Form filling method, device and equipment based on AI intent recognition and product

Abstract

This application discloses a form filling method, apparatus, device, and product based on AI intent recognition, relating to the field of natural language processing technology. The method includes: receiving user query information; recognizing the user query information using an intent recognition model to obtain several user intents and a list of intent confidence scores corresponding to the several user intents; performing multi-level decision-making on the several user intents based on the intent confidence score list to obtain a final user intent ID; and calling a business interface through an intent service mapping table based on the final intent ID; extracting fields and pre-filling the form through the business interface to obtain an encapsulated result; and sending the encapsulated result to the user terminal corresponding to the user query information. This solves the problems of cumbersome traditional multi-system operations and inaccurate intent recognition in existing intelligent customer service systems, which prevents automatic form pre-filling, thus improving the efficiency of form filling.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and in particular to a form filling method, apparatus, device, and product based on AI intent recognition. Background Technology

[0002] As enterprises continue to advance their digital transformation, internal employees frequently need to interact with multiple business systems in their daily work, such as the enterprise knowledge base, internet information retrieval, BI report queries, and HR leave application processes. Under the traditional service model, employees need to log in to each system separately, manually fill out forms, and retrieve information one by one. The operation process is cumbersome, the overall office efficiency is low, and the user experience is poor. To solve this pain point, an enterprise internal intelligent customer service system based on natural language processing is gradually being put into use. It builds a unified service entry point with conversational interaction, and attempts to simplify the employee operation process and improve the efficiency of internal office and service.

[0003] However, existing intelligent customer service technologies still have significant shortcomings. First, the accuracy of intent recognition is low, making it difficult to cope with the diversity and complexity of user expressions. Second, business system integration is difficult, and there is a lack of a unified scheduling mechanism. In addition, there is insufficient automation of processes, and the forms cannot be pre-filled and processes cannot be initiated automatically, still requiring a lot of manual operation.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main purpose of this application is to provide a form filling method, device, equipment and product based on AI intent recognition, which aims to solve the technical problems of cumbersome operation of traditional multi-systems and the inaccurate intent recognition of existing intelligent customer service, which leads to the inability to automatically prefill forms.

[0006] To achieve the above objectives, this application proposes a form filling method based on AI intent recognition, wherein the form filling method based on AI intent recognition includes: Receive user query information, identify the user query information through an intent recognition model, and obtain a list of several user intents and a list of intent confidence scores corresponding to the several user intents; Based on the intent confidence score list, a multi-level decision is made on the several user intents to obtain the final user intent ID, and the business interface is called through the intent service mapping table based on the final intent ID. The business interface is used to extract fields and pre-fill forms to obtain encapsulated results, which are then sent to the user terminal corresponding to the user's question information.

[0007] In one embodiment, the step of identifying the user's question information using an intent recognition model to obtain several user intents and a list of intent confidence scores corresponding to the several user intents includes: The user's semantics are obtained by performing semantic understanding on the user's question information using the intent recognition model. The intent is classified according to the user semantics to obtain several candidate intents, and the confidence scores of the several candidate intents are calculated through the user semantics. The intent recognition model is used to extract entity auxiliary information and keyword auxiliary information from the user's question information. The entity auxiliary information and keyword auxiliary information are integrated into the candidate intents to obtain a list of user intents and a list of intent confidence scores corresponding to the user intents.

[0008] In one embodiment, the step of performing multi-level decision-making on the plurality of user intents based on the intent confidence score list to obtain the final user intent ID includes: Extract the first user intent with the highest score from the intent confidence score list; The confidence score of the first user intent is compared using the first intent threshold and the second intent threshold. If the confidence score of the first user intent is greater than the first intent threshold, then the first user intent is determined to be the final user intent, and the final user intent ID of the final user intent is extracted. If the confidence score of the first user intent is less than the first intent threshold but greater than the second intent threshold, then multiple rounds of clarification dialogue are conducted based on the candidate intents to obtain the final user intent ID. If the confidence score of the first user intent is less than the second intent threshold, the list of intent confidence scores is sent to the human customer service terminal, which then determines the final user intent ID.

[0009] In one embodiment, the step of performing multiple rounds of clarification dialogue based on the plurality of candidate intents to obtain the final user intent ID includes: Based on the aforementioned candidate intentions, clarifying questions are generated; The clarification question is sent to the user terminal, and the user terminal generates a response based on the clarification question. The system receives the reply information sent by the user terminal, performs context splicing and semantic fusion on the reply information, clarification questions, and user questions to obtain fused text; The intent recognition model is used to identify the intent of the fused text to obtain the final user intent ID.

[0010] In one embodiment, the step of calling the business interface through the intent service mapping table based on the final intent ID includes: The business service type is obtained by matching the type of the intent service mapping table based on the final intent ID. Select the business system adapter based on the aforementioned business service type; After performing protocol conversion and authentication information encapsulation on the business system adapter, the business interface is called through the business system adapter.

[0011] In one embodiment, the step of extracting fields and pre-filling forms through the business interface to obtain a packaged result, and sending the packaged result to the user terminal corresponding to the user's question information, includes: The form pre-filling engine extracts the basic user information corresponding to the user terminal, and extracts the entity information from the user's question information based on the end user intent ID; The form metadata of the business system corresponding to the business interface is invoked through the business interface. Based on the form metadata, the user basic information and entity information are formatted and mapped to fields to obtain the field mapping result; Perform a validity check on the field mapping result, and populate the field mapping result into the initial form after the validity check passes; Based on the initial form, a pre-filled form is generated through the business interface, and a form link is generated through the pre-filled form; The form link and the process information of the pre-filled form are encapsulated to obtain an encapsulation result, and the encapsulation result is sent to the user terminal.

[0012] In one embodiment, after the steps of extracting fields and pre-filling forms through the business interface to obtain the encapsulated result, and sending the encapsulated result to the user terminal corresponding to the user's question information, the method further includes: Receive the archiving process form submitted by the user client; Data is collected from the archive process form using the encapsulation result to obtain the user's operation behavior data; The operational behavior data is cleaned to obtain model training samples; The intent recognition model is incrementally fine-tuned using the training samples of the model to obtain the adjustment result.

[0013] Furthermore, to achieve the above objectives, this application also proposes a form filling device based on AI intent recognition, the form filling device based on AI intent recognition comprising: The receiving module is used to receive user query information, identify the user query information through an intent recognition model, and obtain a list of several user intents and a list of intent confidence scores corresponding to the several user intents. The calling module is used to make multi-level decisions on the several user intents based on the intent confidence score, obtain the final user intent ID, and call the business interface through the intent service mapping table based on the final intent ID. The encapsulation module is used to extract fields and pre-fill forms through the business interface to obtain the encapsulation result, and send the encapsulation result to the user terminal corresponding to the user's question information.

[0014] In addition, to achieve the above objectives, this application also proposes a form filling device based on AI intent recognition, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the form filling method based on AI intent recognition as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the form filling method based on AI intent recognition as described above.

[0016] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the form filling method based on AI intent recognition as described above.

[0017] One or more technical solutions proposed in this application have at least the following technical effects: This application proposes a form filling method, apparatus, device, and product based on AI intent recognition. It receives user query information, identifies the user query information using an intent recognition model to obtain several user intents and a list of intent confidence scores corresponding to these intents. Based on the intent confidence score list, it performs multi-level decision-making on the several user intents to obtain a final user intent ID. Based on the final intent ID, it calls a business interface through an intent service mapping table. Through the business interface, it performs field extraction and form pre-filling to obtain an encapsulated result, which is then sent to the user terminal corresponding to the user query information. Thus, by identifying user queries and outputting multiple intents and confidence lists through an intent recognition model, and then accurately determining the final intent through multi-level decision-making, it solves the problem of inaccurate identification. Subsequently, relying on the intent-service mapping table to uniformly call the business interface, it automatically completes field extraction and form pre-filling, eliminating the need for switching between multiple systems and manual filling. This simplifies operations, achieves automated form pre-filling, and solves the problems of cumbersome traditional multi-system operations and inaccurate intent recognition in existing intelligent customer service systems, which prevents automatic form pre-filling. This improves the efficiency of form filling. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an embodiment of the form filling method based on AI intent recognition provided in this application. Figure 2 This is a schematic diagram illustrating the intent recognition and decision-making involved in the form filling method based on AI intent recognition in this application; Figure 3 This is a schematic diagram of the NL2SQL processing flow of the intelligent question module involved in the form filling method based on AI intent recognition in this application; Figure 4 This is a flowchart illustrating Embodiment 2 of the form filling method based on AI intent recognition in this application; Figure 5 This is a schematic diagram illustrating the form filling method based on AI intent recognition in this application. Figure 6A simplified flowchart illustrating the form filling method based on AI intent recognition provided in Embodiment 2 of this application; Figure 7 This is a schematic diagram of the overall system architecture involved in the form filling method based on AI intent recognition in this application; Figure 8 This is a schematic diagram of the module structure of the form filling device based on AI intent recognition in an embodiment of this application; Figure 9 This is a schematic diagram of the device structure of the hardware operating environment involved in the form filling method based on AI intent recognition in the embodiments of this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] The main solution of this application embodiment is as follows: semantic understanding of the user question information is performed through the intent recognition model to obtain user semantics; intent is classified according to the user semantics to obtain several candidate intents, and confidence scores of the several candidate intents are calculated through the user semantics; entity auxiliary information and keyword auxiliary information are extracted from the user question information through the intent recognition model; the entity auxiliary information and keyword auxiliary information are integrated into the several candidate intents to obtain several user intents and a list of intent confidence scores corresponding to the several user intents. Extract the highest-scoring first user intent from the intent confidence score list; compare the confidence score of the first user intent with a first intent threshold and a second intent threshold; if the confidence score of the first user intent is greater than the first intent threshold, then determine the first user intent as the final user intent, and extract the final user intent ID; if the confidence score of the first user intent is less than the first intent threshold but greater than the second intent threshold, then conduct multiple rounds of clarification dialogue based on the candidate intents to obtain the final user intent ID; if the confidence score of the first user intent is less than the second intent threshold, then send the intent confidence score list to the human customer service terminal, where the human customer service terminal determines the final user intent ID. Generate clarification questions based on the candidate intents; send the clarification questions to the user terminal, where the user terminal generates response information based on the clarification questions; receive the response information sent by the user terminal, and perform contextual concatenation and semantic fusion on the response information, clarification questions, and user question information to obtain fused text; perform intent recognition on the fused text using the intent recognition model to obtain the final user intent ID. The intent service mapping table is type-matched based on the final intent ID to obtain the business service type; a business system adapter is selected based on the business service type; after protocol conversion and authentication information encapsulation of the business system adapter, the business interface is called through the business system adapter. The user's basic information corresponding to the user terminal is extracted using a form pre-filling engine, and entity information from the user's question information is extracted based on the final user intent ID; the form metadata of the business system corresponding to the business interface is called through the business interface; the user's basic information and entity information are format-converted and field-mapped based on the form metadata to obtain a field mapping result; the field mapping result is validated for legality, and after passing the validity validation, the field mapping result is filled into the initial form; a pre-filled form is generated through the business interface based on the initial form, and a form link is generated through the pre-filled form; the process information of the form link and the pre-filled form is encapsulated to obtain an encapsulation result, and the encapsulation result is sent to the user terminal.The system receives an archived process form submitted by the user; collects data from the archived process form using the encapsulation result to obtain the user's operational behavior data; cleans the operational behavior data to obtain model training samples; and incrementally fine-tunes the intent recognition model using the model training samples to obtain the adjustment result. This solves the problems of cumbersome traditional multi-system operations and inaccurate intent recognition in existing intelligent customer service systems, which leads to the inability to automatically pre-fill forms, thus enabling form filling and improving form filling efficiency. Based on this invention, addressing the issues of low accuracy in intent recognition in real-world intelligent customer service technologies, difficulty in handling the diversity and complexity of user expressions, difficulty in integrating business systems, lack of a unified scheduling mechanism, insufficient process automation (inability to automatically complete form pre-filling and process initiation, still requiring a large amount of manual operation, resulting in low efficiency), this invention designs a form filling method based on AI intent recognition. The effectiveness of this AI intent recognition-based form filling method is verified during form filling, and the efficiency of form filling is significantly improved using this method.

[0025] In this embodiment, for ease of description, the form filling device based on AI intent recognition will be used as the execution subject for the following description.

[0026] As enterprises continue their digital transformation, employees frequently interact with multiple business systems in their daily work. Under the traditional service model, employee workflows have significant drawbacks. One is the issue of switching between multiple systems, requiring employees to log into each business system separately, manually fill out forms, and retrieve information one by one. These multiple steps and cumbersome processes lead to decreased work efficiency. Another issue is the cumbersome operation itself. The traditional model lacks a unified service entry point, and various business processes need to be handled separately. Without unified scheduling, work efficiency will also decrease significantly. Furthermore, there is a lack of automation. Intelligent customer service should simplify operations, but current technology cannot automatically complete form pre-filling and process initiation, still requiring a lot of manual operation, which also affects work efficiency. Currently, intelligent customer service within enterprises also faces challenges in business adaptability. Since different business systems have different operating logics, terminology standards, and process requirements, if intelligent customer service is not specifically adapted and optimized, efficiency will also decrease.

[0027] This application provides a solution that identifies user questions through an intent recognition model and outputs a list of multiple intents and confidence levels. Then, through multi-level decision-making, the final intent is accurately determined, solving the problem of inaccurate identification. Subsequently, relying on the intent-service mapping table, business interfaces are called uniformly to automatically complete field extraction and form pre-filling, eliminating the need for switching between multiple systems and manual filling. This simplifies the operation and realizes automated form pre-filling, solving the problem of cumbersome traditional multi-system operations and the inaccurate intent recognition of existing intelligent customer service, which leads to the inability to automatically pre-fill forms, thus improving the efficiency of form filling.

[0028] Based on this, this application provides a form filling method based on AI intent recognition, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the form filling method based on AI intent recognition in this application.

[0029] In this embodiment, the form filling method based on AI intent recognition includes steps S01 to S03: Step S01: Receive user question information, identify the user question information through an intent recognition model, and obtain a list of several user intents and a list of intent confidence scores corresponding to the several user intents. Before the implementation of this embodiment, it should be clear that existing intelligent customer service technologies have significant drawbacks. First, the accuracy of intent recognition is low, making it difficult to cope with the diversity and complexity of user expressions. Second, business system integration is difficult, and there is a lack of a unified scheduling mechanism. In addition, there is insufficient automation of the process, and it is impossible to automatically complete form pre-filling and process initiation, still requiring a lot of manual operation.

[0030] Therefore, in order to solve the above problems, this embodiment first receives the question information sent by the user through the enterprise's commonly used interaction channels, preprocesses the question information and inputs it into the intent recognition model, and completes semantic parsing, intent classification and confidence calculation through the intent recognition model, and outputs a list of several user intents and corresponding confidence scores (arranged in descending order of scores), providing data support for subsequent multi-level decision-making.

[0031] Step S02: Perform multi-level decision-making on the several user intents according to the intent confidence score list to obtain the final user intent ID, and call the business interface through the intent service mapping table based on the final intent ID; After completing intent recognition and confidence score acquisition, multi-level decision-making is performed through the confidence score list. The final user intent ID is determined by comparing two thresholds. Then, the intent service mapping table is queried based on the ID to match the corresponding business service type, select the dedicated business system adapter, complete the protocol conversion and authentication encapsulation, and call the corresponding business interface.

[0032] Step S03: Extract fields and pre-fill forms through the business interface to obtain the encapsulation result, and send the encapsulation result to the user terminal corresponding to the user's question information.

[0033] Finally, the form metadata is obtained through the business interface, and the user's basic information and entity information in the question are extracted. After format conversion, field mapping and legality verification, the information is filled into the initial form to generate a pre-filled form. A form link is generated and relevant information is encapsulated and sent to the user for confirmation and submission.

[0034] Specifically, step S01 above, which involves identifying the user's question information using an intent recognition model to obtain several user intents and a list of intent confidence scores corresponding to those intents, includes: Step S011: Perform semantic understanding on the user's question information using the intent recognition model to obtain the user's semantics; Step S012: Classify intents based on the user semantics to obtain several candidate intents, and calculate the confidence scores of the several candidate intents based on the user semantics. Step S013: Extract entity auxiliary information and keyword auxiliary information from the user's question information using the intent recognition model; Step S014: Integrate the entity auxiliary information and keyword auxiliary information into the plurality of candidate intents to obtain a plurality of user intents and a list of intent confidence scores corresponding to the plurality of user intents.

[0035] First, the preprocessed user query information is input into the intent recognition model. The model uses its built-in natural language understanding module (adopting a Transformer architecture combined with an attention mechanism) to perform deep semantic analysis on the query information. It can not only recognize the literal meaning, but also understand complex situations such as contextual relationships, ambiguous expressions, and ellipsis. For example, if a user asks "Please take leave tomorrow", the model can accurately understand that the core need is "to initiate the leave application process", rather than simply informing the user of the time. Finally, the model obtains structured user semantics. The semantic data is stored in JSON format and includes fields such as core needs, contextual information, and expression methods. In addition, the process of deep semantic analysis through the natural language understanding module in this embodiment includes word segmentation, part-of-speech tagging, and syntactic analysis of the user query. Then, it combines the enterprise's internal business dictionary and contextual information to eliminate ambiguity in expression, and finally outputs structured user semantic data, providing a unified semantic foundation for subsequent intent classification and entity extraction.

[0036] Based on structured user semantics, and combined with a pre-defined business scenario classification system (such as intelligent question answering, internet question answering, intelligent data analysis, intelligent leave application, expense reimbursement application, and office supplies application, etc., with each primary scenario containing several secondary sub-scenarios), intent classification is performed. The classification process uses a multi-label classification algorithm, which is trained and optimized in conjunction with the enterprise's internal business corpus to ensure the accuracy of classification. Several candidate intents that highly match the user semantics are selected. The number of candidate intents is usually controlled to 3-5 to avoid the impact of too many on the efficiency of subsequent decision-making.

[0037] Simultaneously, the model's built-in confidence score calculation algorithm (Softmax function) combines the user's semantics with the matching degree of each candidate intent to calculate a confidence score for each candidate intent. The score ranges from 0 to 1, with higher scores indicating higher matching degrees. For example, if a user asks "Check last month's sales report," the confidence score for the candidate intent "Intelligent Data Inquiry - Sales Report Query" might be 0.92, the confidence score for "Intelligent Q&A - Sales Data Consultation" might be 0.05, and the confidence score for "Intelligent Leave Request" might be 0.03. The process of calculating the confidence score using the confidence score calculation algorithm involves first performing similarity matching between the user's semantic features and the feature vectors of each candidate intent to obtain a matching score; then, inputting the matching scores of all candidate intents into the Softmax function for normalization processing, mapping the scores to the 0-1 interval to obtain the confidence score for each candidate intent, ensuring that the sum of the confidence scores of all candidate intents is 1, thereby quantifying the matching degree between each intent and the user's needs.

[0038] Subsequently, the model's entity recognition module employs the Named Entity Recognition (NER) algorithm to extract entity auxiliary information and keyword auxiliary information from the user's query information. Entity auxiliary information includes time entities (such as "last month" and "tomorrow afternoon"), person entities (such as "Zhang San" and "department manager"), numerical entities (such as "half a day" and "500 yuan"), and category entities (such as "annual leave" and "sales report"). Keyword auxiliary information includes core words related to the business scenario (such as "report", "annual leave", "reimbursement", and "leave request") and action words (such as "check", "please", and "apply"). This information will serve as supplementary evidence for intent recognition, further improving the accuracy of intent recognition.

[0039] Finally, the extracted entity auxiliary information and keyword auxiliary information are associated and integrated with the corresponding candidate intents. Auxiliary information tags are added to each candidate intent to improve the information dimensions of each candidate intent. For example, tags such as the time entity "last month" and the keywords "report" and "query" are added to the candidate intent "intelligent data query - sales report query". In the end, an intent confidence score list containing several user intents, corresponding confidence scores and auxiliary information tags is formed. After being sorted in descending order of confidence scores, it provides comprehensive and accurate data support for subsequent multi-level decision-making.

[0040] More specifically, step S02 above, which involves making multi-level decisions on the several user intentions based on the intention confidence score list to obtain the final user intention ID, includes: Step S021: Extract the first user intent with the highest score from the intent confidence score list; Step S022: Compare the confidence score of the first user intent using the first intent threshold and the second intent threshold; Step S023: If the confidence score of the first user intent is greater than the first intent threshold, then the first user intent is determined to be the final user intent, and the final user intent ID of the final user intent is extracted. Step S024: If the confidence score of the first user intent is less than the first intent threshold and greater than the second intent threshold, then perform multiple rounds of clarification dialogue based on the several candidate intents to obtain the final user intent ID. Step S025: If the confidence score of the first user intent is less than the second intent threshold, the list of intent confidence scores is sent to the human customer service terminal, which then determines the final user intent ID.

[0041] like Figure 2 As shown, the system first quickly selects the user intent with the highest confidence score from the list of intent confidence scores sorted in descending order as the first user intent. At the same time, it calls the preset first intent threshold and second intent threshold from the dynamic configuration center through the interface. The initial value of the first intent threshold is set to 0.9 to judge the accuracy of intent recognition, and the initial value of the second intent threshold is set to 0.6 to judge the ambiguity of intent recognition. Both thresholds can be dynamically adjusted according to factors such as the complexity of the enterprise's business scenario and user expression habits. For example, for internal employees with standardized expression, the first threshold can be appropriately reduced to 0.85, and for scenarios with ambiguous expression, the second threshold can be appropriately increased to 0.65.

[0042] The confidence score of the first user intent is compared one by one with two thresholds to initiate a multi-level decision-making process: (1) If the confidence score of the first user intent is greater than the threshold of the first intent, it means that the intent recognition accuracy is extremely high and there is no ambiguity. The intent can be directly identified as the final user intent, and its corresponding final user intent ID can be extracted from the intent database. This ID is a unique identifier that corresponds one-to-one with the business service. After extraction, it directly enters the subsequent business interface call stage to ensure efficient process advancement.

[0043] (2) If the confidence score of the first user intent is between the second threshold and the first threshold, it indicates that the user intent is somewhat ambiguous and cannot be directly determined. At this time, a multi-round clarification dialogue process needs to be triggered to further clarify the user's core needs through interaction with the user. During the clarification dialogue process, targeted questions will be generated by combining the candidate intent and the context of the user's question until the final intent ID with the confidence level is obtained.

[0044] (4) If the confidence score of the first user intent is less than the second threshold, it means that the model cannot effectively identify the user intent. This may be because the user’s expression is too vague, there are obscure terms, or the business scenario is not covered by the model. In this case, the relevant data such as the intent confidence score list, user question information, and conversation context are packaged and sent to the human customer service terminal. The human customer service terminal will pop up a prompt message. Customer service personnel can combine this data to manually analyze the user’s core needs, determine the final user intent ID, and feed the ID back to the system to ensure the accuracy and comprehensiveness of intent recognition, avoid process interruption due to recognition failure, and improve the user experience.

[0045] Further, step S02 above, the step of calling the business interface through the intent service mapping table based on the final intent ID, includes: Step S026: Perform type matching on the intent service mapping table based on the final intent ID to obtain the business service type; Step S027: Select a business system adapter based on the business service type; Step S028: After performing protocol conversion and authentication information encapsulation on the business system adapter, the business interface is called through the business system adapter.

[0046] After obtaining the end-user intent ID, the system queries the intent service mapping table stored in the dynamic configuration center (such as Nacos or Apollo) through a preset interface. This mapping table adopts a dynamically updatable configuration form, which supports adding, modifying, and deleting the association between intents and business services online. It takes effect without restarting the system, which facilitates the subsequent business expansion of enterprises.

[0047] The mapping table contains several core fields, including intent ID (which corresponds one-to-one with the end-user intent ID), service name (such as "Enterprise Knowledge Base Service", "BI Platform Service", "HR Leave Service"), business service type (first-level category, such as "Intelligent Question Answering", "Intelligent Question Data", "HR Business"), adapter class name (used to locate the corresponding business system adapter), interface URL (access address of the business system interface), authentication method (such as OAuth2.0, API Key, username and password), request parameter format (such as JSON, XML), response parameter format, timeout, number of retries, and other detailed information.

[0048] The mapping table is precisely matched based on the end-user intent ID. The key-value pair query method is used to quickly locate the corresponding business service type. For example, when the end-user intent ID is "3", the matched business service type is "intelligent data inquiry" and the corresponding service name is "BI platform service". When the end-user intent ID is "4", the matched business service type is "HR business" and the corresponding service name is "HR leave service".

[0049] Based on the matched business service type, the corresponding business system adapter is selected from the preset set of business system adapters. Each business system has its own dedicated adapter. The adapter adopts a plug-in design, which can be deployed and updated independently. The interface call details, protocol specifications, exception handling logic (such as interface timeout and retry mechanism for call failure), data format conversion rules, etc. of the system are encapsulated internally, so there is no need to develop additional adaptation code, reducing the difficulty of system integration.

[0050] The selected business system adapter performs protocol conversion processing, supporting mutual adaptation of multiple mainstream protocols such as HTTP / REST, RPC, and SOAP. For example, the BI platform uses the RPC protocol, while the system core uses the HTTP / REST protocol. The adapter can automatically complete the conversion between the two protocols to ensure data transmission compatibility. At the same time, it uniformly encapsulates the authentication information of the business system and automatically selects the corresponding authentication method according to the requirements of the business system. For example, the BI platform uses API Key signature authentication, and the API Key and request parameters are encrypted and injected into the request header. The HR system uses OAuth2.0 token authentication, and the authentication interface is called first to obtain the token, and then the token is injected into the request header to ensure the security and legitimacy of the interface call.

[0051] After completing protocol conversion and authentication encapsulation, the business system adapter sends a call request to the corresponding business system interface. The request contains preset request parameters (dynamically generated according to the business scenario), and sets the timeout and number of retries. If the interface call fails, the adapter will retry according to the preset exception handling logic. If the retry fails, it will return a call failure message and indicate the specific reason for the failure, which will help the system troubleshoot the problem and achieve accurate and stable connection between intent and business service.

[0052] To illustrate, consider the scenario of intelligent data querying, such as... Figure 3 As shown, when the final intent ID is determined to be Smart Question (intent ID=3), the BI adapter first receives the interface call instruction sent by the system. After the adapter starts, it automatically initializes the internal configuration, including the core parameters such as the BI platform's interface URL, authentication method (API Key signature authentication is used by default), timeout (set to 30 seconds), and number of retries (set to 2 times), to ensure that it matches the communication parameters of the BI platform.

[0053] Subsequently, the BI adapter encapsulates the final intent ID, user identity identifier, session ID, and extracted entity auxiliary information (such as the time entity "last month" and the category entity "sales report") into standard RPC request parameters according to the preset protocol format. At the same time, it encrypts the API key and injects it into the request header to complete the authentication encapsulation, thus preventing the request from being illegally intercepted or tampered with.

[0054] After encapsulation, the BI adapter sends a call request to the report query interface of the BI platform. The request clearly specifies key information such as the report type, time range, and data dimension to be queried. This information is all derived from the entity extraction results in the user's query information.

[0055] After receiving the request, the BI platform verifies the validity of the request parameters. If the verification is successful, it retrieves the corresponding report data, organizes it according to the preset format, and generates structured data containing report name, query time, data details, and statistical results. The data is then fed back to the BI adapter via the RPC protocol.

[0056] After receiving feedback data, the BI adapter automatically performs protocol conversion, converting the RPC response format into the JSON format supported by the system core. At the same time, it cleans the data, removing invalid data and supplementing missing fields to ensure the integrity and standardization of the data.

[0057] The BI adapter then extracts core fields related to report queries (such as report ID, query time, and data statistics) from the processed structured report data, combined with form pre-filling requirements. These fields are then linked and integrated with the user's basic information to generate field data adapted for form pre-filling. This data is then synchronously sent to the form pre-filling engine, providing accurate data support for form pre-filling in subsequent intelligent query scenarios (such as report export forms and data subscription forms). The entire process requires no manual operation from the user, achieving a fully automated workflow from intent recognition to business interface calls and data processing.

[0058] This embodiment, through the above-described scheme, specifically receives user query information, identifies the user query information using an intent recognition model, obtains several user intents and a list of intent confidence scores corresponding to these intents, performs multi-level decision-making based on the intent confidence score list, obtains a final user intent ID, and calls a business interface through an intent service mapping table based on the final intent ID, extracts fields and pre-fills forms through the business interface, obtains an encapsulated result, and sends the encapsulated result to the user terminal corresponding to the user query information. Thus, by using an intent recognition model to identify user queries and output multiple intents and confidence lists, and then accurately determining the final intent through multi-level decision-making, the problem of inaccurate identification is solved. Subsequently, relying on the intent-service mapping table to uniformly call the business interface, field extraction and form pre-filling are automatically completed, eliminating the need for switching between multiple systems and manual filling, thereby simplifying operations and achieving automated form pre-filling. This solves the problem of cumbersome traditional multi-system operations and the inaccurate intent recognition of existing intelligent customer service systems, which leads to the inability to automatically pre-fill forms, thus improving the efficiency of form filling.

[0059] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 In step S024, the form filling method based on AI intent recognition further includes steps S0241-S0244, which involves conducting multiple rounds of clarification dialogue based on the candidate intents to obtain the final user intent ID. Step S0241: Generate clarifying questions based on the candidate intentions; Step S0242: The clarification question is sent to the user terminal, and the user terminal generates a response based on the clarification question; Step S0243: Receive the reply information sent by the user terminal, and perform context splicing and semantic fusion on the reply information, clarification question and user question information to obtain fused text; Step S0244: Perform intent recognition on the fused text using the intent recognition model to obtain the final user intent ID.

[0060] First, the top three candidate intents (with confidence scores between the second and first thresholds) are selected from the intent confidence score list. The core features, entity auxiliary information, and keywords of each candidate intent are extracted. Combined with the context of the user's original question, concise and targeted clarifying questions are generated, avoiding redundant wording and ensuring the user can quickly understand and respond. For example, if the user's original question is "I want to check the report," and the candidate intents are "Smart Data - Sales Report Query," "Smart Data - Financial Report Query," and "Smart Q&A - Report Operation Consultation," then the generated clarifying question would be "Do you want to query the sales report, financial report, or consult about report-related operation methods?" The question clearly lists the core differences between the candidate intents, guiding the user to make an accurate selection.

[0061] The generated clarification question is then encapsulated into a message format that can be recognized by the user (adapting to different interaction channels such as mobile and PC), and sent to the user through the message push interface. At the same time, the clarification question, candidate intent list and conversation context are cached locally in the system to ensure that subsequent reply information can be accurately associated.

[0062] After receiving a clarification question, the user client can generate a reply (such as a "sales report") using convenient methods such as text or radio buttons, and then send the reply to the system.

[0063] After receiving the user's reply, the system first preprocesses the reply (removing invalid characters and semantic standardization). Then, it concatenates the reply, clarification question, and the user's original question in chronological order to form a complete conversational text: "User asks: I want to check the report → System clarifies: Do you want to check the sales report, financial report, or consult about report-related operations? → User replies: Sales report." Then, a semantic fusion algorithm is used to deeply fuse the semantics of the three, eliminate differences in expression, and extract the core semantics ("The user needs to check the sales report"), resulting in a logically coherent and semantically clear fused text.

[0064] Finally, the fused text is input into the intent recognition model. The model combines the core semantics, contextual relationships, and candidate intent features in the fused text to re-recognize intent and calculate confidence. At this point, due to the addition of the user's clarification response, the accuracy of intent recognition is greatly improved. It can usually obtain the final intent with a confidence score greater than the first intent threshold, and then extract the corresponding final user intent ID (such as ID=3 for "Intelligent Questions - Sales Report Query") to complete the entire process of multi-round clarification dialogue, ensuring the accuracy of the final intent ID, while avoiding triggering human customer service intervention and improving process efficiency.

[0065] Specifically, in the above embodiments, step S03, which involves extracting fields and pre-filling forms through the business interface to obtain a packaged result, and then sending the packaged result to the user terminal corresponding to the user's question information, includes: Step S031: Extract the basic user information corresponding to the user terminal through the form pre-filling engine, and extract the entity information in the user question information according to the final user intent ID; Step S032: Call the form metadata of the business system corresponding to the business interface through the business interface; Step S033: Based on the form metadata, perform format conversion and field mapping on the user basic information and entity information to obtain the field mapping result; Step S034: Perform a validity check on the field mapping result, and after the validity check passes, fill the field mapping result into the initial form; Step S035: Based on the initial form, generate a pre-filled form through the business interface, and generate a form link through the pre-filled form; Step S036: Encapsulate the form link and the process information of the pre-filled form to obtain the encapsulation result, and send the encapsulation result to the user terminal.

[0066] After the form pre-filling engine starts, it first obtains the user token (JWT) through the session context, calls the user information query interface of the Enterprise Unified Identity Authentication Service (SSO), passes in the user token for identity verification, and after successful verification, extracts the user's basic information related to the current business scenario from the user information database. For example, in the smart leave application scenario, it extracts the employee number, name, department, and remaining leave balance, etc., and in the smart expense reimbursement scenario, it extracts the employee number, name, reimbursement amount, bank card information, etc. After extraction, the user's basic information is cached in the local context to avoid repeated API calls.

[0067] Simultaneously, the current business scenario is determined based on the end-user intent ID, and the built-in lightweight NER model is invoked to accurately extract entity information from the user's question information. The scope of entity information extraction is strongly correlated with the business scenario. For example, in the intelligent leave application scenario, leave time, leave type, leave duration, and leave reason are extracted. In the intelligent data query scenario, report type, query time, data dimension, etc. are extracted. After extraction, the entity information is initially verified to remove invalid or incomplete entities (such as vague time descriptions).

[0068] Subsequently, a form metadata query request is sent to the corresponding business system through the invoked business interface. The request carries the end-user intent ID and user identity identifier. After receiving the request, the business system matches the corresponding form template according to the intent ID, extracts all metadata information of the form template, including field name, field type, field length, required field identifier, default value, enumeration value range, validation rules, etc., and encapsulates the metadata information into JSON format before feeding it back to the system.

[0069] Then, based on the received form metadata, a mapping relationship is established between user basic information, entity information and form fields, and format conversion and field mapping are performed: field name mapping converts general semantic fields into business system form fields (such as "Name" → "emp_name", "Department" → "dept_code"), and value conversion converts non-standard values ​​expressed by users into standard values ​​that the form can recognize (such as "Half a day" → 0.5, "Annual leave" → enumerated value "1", "Tomorrow" → "YYYY-MM-DD" standard date), and finally obtains standardized and adapted field mapping results.

[0070] The field mapping results also need to undergo multi-dimensional validity checks, including field format checks (whether the date, value, and enumeration value meet the requirements), required field completeness checks (whether any required fields are missing), and business rule compliance checks (such as leave duration ≤ remaining holiday balance, reimbursement amount ≤ reimbursement limit). If the checks fail, a prompt message containing the reason for the failure (such as "Leave duration exceeds remaining holiday balance, please adjust") is generated and fed back to the user.

[0071] If the validation passes, the field mapping results will be automatically filled into the preset initial form template to generate a pre-filled form. The pre-filled form will mark the pre-filled fields and the required fields that have not been filled, making it easy for users to view.

[0072] Next, the form creation interface of the business system is called through the business interface to send the pre-filled form data to the business system. The business system generates a draft ID or application number of the pre-filled form. Combined with the basic access URL of the business system, a form link containing identity authentication parameters (user token) is generated to ensure that only the current user can access the form and avoid information leakage.

[0073] Finally, the form link, the core information of the pre-filled form (list of pre-filled fields, required fields not filled, form submission process steps, estimated processing time), API call status, error codes, etc., are encapsulated into a standard JSON format and sent to the user terminal via streaming. After receiving the data, the user can directly click the form link to enter the pre-filled page, confirm that the information is correct, fill in the missing content, and submit, greatly simplifying the user operation.

[0074] like Figure 5As shown, taking a user-initiated leave application process (final intent ID=4, corresponding to HR business - intelligent leave application service) as an example, the user sends the question information "Please take half a day of annual leave tomorrow, the reason is family matters" through WeChat Work. After the system determines that the final intent ID is intelligent leave application, the form pre-filling engine calls the SSO service through the user token to extract the user's basic information (employee number: 10001, name: Zhang San, department: marketing department, remaining annual leave balance: 3 days), and at the same time extracts the entity information through the NER model (leave time: tomorrow, leave type: annual leave, leave duration: 0.5 days, leave reason: family matters).

[0075] Subsequently, the HR system's leave form metadata interface is called through the HR system adapter to obtain the metadata of the leave form (fields include emp_id, emp_name, dept_code, leave_type, leave_start_time, leave_duration, leave_reason, etc., where leave_type is an enumeration value, annual leave = 1, personal leave = 2, and leave_duration is in days). Based on the obtained leave form metadata, field mapping and format conversion are performed, mapping "Name: Zhang San" to "emp_name: Zhang San", "Department: Marketing Department" to "dept_code: SC001", "Annual leave" to "leave_type: 1", "Tomorrow" to "leave_start_time: YYYY-MM-DD", and "0.5 days" to "leave_duration: 0.5", thus obtaining the field mapping results.

[0076] Validate the field mapping results to confirm that the leave duration is 0.5 days or less and the remaining annual leave balance is 3 days, all required fields are complete, and the enumeration values ​​match. After successful validation, fill the form into the initial leave application form, call the HR system's leave draft creation interface to generate a draft ID: QY2024001, and concatenate the form link "http: / / hr system domain / leave / form?draftId=QY2024001&token=xxx".

[0077] Finally, the form link, pre-filled information (pre-filled employee information and leave-related information), and process steps (confirm information → submit for approval → wait for approval) are packaged and sent to the user. After clicking the link, the user will be directly taken to the pre-filled leave form page. There is no need to manually fill in the pre-filled information such as employee information and leave time. The user only needs to confirm that the information is correct and click "submit" to complete the leave application process.

[0078] More specifically, after step S03 above, which involves extracting fields and pre-filling forms through the business interface to obtain a packaged result, and then sending the packaged result to the user terminal corresponding to the user's question information, the method further includes: Step S04: Receive the archiving process form submitted by the user terminal; Step S05: Collect data from the archive process form using the encapsulation result to obtain the user's operation behavior data; Step S06: Clean the operation behavior data to obtain model training samples; Step S07: Incrementally fine-tune the intent recognition model using the model training samples to obtain the adjustment result.

[0079] After the user submits a pre-filled form via a form link, the business system completes the form review and archiving process. Then, it sends the archived process form (containing complete information submitted by the user, form processing status, approval records, etc.) to this system through the business interface. The system receives the archived process form, extracts the core data from the form (user ID, final intent ID, form type, submission time, completeness of pre-filled fields, supplementary fields, etc.), and associates and stores it with the previously sent encapsulated results to ensure data traceability.

[0080] Based on the encapsulation results, comprehensive data collection is carried out on the archived process forms. The collected operational behavior data includes user operation behavior (such as whether pre-filled fields are modified, the type of field modified, the content to be filled in, and the time taken to submit the form), form pre-filling effect data (accuracy rate of pre-filled fields, completion rate of required field pre-filling), intent matching data (the degree of matching between the final intent ID and the user's actual needs), etc. All collected data are associated with user identity and session ID to form a structured behavior dataset.

[0081] The collected operational behavior data then undergoes multi-step data cleaning to remove invalid data (such as abnormally submitted empty forms and duplicate data submissions), correct erroneous data (such as incorrectly formatted times and values), and eliminate outliers (such as data with excessively long or short submission times). At the same time, the data is standardized (such as unifying time formats and standardizing field naming). Data that meets the requirements for model training is selected and divided into training and validation sets according to a preset ratio (such as 8:2) to obtain high-quality model training samples, ensuring the accuracy and effectiveness of the training samples.

[0082] Then, the intention recognition model is optimized by incremental fine-tuning. The training set is input into the model, the underlying parameters of the model are frozen, and only the upper-layer parameters related to intention recognition and entity extraction are fine-tuned. The model performance (intention recognition accuracy and confidence stability) is monitored in real time with the validation set to avoid model overfitting.

[0083] After fine-tuning, the model adjustment results are obtained, including the fine-tuned model parameters, data on the improvement in intent recognition accuracy, data on the improvement in entity extraction accuracy, etc. The system automatically updates the adjusted model parameters to the production environment, realizing continuous optimization of the intent recognition model, making the model more in line with the expression habits and business needs of enterprise users, and further improving the accuracy of intent recognition.

[0084] This embodiment, through the above-described scheme, specifically generates a clarifying question based on several candidate intents; sends the clarifying question to the user terminal, where the user terminal generates a response based on the clarifying question; receives the response from the user terminal, performs contextual concatenation and semantic fusion on the response, clarifying question, and user question information to obtain fused text; and uses the intent recognition model to perform intent recognition on the fused text to obtain the final user intent ID. Thus, by using the intent recognition model to identify user questions and output a list of multiple intents and confidence levels, and then accurately determining the final intent through multi-level decision-making, the inaccurate identification problem is solved. Subsequently, relying on the intent-service mapping table to uniformly call business interfaces, field extraction and form pre-filling are automatically completed, eliminating the need for switching between multiple systems and manual filling, thereby simplifying operations and achieving automated form pre-filling. This solves the problem of cumbersome traditional multi-system operations and the inaccurate intent recognition of existing intelligent customer service systems, which leads to the inability to automatically pre-fill forms, thus improving the efficiency of form filling.

[0085] For example, to help understand the implementation process of the form filling method based on AI intent recognition obtained by combining this embodiment with the above embodiment one, please refer to... Figure 6 , Figure 6 A simplified flowchart of a form filling method based on AI intent recognition is provided, specifically: First, users input and submit questions in natural language through interactive channels such as mobile apps, PC web pages, and WeChat for business. After receiving the information, the system performs preprocessing operations such as text cleaning, word segmentation, and noise reduction. Then, the standardized information is input into the intent recognition model, which performs semantic parsing, intent classification, and confidence calculation, and outputs a list of multiple candidate user intents and their corresponding confidence scores.

[0086] Secondly, the system performs multi-level decision-making based on a list of confidence scores. It compares scores with a first threshold and a second threshold, directly identifies high-confidence intentions, initiates multiple rounds of clarification dialogue for medium-confidence intentions, and transfers low-confidence intentions to manual processing, ultimately obtaining a unique end-user intention ID.

[0087] Next, the intent is matched with the business service based on the intent-service mapping table, and the corresponding business system interface is called. Then, the form metadata of the target business system is obtained through the called business interface, and basic information is extracted from the user identity information, and entity information such as time, type and duration are extracted from the user's question.

[0088] According to the form metadata requirements, the above information is formatted, mapped, and validated. The validated content is automatically filled into the initial form to generate a pre-filled business form. Then, a form link that can be directly accessed and edited is generated based on the pre-filled form. The link and process information are uniformly encapsulated and sent to the user terminal.

[0089] Finally, after the user confirms and submits the form, the business system completes the process archiving. The system receives the archived process form, collects user action data during the form filling and submission process based on the encapsulated results, cleans, deduplicates, and standardizes the data to generate high-quality training samples, and then uses the samples to incrementally fine-tune and update the parameters of the intent recognition model to continuously improve the intent recognition and form pre-filling effects.

[0090] This application can achieve the following: Figure 6 The process shown relies on, for example, Figure 7 The system architecture shown is as follows: The system in this application uses a user interaction module as the front-end entry point to receive user questions, display clarification dialogue content, and return form links and process results to the user.

[0091] After a user requests to enter the system, the request is first transmitted to the intent recognition engine, which performs semantic parsing, multi-intent output, and confidence calculation to obtain a list of candidate intents and confidence scores.

[0092] The system then proceeds to the multi-level decision-making and routing module. This module performs multi-level judgments based on the confidence threshold, determines the final intent ID through direct confirmation, multiple rounds of clarification, or manual assistance, and automatically matches and calls the corresponding business interface according to the intent-service mapping table.

[0093] Business interface calls are completed by a unified business adapter layer. This layer encapsulates the interface logic of multiple business systems such as knowledge base, BI platform, and HR system, and realizes protocol conversion, authentication encapsulation, parameter adaptation and result parsing, thus shielding the differences between heterogeneous systems.

[0094] After determining the business scenario, the form pre-filling engine performs field extraction, information mapping, format conversion, and legality verification, automatically filling in the form and generating an accessible link. After encapsulation, the link is returned to the user. Meanwhile, the system has a feedback learning and model optimization module, which collects user submission, modification, confirmation, and other operation data. After data cleaning, training samples are formed to incrementally fine-tune the intent recognition model, enabling continuous iteration of the system's capabilities.

[0095] pass Figure 7 The architecture shown enables this application to stably achieve accurate recognition of multiple intents, unified scheduling of multiple systems, and automated form filling, with high overall efficiency and strong scalability.

[0096] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the form filling method based on AI intent recognition in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0097] This application also provides a form filling device based on AI intent recognition, please refer to... Figure 8 The form filling device based on AI intent recognition includes: The receiving module 10 is used to receive user question information, identify the user question information through an intent recognition model, and obtain a list of several user intents and a list of intent confidence scores corresponding to the several user intents. The calling module 20 is used to make multi-level decisions on the several user intentions based on the intention confidence score, obtain the final user intention ID, and call the business interface through the intention service mapping table based on the final intention ID. The encapsulation module 30 is used to extract fields and pre-fill forms through the business interface to obtain the encapsulation result, and send the encapsulation result to the user terminal corresponding to the user question information.

[0098] The form filling device based on AI intent recognition provided in this application, employing the form filling method based on AI intent recognition in the above embodiments, can solve the technical problems of cumbersome traditional multi-system operations and inaccurate intent recognition in existing intelligent customer service systems, thus preventing automatic form pre-filling. Compared with the prior art, the beneficial effects of the form filling device based on AI intent recognition provided in this application are the same as those of the form filling method based on AI intent recognition provided in the above embodiments, and other technical features in the form filling device based on AI intent recognition are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0099] This application provides a form filling device based on AI intent recognition. The form filling device based on AI intent recognition includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the form filling method based on AI intent recognition in the above embodiment 1.

[0100] The following is for reference. Figure 9 This document illustrates a structural diagram of a form-filling device based on AI intent recognition suitable for implementing embodiments of this application. The form-filling device based on AI intent recognition in these embodiments may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 9 The AI ​​intent recognition-based form filling device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0101] like Figure 9As shown, the AI ​​intent recognition-based form filling device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the AI ​​intent recognition-based form filling device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the AI ​​intent recognition-based form filling device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows an AI intent recognition-based form filling device with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0102] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0103] The form filling device based on AI intent recognition provided in this application, employing the form filling method based on AI intent recognition in the above embodiments, can solve the technical problems of cumbersome traditional multi-system operations and inaccurate intent recognition in existing intelligent customer service systems, thus preventing automatic form pre-filling. Compared with the prior art, the beneficial effects of the form filling device based on AI intent recognition provided in this application are the same as those of the form filling method based on AI intent recognition provided in the above embodiments, and other technical features in this form filling device based on AI intent recognition are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0104] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0105] The above description is merely a specific embodiment 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.

[0106] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the form filling method based on AI intent recognition in the above embodiments.

[0107] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0108] The aforementioned computer-readable storage medium may be included in an AI intent recognition-based form filling device; or it may exist independently and not assembled into an AI intent recognition-based form filling device.

[0109] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by an AI intent recognition-based form filling device, the AI ​​intent recognition-based form filling device: receives user query information; identifies the user query information using an intent recognition model to obtain several user intents and a list of intent confidence scores corresponding to the several user intents; performs multi-level decision-making on the several user intents based on the intent confidence score list to obtain a final user intent ID; and calls a business interface through an intent service mapping table based on the final intent ID; performs field extraction and form pre-filling through the business interface to obtain an encapsulation result; and sends the encapsulation result to the user terminal corresponding to the user query information.

[0110] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0111] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0112] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0113] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described form filling method based on AI intent recognition. This solves the technical problems of cumbersome traditional multi-system operations and inaccurate intent recognition in existing intelligent customer service systems, which prevents automatic form pre-filling. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the form filling method based on AI intent recognition provided in the above embodiments, and will not be repeated here.

[0114] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the form filling method based on AI intent recognition as described above.

[0115] The computer program product provided in this application can solve the technical problems of cumbersome operation of traditional multi-systems and inaccurate intent recognition of existing intelligent customer service, which leads to the inability to automatically pre-fill forms. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the form filling method based on AI intent recognition provided in the above embodiments, and will not be repeated here.

[0116] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A form filling method based on AI intent recognition, characterized in that, The form filling method based on AI intent recognition includes: Receive user query information, identify the user query information through an intent recognition model, and obtain a list of several user intents and a list of intent confidence scores corresponding to the several user intents; Based on the intent confidence score list, a multi-level decision is made on the several user intents to obtain the final user intent ID, and the business interface is called through the intent service mapping table based on the final intent ID. The business interface is used to extract fields and pre-fill forms to obtain encapsulated results, which are then sent to the user terminal corresponding to the user's question information.

2. The form filling method based on AI intent recognition as described in claim 1, characterized in that, The step of identifying the user's question information using an intent recognition model to obtain several user intents and a list of intent confidence scores corresponding to the several user intents includes: The user's semantics are obtained by performing semantic understanding on the user's question information using the intent recognition model. The intent is classified according to the user semantics to obtain several candidate intents, and the confidence scores of the several candidate intents are calculated through the user semantics. The intent recognition model is used to extract entity auxiliary information and keyword auxiliary information from the user's question information. The entity auxiliary information and keyword auxiliary information are integrated into the candidate intents to obtain a list of user intents and a list of intent confidence scores corresponding to the user intents.

3. The form filling method based on AI intent recognition as described in claim 1, characterized in that, The step of performing multi-level decision-making on the plurality of user intentions based on the intention confidence score list to obtain the final user intention ID includes: Extract the first user intent with the highest score from the intent confidence score list; The confidence score of the first user intent is compared using the first intent threshold and the second intent threshold. If the confidence score of the first user intent is greater than the first intent threshold, then the first user intent is determined to be the final user intent, and the final user intent ID of the final user intent is extracted. If the confidence score of the first user intent is less than the first intent threshold but greater than the second intent threshold, then multiple rounds of clarification dialogue are conducted based on the candidate intents to obtain the final user intent ID. If the confidence score of the first user intent is less than the second intent threshold, the list of intent confidence scores is sent to the human customer service terminal, which then determines the final user intent ID.

4. The form filling method based on AI intent recognition as described in claim 3, characterized in that, The step of conducting multiple rounds of clarification dialogue based on the candidate intents to obtain the final user intent ID includes: Based on the aforementioned candidate intentions, clarifying questions are generated; The clarification question is sent to the user terminal, and the user terminal generates a response based on the clarification question. The system receives the reply information sent by the user terminal, performs context splicing and semantic fusion on the reply information, clarification questions, and user questions to obtain fused text; The intent recognition model is used to identify the intent of the fused text to obtain the final user intent ID.

5. The form filling method based on AI intent recognition as described in claim 1, characterized in that, The step of calling the business interface through the intent service mapping table based on the final intent ID includes: The business service type is obtained by matching the type of the intent service mapping table based on the final intent ID. Select the business system adapter based on the aforementioned business service type; After performing protocol conversion and authentication information encapsulation on the business system adapter, the business interface is called through the business system adapter.

6. The form filling method based on AI intent recognition as described in claim 1, characterized in that, The step of extracting fields and pre-filling forms through the business interface to obtain a packaged result, and sending the packaged result to the user terminal corresponding to the user's question information, includes: The form pre-filling engine extracts the basic user information corresponding to the user terminal, and extracts the entity information from the user's question information based on the end user intent ID; The form metadata of the business system corresponding to the business interface is invoked through the business interface. Based on the form metadata, the user basic information and entity information are formatted and mapped to fields to obtain the field mapping result; Perform a validity check on the field mapping result, and populate the field mapping result into the initial form after the validity check passes; Based on the initial form, a pre-filled form is generated through the business interface, and a form link is generated through the pre-filled form; The form link and the process information of the pre-filled form are encapsulated to obtain an encapsulation result, and the encapsulation result is sent to the user terminal.

7. The form filling method based on AI intent recognition as described in claim 1, characterized in that, After the steps of extracting fields and pre-filling forms through the business interface to obtain the encapsulated result, and sending the encapsulated result to the user terminal corresponding to the user's question information, the method further includes: Receive the archiving process form submitted by the user client; Data is collected from the archive process form using the encapsulation result to obtain the user's operation behavior data; The operational behavior data is cleaned to obtain model training samples; The intent recognition model is incrementally fine-tuned using the training samples of the model to obtain the adjustment result.

8. A form filling device based on AI intent recognition, characterized in that, The form filling device based on AI intent recognition includes: The receiving module is used to receive user query information, identify the user query information through an intent recognition model, and obtain a list of several user intents and a list of intent confidence scores corresponding to the several user intents. The calling module is used to make multi-level decisions on the several user intents based on the intent confidence score, obtain the final user intent ID, and call the business interface through the intent service mapping table based on the final intent ID. The encapsulation module is used to extract fields and pre-fill forms through the business interface to obtain the encapsulation result, and send the encapsulation result to the user terminal corresponding to the user's question information.

9. A form filling device based on AI intent recognition, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the form filling method based on AI intent recognition as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the form filling method based on AI intent recognition as described in any one of claims 1 to 7.

Images