre-built knowledge graph, and the relationship network corresponding to each candidate related business intent, wherein each relationship network is the entity result corresponding to the candidate related business intent and the dialogue branch pointed to by the entity result; identifying the entity contained in the initial identification intent; searching for target candidate related business intents containing entities from the multiple candidate related business intents; and determining the target candidate related business intent in the most recent historical session as the target intent corresponding to the target question according to the session order of the target historical session.

[0011] In one possible implementation, the step of determining multiple associated business intents corresponding to the initial identification intent includes: determining the connection intent type of the initial identification intent, wherein the connection intent type includes a first-level connection intent and a second-level connection intent, wherein the first-level connection intent does not include a business topic and a business type, and the second-level connection intent does not include a business topic; if the connection intent type of the initial identification intent is a first-level connection intent, then the business topics and business types of multiple historical business intents are associated with the initial identification intent to obtain multiple associated business intents; if the connection intent type of the initial identification intent is a second-level connection intent, then the business topics of multiple historical business intents are associated with the initial identification intent to obtain multiple associated business intents.

[0012] In one possible implementation, the step of determining multiple associated business intents corresponding to the initial identification intent includes: determining the association method of the initial identification intent, whereby the association method includes business type tag association and business topic tag association; if the association method of the initial identification intent is business type tag association, then based on the knowledge graph, determining multiple associated business intents corresponding to the business type tag association method; if the association method of the initial identification intent is business topic tag association, then based on the knowledge graph, determining multiple associated business intents corresponding to the business topic tag association method.

[0013] In one possible implementation, the step of determining multiple candidate related business intentions and the relationship network corresponding to each candidate related business intention from multiple related business intentions based on a pre-constructed knowledge graph includes: comparing multiple related business intentions with preset business intentions in the knowledge graph; taking the preset business intentions in the knowledge graph that correspond to the multiple related business intentions as candidate related business intentions; and sequentially obtaining the relationship network corresponding to the candidate related business intentions from the nearest to the furthest in the target historical session according to the session order in the target historical session.

[0014] In one possible implementation, if the initial identification intent belongs to a business intent, the step of determining the initial identification intent as the target intent of the target problem includes: comparing the highest confidence value with a first threshold; if the highest confidence value is greater than or equal to the first threshold, determining the initial identification intent as the target intent of the target problem; if the highest confidence value is less than the first threshold, determining the second highest confidence value among the confidence values ​​corresponding to multiple intent results; determining whether the difference between the highest confidence value and the second highest confidence value is greater than or equal to a second threshold; if the difference is greater than or equal to the second threshold, determining the initial identification intent as the target intent of the target problem.

[0015] In one possible implementation, the step of providing the user with response information for the target intent includes: determining whether the target relationship network corresponding to the requested business intent is complete based on a pre-built knowledge graph; if incomplete, determining the missing entity information based on the target relationship network, generating a guiding statement based on the missing entity information, and providing the guiding statement to the user to guide the user to provide the missing entity information.

[0016] Secondly, embodiments of this application also provide a dialogue interaction device, comprising: an intent recognition module, configured to recognize the intent of a target question input by a user, and obtain entity results and multiple intent results, each intent result including intent content and a confidence value corresponding to the intent result, and the entity result including an intent entity used to characterize the user's intent and entity content corresponding to the intent entity; a first determination module, configured to determine the intent content contained in the intent result with the highest confidence value as the initial identified intent corresponding to the target question; a second determination module, configured to determine the target intent of the target question based on the target historical conversation and the initial identified intent if the initial identified intent is a connecting intent; and a response feedback module, configured to provide the user with response information for the target intent.

[0017] This application provides a dialogue interaction method and apparatus that performs intent recognition on a user-input target question to obtain entity results and multiple intent results. Each intent result includes intent content and a corresponding confidence value. The entity result includes an intent entity representing the user's intent and the entity content corresponding to the intent entity. The intent content contained in the intent result with the highest confidence value is determined as the initial identified intent corresponding to the target question. If the initial identified intent is a connecting intent, the target intent of the target question is determined based on the target's historical conversations and the initial identified intent. Response information for the target intent is then fed back to the user. By associating intents with unclear topics or business functions with topics or business functions in historical conversations, the user's true intent can be more accurately identified, providing effective information and thus improving work efficiency.

[0018] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of a dialogue interaction method provided in an embodiment of this application is shown;

[0021] Figure 2 This application illustrates a flowchart of steps for determining the target intent of a target problem, as provided in an embodiment of this application. Figure 1 ;

[0022] Figure 3 This application illustrates a flowchart of steps for determining the target intent of a target problem, as provided in an embodiment of this application. Figure 2 ;

[0023] Figure 4 A schematic diagram of the structure of a dialogue interaction device provided in an embodiment of this application is shown. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0025] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0026] Please see Figure 1 , Figure 1 This is a flowchart illustrating a dialogue interaction method provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the dialogue interaction method includes the following steps:

[0027] S100: Perform intent recognition on the target question input by the user to obtain entity results and multiple intent results.

[0028] In practice, chatbots can engage in conversations with users in different business scenarios. Users can input their target questions into the chatbot via voice or text. For example, a user might ask the chatbot, "How do I apply for a residence permit?"

[0029] Chatbots can identify the target question input by the user. Each intent result includes the intent content and the corresponding credibility value. Entity results include the intent entity used to represent the user's intent and the entity content corresponding to the intent entity. Specifically, all entity information contained in the target question can be extracted through techniques such as keyword matching, regular expressions, and information extraction.

[0030] In a preferred embodiment, an intent recognition model needs to be created first based on the application scenario of the chatbot. This model is then used to identify the target question input by the user. The intent recognition model can be created in different ways. Specifically, when identifying the intent type by intent name, the intent recognition model is created in the following way:

[0031] First, multiple preset intents are created based on the application scenarios of the chatbot. Each preset intent includes a preset intent name, and each preset intent name must meet preset naming rules. Specifically, according to different preset naming rules, preset intents are divided into business intents and connecting intents. Therefore, preset intents also include preset business intents and preset connecting intents. In the preset naming rules, preset intents whose names include business theme, business type, and business purpose are identified as preset business intents. Preset intents whose names do not include business theme and / or business type are identified as preset connecting intents. Among them, preset intents whose names do not include business theme or business type are identified as first-level connecting intents, and preset intents whose names do not include business theme are identified as second-level connecting intents. For example, the pre-set "Residence Permit_Application_Application Process", "Residence Permit_Agency_Application Materials", and "Housing Rental Registration_Application_Application Materials" are all business intents, while "Replacement_Application Materials" and "Application Materials" are connecting intents. Among them, "Replacement_Application Materials" is a second-level connecting intent, and "Application Materials" is a first-level connecting intent.

[0032] You can also distinguish between business intents and connection intents by adding a fixed prefix to the name of the preset intent in the preset connection intent. For example, "General_Replace_Processing Materials" and "General_Processing Materials" are connection intents.

[0033] Secondly, based on multiple preset intentions, we collect the question data of users when consulting related preset intentions. That is, we collect the dialogue scripts that users may use when they talk to the chatbot in the corresponding business application scenarios. These scripts are all related to multiple preset intentions, such as "residence permit_agency_application materials": "What materials do I need to apply for a residence permit for my child?" "What do I need to prepare to apply for a residence permit for my elderly family member?" etc. The richness and diversity of the data is the key to the success of the model.

[0034] Finally, collecting entity information refers to the information that users need to provide in real-world application scenarios to clarify their intentions and purposes, such as "representative: elderly", "representative: child", "place of household registration: local", "place of household registration: non-local".

[0035] After creating the preset intent, collecting the question corpus and entity information, the data is used for recognition training to generate an intent recognition model.

[0036] In another preferred embodiment, when identifying the intent type via the intent type identifier, the intent identification model is created in the following manner:

[0037] First, create multiple preset intents based on the application scenarios of the chatbot. Each preset intent includes a preset intent name. Specifically, different names can be used according to different intent requirements and the developers' usage habits, such as "materials for residence permit application". Since there is no preset naming rule at this time, an intent type identifier should be added to each preset intent according to its type to mark whether the corresponding preset intent is a business intent or a connecting intent. The intent type identifier includes a business intent identifier and a connecting intent identifier. In addition, two identifiers need to be added to mark the topic tag and business tag corresponding to each preset intent. For example, the topic tag of the preset intent "materials for residence permit application" is "residence permit", and the business tag is "residence permit application". The topic tag of the preset intent "materials for housing rental registration agency" is "housing rental registration", and the business tag is "housing rental registration agency".

[0038] After creating multiple preset intent names, question corpora and entity information are collected for multiple preset intents. Then, the collected information is used to train the intent recognition model to generate the intent recognition model. The specific process of collecting question corpora and entity information and generating the intent recognition model after training can be referred to the corresponding process in the above method embodiments, and will not be repeated here.

[0039] In a preferred embodiment, after the intent recognition model is created, a knowledge graph is also created. When the intent type is identified by the intent name, the knowledge graph is created in the following way:

[0040] Specifically, the entity information collected in the intent recognition model is associated with the corresponding preset business intents in the model. Different entity information points to different preset dialogue branches, generating a relationship network corresponding to each preset business intent. Furthermore, the relationship network corresponding to each preset business intent ultimately points to different response results depending on the entities it contains. Multiple relationship networks corresponding to multiple preset business intents constitute a knowledge graph. Here, the relationship network corresponding to each preset business intent may contain one entity, multiple entities, or no relationship network at all. For example, in the phrase "Residence Permit_Agency_Processing Materials," the relevant entity includes "Agency Target," and different processing material descriptions are provided based on different agency targets.

[0041] The purpose of configuring a knowledge graph is to enable chatbots to identify the entity information required to respond to a user's target question after recognizing it, and to determine the appropriate questioning techniques to guide the user to provide the corresponding entity information. Only after collecting this entity information can the chatbot clarify the user's true intent and provide a targeted response. For example, in the context of "residence permit_agency_application materials," the relevant entity includes "agency recipient," and different application materials can be provided based on the different agency recipients.

[0042] In another preferred embodiment, when identifying intent types through intent type identifiers, after creating a knowledge graph in the manner described above, it is also necessary to associate the preset connection intents among multiple preset intents with multiple preset business intents and configure their association methods.

[0043] The specific association methods are divided into topic tag association and business tag association. For example, if the preset connection intention "Required Materials" lacks business information, it needs to be associated through business tags to obtain preset business intentions such as "Materials for Residence Permit Application", "Materials for Residence Permit Agency Application", "Materials for Housing Rental Registration", etc.; if the preset connection intention "Agency Materials" lacks relevant topics, it needs to be associated through topic tags to obtain preset business intentions such as "Materials for Residence Permit Agency Application" "Materials for Housing Rental Registration Agency Application", etc.

[0044] S200. The intent content contained in the intent result with the highest confidence value is determined as the initial identification intent corresponding to the target question.

[0045] In a preferred embodiment, a user poses a target question to the chatbot, such as "I'm renting a house and want to apply for a residence permit, what should I do?". Upon receiving the target question, the chatbot invokes the aforementioned intent recognition model, which utilizes NLU (Natural Language Undefined)... The Natural Language Processing (NLP) algorithm and the logistic regression model calculate multiple confidence values ​​for each preset intent in the intent recognition model and the target question input by the current user. The highest confidence value is less than 1, and the lowest is greater than 0. For example, the intent recognition model outputs intent results such as: "'intent': 'Residential Permit Application Process', 'score': '0.995'", "'intent': 'Materials for Residential Permit Application', 'score': '0.002'", and "'intent': 'Housing Rental Registration Process', 'score': '0.001'". Here, "intent" represents the intent content used to indicate the theme and / or intent type of the diagram, and "score" represents the confidence value corresponding to each intent result, which indicates the degree of matching between the target question and multiple preset intents in the intent recognition model. In this embodiment, 'score': '0.995' is the highest confidence value output by the intent recognition model, so the intent content 'intent': 'Residential Permit Application Process' corresponding to 'score': '0.995' is determined as the initial recognition intent.

[0046] S300. If the initial identification intent is a connection intent, determine the target intent of the target question based on the target's historical conversation and the initial identification intent.

[0047] In a preferred embodiment, it is also necessary to determine the type of the initial identification intent. When identifying the intent type by intent name, the step of determining the type of the initial identification intent includes:

[0048] Determine whether the intent name of the initial identification intent includes a business theme and a business type. If the intent name of the initial identification intent includes a business theme and a business type, then the intent type of the initial identification intent is determined to be a business intent; if the intent name of the initial identification intent does not include a business theme, then the intent type of the initial identification intent is determined to be a connection intent.

[0049] In one specific embodiment, if the initial identification intent is "Residence Permit_Agency_Processing Materials", then the intent name of the initial identification intent includes the business subject "Residence Permit", the business type "Agency", and the business purpose "Processing Materials". Therefore, the intent type of "Residence Permit_Agency_Processing Materials" belongs to business intent. If the initial identification intent is "General_Replacement_Processing Materials", then based on its intent name, it can be determined that the initial identification intent lacks a business subject. Therefore, the intent type of the initial identification intent belongs to connecting intent.

[0050] In another preferred embodiment, when identifying the intent type via the intent type identifier, the step of determining the intent type of the initially identified intent includes:

[0051] Obtain the intent type identifier of the initial identification intent. If the intent type identifier of the initial identification intent is a business intent identifier, then the intent type of the initial identification intent is determined to be a business intent. If the intent type identifier of the initial identification intent is a connection intent identifier, then the intent type of the initial identification intent is determined to be a connection intent.

[0052] In one specific embodiment, if the initial identification intent is "materials for residence permit application" and the initial identification intent carries a business intent identifier, then the intent type of the initial identification intent is determined to be a business intent; if the initial identification intent is "required materials" and the initial identification intent carries a connecting intent identifier, then the intent type of the initial identification intent is determined to be a connecting intent.

[0053] In one specific embodiment, please refer to Figure 2 , Figure 2 This application illustrates a flowchart of steps for determining the target intent of a target problem, as provided in an embodiment of this application. Figure 1 .like Figure 2 As shown, the steps to determine the target intent of the target question based on the target's historical sessions and initial identification intent include:

[0054] S301. Determine the multiple associated business intents corresponding to the initial identification intent.

[0055] In one specific embodiment, when the initial identification intent is a connecting intent, it is necessary to associate the initial identification intent with other historical business intents in the historical session to clarify the business theme and / or business type of the initial identification intent, thereby obtaining the true business intent corresponding to the initial identification intent. The target historical session is a historical session within a preset historical session layer range. In a specific session scenario, there may be multiple historical business intents corresponding to multiple business themes and / or business types. Therefore, the chatbot needs to perform contextual intent association within a certain number of dialogue layers. The specific historical session layer limit should be flexibly configurable in different business scenarios. For example, in the government affairs field, it is generally configured as 5 layers. Therefore, it is necessary to pre-set a preset historical session layer, determine the sessions within the preset historical session layer as the target historical session, and obtain multiple historical business intents in the target historical session.

[0056] In a preferred embodiment, when the intent type is identified by intent name, multiple associated service intents are obtained in the following manner:

[0057] The process involves determining the connecting intent type of the initial identification intent. If the connecting intent type is a Level 1 connecting intent, then the business themes and business types of multiple historical business intents are associated with the initial identification intent to obtain multiple associated business intents. If the connecting intent type is a Level 2 connecting intent, then the business themes of multiple historical business intents are associated with the initial identification intent to obtain multiple associated business intents. Connecting intent types include both Level 1 and Level 2 connecting intents. Level 1 connecting intents do not include business themes and business types, and Level 2 connecting intents do not include business themes. Specifically, if the historical business intents, from oldest to newest, are "Residence Permit_Application_Application Process" and "Housing Rental Registration_Application_Application Process," and the current initial identification intent is "General_Application Materials," then the combined result will be "Residence Permit_Application_Application Materials" and "Housing Rental Registration_Application_Application Materials."

[0058] In another preferred embodiment, when identifying the intent type via the intent type identifier, multiple associated service intents are obtained in the following manner:

[0059] The association method for the initial identification intent is determined. Association methods include business type tag association and business topic tag association. If the initial identification intent is associated using business type tag association, then based on the knowledge graph, all pre-defined business intents that can be associated with the business type tag association method are determined, and all pre-defined business intents that can be associated are identified as multiple associated business intents. If the initial identification intent is associated using business topic tag association, then based on the pre-defined knowledge graph, all pre-defined business intents that can be associated with the business topic tag association method are determined, and all pre-defined business intents that can be associated are identified as multiple associated business intents. Specifically, the current initial... The identified intent is "required materials". According to the previously generated knowledge graph, the connection intent "required materials" lacks business type information. It is necessary to determine multiple associated preset business intents corresponding to the business type tag in the knowledge graph through business tags. Associated business intents include "materials for residence permit application", "materials for residence permit agency", "materials for housing rental registration", etc. The current initial identified intent is "materials for agency", which lacks related topics. It is necessary to determine multiple associated preset business intents corresponding to the business topic tag in the preset knowledge graph through business topic tags. Associated intents include "materials for residence permit agency" and "materials for housing rental registration agency".

[0060] S302. Based on the pre-built knowledge graph, determine multiple candidate related business intentions from multiple related business intentions, and the relationship network corresponding to each candidate related business intention.

[0061] Specifically, each relationship network consists of the entity result corresponding to the candidate associated business intent and the different dialogue branches pointed to by the entity result.

[0062] In a preferred embodiment, when identifying the intent type by intent name, multiple candidate associated service intents and the relationship network corresponding to each candidate associated service intent are determined in the following way:

[0063] Multiple related business intents are compared with preset business intents in the knowledge graph. The preset business intents in the knowledge graph that correspond to multiple related business intents are used as candidate related business intents. The relationship network corresponding to the candidate related business intents is obtained in order from the nearest to the farthest in the target historical session.

[0064] Specifically, the system determines whether a pre-defined business intent exists in the knowledge graph for multiple related business intents. If a pre-defined business intent exists in the knowledge graph corresponding to any one of the multiple related business intents, that pre-defined business intent is taken as a candidate related business intent. The system then retrieves the relationship network corresponding to the candidate related business intents sequentially from the nearest to the furthest in the target historical session. For example, if the combined related business intents are "Residence Permit_Application_Application Materials" and "Housing Rental Registration_Application_Application Materials", and the knowledge graph contains pre-defined business intents for "Residence Permit_Application_Application Materials" and "Housing Rental Registration_Application_Application Materials", then "Residence Permit_Application_Application Materials" and "Housing Rental Registration_Application_Application Materials" are identified as candidate related business intents. However, the system retrieves the relationship network corresponding to "Residence Permit_Application_Application Materials" and "Housing Rental Registration_Application_Application Materials" sequentially from the nearest to the furthest in the target historical session.

[0065] In another preferred embodiment, when identifying the intent type through the intent type identifier, multiple candidate related business intents and their corresponding multiple relationship networks are determined in the following manner: matching the business type tags and / or business topic tags of multiple historical business intents with the business type tags and / or business topic tags of multiple related business intents in a preset knowledge graph; determining whether there exists a related business intent in the knowledge graph that has the same business type tag and / or business topic tag as any one of the multiple historical business intents; if there is a related business intent that has the same business topic tag and / or business type tag as any historical business intent, then the related business intent is taken as a candidate related business intent, and the relationship network corresponding to the candidate related business intent is obtained sequentially from near to far according to the session order in the target historical session.

[0066] For example, if the initial identification intent is "required materials", according to the preset knowledge graph, the initial identification intent "required materials" lacks business information and needs to be associated through business type tags. According to the corresponding association method in the preset knowledge graph, the initial identification intent "required materials" corresponds to multiple associated business intents in the preset knowledge graph as "materials for residence permit application", "materials for residence permit agency", and "materials for housing rental registration". The historical business intents are "process of residence permit application" and "process of housing rental registration". Based on the topic tags of the historical business intents, it is found that among the multiple associated business intents in the knowledge graph, there are associated business intents "materials for residence permit application" and "materials for housing rental registration" that correspond to the business topic tags of "process of residence permit application" and "process of housing rental registration". Therefore, "materials for residence permit application" and "materials for housing rental registration" are identified as candidate associated business intents. The relationship network corresponding to "materials for residence permit application" and "materials for housing rental registration" is retrieved from the knowledge graph in the order of the conversation in the target historical session from near to far.

[0067] S303. Identify entities contained in the initial identification intent.

[0068] Specifically, when the target question is input into the intent recognition model, an entity result corresponding to the target question is obtained. This entity result is the corresponding entity, and the entity of the initial recognition intent is determined based on the output of the intent recognition model.

[0069] S304. Search for the target candidate related business intent that contains the entity from multiple candidate related business intents.

[0070] Specifically, the intent recognition model will identify the entities in the initial recognition intent. If the initial recognition intent is "to handle materials" and includes the entity result "elderly", then it will sequentially search the multiple relationship networks corresponding to the multiple candidate related business intents obtained to find the nearest relationship network containing the entity, and determine the candidate related business intent corresponding to the relationship network containing the entity as the target candidate related business intent.

[0071] S305. Based on the session order of the target historical sessions, determine the target candidate related business intent in the most recent historical session as the target intent of the target question.

[0072] Based on the order of historical sessions, the target candidate related business intent in the most recent historical session is determined as the target intent corresponding to the target question. This target intent is the true business intent corresponding to the user's current target question. For example, based on the initial identified intent "materials for agency services," the candidate related business intents determined from the knowledge graph are "materials for residence permit agency services" and "materials for housing rental registration agency services." In the knowledge graph, it is found that the relationship network of "materials for residence permit agency services" contains the corresponding entity result, while the relationship network of "materials for housing rental registration agency services" in the knowledge graph does not include the above-mentioned corresponding entity result. Therefore, the target candidate related business intent is "materials for residence permit agency services," and this target candidate related business intent "materials for residence permit agency services" is the true business intent corresponding to the current target question.

[0073] In another preferred embodiment of this application, if an entity result not output by the intent recognition model is identified, the candidate associated business intent in the most recent historical session is determined as the target intent of the target question.

[0074] Specifically, if the intent recognition model does not output entity results, it is assumed that the initial recognition intent does not contain entity information. The candidate related business intents determined within the range of historical session layers are traversed, and the candidate related business intents in the most recent historical session are determined as the target intent corresponding to the target question.

[0075] In another preferred embodiment, if the initial identification intent belongs to the business intent, then the initial identification intent is determined as the target intent of the target problem.

[0076] In one specific embodiment, when the initial identification intent is a business intent, since the business intent clearly includes a business topic and a business type, it is only necessary to directly determine the response script corresponding to the initial identification intent based on the knowledge graph. If the initial identification intent corresponding to the highest confidence value output by the intent identification model is incorrect, that is, the business intent is actually incorrectly identified, then, based on the corresponding relationship network in the knowledge graph, some response scripts that are completely unrelated to the historical conversation may be fed back. However, when the initial identification intent is a connecting intent, even if the connecting intent corresponding to the highest score output by the intent identification model is incorrectly identified, the connecting intent feedback is still a response script related to the historical conversation. Therefore, if the initial identification intent belongs to a business intent, it is still necessary to determine whether the initial identification intent is credible; if the initial identification intent belongs to a connecting intent, it is not necessary to determine whether it is credible.

[0077] In one specific embodiment, please refer to Figure 3 , Figure 3 This application illustrates a flowchart of steps for determining the target intent of a target problem, as provided in an embodiment of this application. Figure 2 ,like Figure 3As shown, if the initial identification intent belongs to the business intent, the steps to determine the initial identification intent as the target intent of the target question include:

[0078] S401. Compare the highest confidence value with the first threshold.

[0079] In an optional embodiment, if the initial identification intent belongs to a business intent, it is necessary to judge the credibility of the initial identification intent. Specifically, the highest credibility value corresponding to the initial identification intent is determined and compared with a pre-set first threshold. The first threshold is used to judge whether the identification result of the intent recognition model is credible.

[0080] S402. If the highest confidence value is greater than or equal to the first threshold, the initial identification intent is determined as the target intent of the target problem.

[0081] Specifically, the higher the confidence value, the closer the preset intent corresponding to the confidence value is to the target problem. When the highest confidence value is not less than the first threshold, the initial recognition intent identified by the intent recognition model corresponding to the highest confidence value is considered to be a credible initial recognition intent, and the initial recognition intent can be determined as the target intent corresponding to the target problem.

[0082] S403. If the highest confidence value is less than the first threshold, determine the second highest confidence value among the confidence values ​​corresponding to multiple intent results.

[0083] Specifically, when the highest confidence value is less than the first threshold, a second authentication is required for the highest confidence value. In other words, the second highest confidence value among multiple intent results needs to be obtained again.

[0084] S404. Determine whether the difference between the highest confidence value and the second highest confidence value is greater than or equal to the second threshold.

[0085] If the highest confidence value is less than the first threshold, it is also necessary to calculate the difference between the highest confidence value and the second highest confidence value, and determine whether the difference between the highest confidence value and the second highest confidence value is greater than the second threshold.

[0086] S405. If the difference is greater than or equal to the second threshold, the initial identification intent is determined as the target intent of the target problem.

[0087] If the highest confidence value is less than the first threshold, and the difference between the highest confidence value and the second highest confidence value is greater than or equal to the second threshold, then the initial identification intent corresponding to the highest confidence value is considered a credible initial identification intent, and this initial identification intent is the target intent corresponding to the target question.

[0088] S406. If the difference is less than the second threshold, then the initial identification intent corresponding to the highest confidence value is determined to be untrustworthy.

[0089] In an optional implementation, the target question can also be sent to the target auxiliary identification module, which will identify the question and return the identification result. The result may be valid, or it may not be valid and will directly reply with a user-friendly fallback message.

[0090] By introducing a credibility value for the intent results output by the intent recognition model, developers can better maintain and adjust the intent recognition model, and continuously improve the accuracy of the recognition results output by the intent recognition model.

[0091] In one specific embodiment, if the initial identification intent belongs to a business intent and its result is credible after credibility judgment, then directly determining the initial identification intent as the request business intent corresponding to the target question further includes: obtaining the relationship network in the knowledge graph corresponding to the initial identification intent based on the knowledge graph.

[0092] return Figure 1 S400: Provide the user with response information based on the target intent.

[0093] In one specific embodiment, the step of providing feedback to the user on the response information in response to the target intent includes:

[0094] Based on the pre-constructed knowledge graph, it is determined whether the target relationship network corresponding to the target intent is complete. In a specific embodiment, it is determined whether the target relationship network corresponding to the target intent is complete, that is, whether the entity information covered by the target relationship network corresponding to the target intent is complete.

[0095] If complete, the system will provide the user with a response to the target intent. In a specific embodiment, if the target intent "materials for residence permit application" also includes the entity information "elderly person", then the target intent is actually "to obtain materials for the elderly person's residence permit application". In other words, the target relationship network of the target intent is complete and contains the corresponding entity information. The system can obtain the response information for the target intent pointed to by the entity information "elderly person" through the target relationship network, obtain the response information and provide it to the user.

[0096] If incomplete, the missing entity information is determined based on the target relationship network, and a guiding statement is generated based on the missing entity information. This guiding statement is then fed back to the user to guide them in providing the missing entity information.

[0097] For example, if a target intent is "residence permit application materials", there is no corresponding entity for this target intent. Based on the missing entity information, a guiding statement is generated and fed back to the user to guide the user to provide the missing entity information. For example, the guiding statement could be "Which age group residence permit do you need to apply for?"

[0098] Based on the same application concept, this application also provides a dialogue interaction device corresponding to the dialogue interaction method provided in the above embodiments. Since the principle of the device in this application to solve the problem is similar to the dialogue interaction method in the above embodiments of this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0099] Reference Figure 4 , Figure 4 This is a schematic diagram of the structure of a dialogue interaction device provided in an embodiment of this application.

[0100] like Figure 4 As shown, the dialogue interaction device includes:

[0101] The intent recognition module 510 is used to recognize the intent of the target question input by the user, and obtain entity results and multiple intent results. Each intent result includes intent content and a confidence value corresponding to the intent result. The entity results include intent entities used to represent the user's intent and entity content corresponding to the intent entities.

[0102] The first determining module 520 is used to determine the intent content contained in the intent result with the highest confidence value as the initial identification intent corresponding to the target question.

[0103] The second determining module 530 is used to determine the target intent of the target question based on the target historical session and the initial identifying intent if the initial identification intent belongs to the connecting intent.

[0104] The response feedback module 540 is used to provide the user with response information based on the target intent.

[0105] Optionally, the dialogue interaction device further includes a third determining module, used to determine the initial identification intent as the target intent of the target question if the initial identification intent belongs to the business intent.

[0106] Optionally, the dialogue interaction device further includes a type determination module, used to determine whether the intent name of the initial identification intent includes a business topic and a business type; if the intent name of the initial identification intent includes a business topic and a business type, then the intent type of the initial identification intent is determined to be a business intent; if the intent name of the initial identification intent does not include a business topic, then the intent type of the initial identification intent is determined to be a connection intent.

[0107] Optionally, if the initial identification intent is a connecting intent, the second determining module 530 is further configured to: determine multiple related business intents corresponding to the initial identification intent; determine multiple candidate related business intents and the relationship network corresponding to each candidate related business intent from the multiple related business intents according to the pre-constructed knowledge graph, wherein each relationship network is the entity result corresponding to the candidate related business intent and the dialogue branch pointed to by the entity result; identify the entity contained in the initial identification intent; search for the target candidate related business intent containing the entity from the multiple candidate related business intents; and determine the target candidate related business intent in the most recent historical session as the target intent corresponding to the target question according to the session order of the target historical session.

[0108] Optionally, the second determining module 530 is further configured to determine the connection intent type of the initial identification intent. The connection intent type includes a first-level connection intent and a second-level connection intent. The first-level connection intent does not include a business topic and a business type, and the second-level connection intent does not include a business topic. If the connection intent type of the initial identification intent is a first-level connection intent, then the business topics and business types of multiple historical business intents contained in the target historical session are associated with the initial identification intent to obtain multiple associated business intents. If the connection intent type of the initial identification intent is a second-level connection intent, then the business topics of multiple historical business intents contained in the target historical session are associated with the initial identification intent to obtain multiple associated business intents.

[0109] Optionally, the second determining module 530 is further configured to: determine the association method of the initial identification intent, the association method including business type tag association and business topic tag association; if the association method of the initial identification intent is business type tag association, then determine multiple associated business intents corresponding to the business type tag association method according to the knowledge graph; if the association method of the initial identification intent is business topic tag association, then determine multiple associated business intents corresponding to the business topic tag association method according to the knowledge graph.

[0110] Optionally, the second determining module 530 is further configured to: compare multiple related business intents with preset business intents in the knowledge graph; take the preset business intents in the knowledge graph that correspond to the multiple related business intents as candidate related business intents, and obtain the relationship network corresponding to the candidate related business intents in order from near to far according to the session order in the target historical session.

[0111] Optionally, the third determining module is further configured to: compare the highest confidence value with a first threshold; if the highest confidence value is greater than or equal to the first threshold, determine the initial identification intent as the target intent of the target problem; if the highest confidence value is less than the first threshold, determine the second highest confidence value among the confidence values ​​corresponding to multiple intent results; determine whether the difference between the highest confidence value and the second highest confidence value is greater than a second threshold; if the difference is greater than or equal to the second threshold, determine the initial identification intent as the target intent of the target problem.

[0112] Optionally, the second determining module 530 is further configured to: determine whether the target relationship network corresponding to the target intent is complete based on the pre-constructed knowledge graph; if it is incomplete, determine the missing entity information based on the target relationship network, generate a guiding statement based on the missing entity information, and feed the guiding statement back to the user to guide the user to provide the missing entity information.

[0113] Based on the same concept, this application also provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions that can be executed by the processor. When the electronic device is running, the processor and the memory communicate through the bus. The machine-readable instructions are executed by the processor to perform the steps of any of the dialogue interaction methods in the above embodiments.

[0114] Based on the same concept, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the dialogue interaction method provided in the above embodiments.

[0115] Specifically, the storage medium can be a general-purpose storage medium, such as a removable disk or hard disk, and the computer program on the storage medium can execute the above-mentioned dialogue interaction method when it is run.

[0116] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0117] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0118] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0119] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0120] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A conversational interaction method, characterized by, The dialogue interaction method comprises: An intent recognition model is called to recognize the intent of the target question input by the user, to obtain entity results and multiple intent results, each intent result comprising intent content and a credibility value corresponding to the intent result, and the entity results comprising intent entities for representing the intent of the user and entity content corresponding to the intent entities, wherein the intent recognition model identifies business topics and business types through intent names or intent type identifiers, and determines the intent type through the business topics and business types; The intent content contained in the intent result with the highest credibility value is determined as the initial recognition intent corresponding to the target question; If the initial recognition intent belongs to a connection intent, a target intent of the target question is determined based on a target historical conversation, a relationship network indicated in a pre-constructed knowledge graph, entity results corresponding to the initial recognition intent, and the initial recognition intent, the relationship network describing the entity results corresponding to the intent; Response information for the target intent is fed back to the user.

2. The conversational interaction method of claim 1, wherein, The dialogue interaction method further comprises: If the initial recognition intent belongs to a business intent, the initial recognition intent is determined as the target intent of the target question.

3. The conversational interaction method of claim 1, wherein, The intent content comprises an intent name, and after the step of determining the intent content contained in the intent result with the highest credibility value as the initial recognition intent corresponding to the target question, the method further comprises: It is determined whether the business topics and business types are included in the intent name of the initial recognition intent; If the business topics and business types are included in the intent name of the initial recognition intent, it is determined that the intent type of the initial recognition intent belongs to a business intent; If the business topics are not included in the intent name of the initial recognition intent, it is determined that the intent type of the initial recognition intent belongs to a connection intent.

4. The conversational interaction method of claim 1, wherein, If the initial recognition intent belongs to a connection intent, a target intent of the target question is determined based on a target historical conversation and the initial recognition intent, the step comprising: Multiple associated business intents corresponding to the initial recognition intent are determined; According to a pre-constructed knowledge graph, multiple candidate associated business intents and relationship networks corresponding to each candidate associated business intent are determined from the multiple associated business intents, each relationship network being an entity result corresponding to a candidate associated business intent and a dialogue branch pointed to by the entity result; An entity included in the initial recognition intent is recognized; A target candidate associated business intent containing the entity is searched from the multiple candidate associated business intents; According to the conversation order of the target historical conversation, the target candidate associated business intent in the most recent historical conversation is determined as the target intent corresponding to the target question.

5. The conversational interaction method of claim 4, wherein, The step of determining the multiple associated business intents corresponding to the initial recognition intent comprises: A connection intent type of the initial recognition intent is determined, the connection intent type comprising a first-level connection intent and a second-level connection intent, the first-level connection intent not comprising business topics and business types, and the second-level connection intent not comprising business topics; If the connection intention type of the initially identified intention is a first-level connection intention, a business theme and a business type of a plurality of historical business intentions contained in the target historical session are associated to the initially identified intention to obtain a plurality of associated business intentions; If the connection intention type of the initially identified intention is a second-level connection intention, a business theme of a plurality of historical business intentions contained in the target historical session is associated to the initially identified intention to obtain a plurality of associated business intentions.

6. The conversational interaction method of claim 4, wherein, The determining the plurality of associated business intentions corresponding to the initially identified intention comprises: determining an association manner of the initially identified intention, the association manner comprising a business type label association and a business theme label association; if the association manner of the initially identified intention is the business type label association, determining the plurality of associated business intentions corresponding to a manner of association of the business type label according to the knowledge graph; if the association manner of the initially identified intention is the business theme label association, determining the plurality of associated business intentions corresponding to a manner of association of the business theme label according to the knowledge graph.

7. The conversational interaction method of claim 4, wherein, The determining the plurality of candidate associated business intentions and the relationship network corresponding to each candidate associated business intention from the plurality of associated business intentions according to the pre-constructed knowledge graph comprises: comparing the plurality of associated business intentions with preset business intentions in the knowledge graph; taking the preset business intentions in the knowledge graph corresponding to the plurality of associated business intentions as candidate associated business intentions; obtaining the relationship network corresponding to the candidate associated business intention in the order from near to far in the target historical session.

8. The conversational interaction method of claim 2, wherein, If the initially identified intention belongs to a business intention, the determining the initially identified intention as the target intention of the target question comprises: comparing the highest confidence value with a first threshold value; if the highest confidence value is greater than or equal to the first threshold value, determining the initially identified intention as the target intention of the target question; if the highest confidence value is less than the first threshold value, determining a second highest confidence value in the confidence values corresponding to the plurality of intention results; determining whether a difference between the highest confidence value and the second highest confidence value is greater than or equal to a second threshold value; if the difference is greater than or equal to the second threshold value, determining the initially identified intention as the target intention of the target question.

9. The conversational interaction method of claim 1, wherein, The feeding back, to the user, response information for the target intention comprises: determining, according to a pre-constructed knowledge graph, whether a target relationship network corresponding to the target intention is complete; if not, determining missing entity information according to the target relationship network; generating a guide sentence according to the missing entity information, and feeding back the guide sentence to the user to guide the user to provide the missing entity information.

10. A dialog interaction device, characterized by The dialog interaction device comprises: The intent recognition module is configured to invoke a pre-created intent recognition model to perform intent recognition on a target question input by a user, to obtain an entity result and a plurality of intent results, each of the intent results including intent content and a confidence value corresponding to the intent result, and the entity result including an intent entity used to represent a user intent and entity content corresponding to the intent entity, wherein the intent recognition model identifies a business theme and a business type through an intent name or an intent type identifier, and determines an intent type through the business theme and the business type; The first determination module is configured to determine the intent content contained in the intent result with the highest confidence value as an initial recognition intent corresponding to the target question. The second determination module is configured to, if the initial recognition intent belongs to a coherent intent, determine a target intent of the target question based on a target historical conversation, a relationship network indicated in a pre-constructed knowledge graph, an entity result corresponding to the initial recognition intent, and the initial recognition intent, the relationship network describing the entity result corresponding to the intent; The response feedback module is configured to feed back response information for the target intent to the user.

Images