Self-supervision method and system for promoting chat robot to utilize unintegrated service

Abstract

The invention provides a self-supervision method and system for promoting a chat robot to utilize unintegrated services, and belongs to the technical field of computer service and software engineering. The method comprises the following steps: firstly, receiving a user input demand, judging whether an unintegrated service is needed or not, if not, calling an integrated service, returning a user answer, and if so, searching a related service set; performing multi-level knowledge collection; performing structured analysis, and preliminarily generating a service document; generating-evaluating an iterative mechanism optimization service document, then judging whether the document is successfully called, if the document is successfully called, adding the document to the integrated service, then calling the integrated service, returning a user answer, and ending a program; and if the calling is not successful, returning to the multi-level knowledge acquisition step. According to the method and the system, autonomous discovery, understanding and calling of the unintegrated service are realized, the service calling coverage rate and the service calling success rate are improved, and the manual intervention cost is remarkably reduced.

Description

Technical Field

[0001] This invention relates to a self-monitoring method and system for promoting chatbots to utilize unintegrated services, belonging to the fields of computer service technology and software engineering technology. Background Technology

[0002] To accelerate the development of the digital service industry, it is essential to deepen the innovative application of artificial intelligence technology and construct dynamic method systems capable of autonomously discovering, understanding, and invoking unintegrated services. This will break through the traditional chatbot's dependence on pre-integrated services, enabling adaptive expansion of service capabilities and a leap in efficiency. Traditional methods collect and construct diverse datasets from real-world service scenarios, fine-tuning the generative models relied upon by chatbots to give them the ability to generalize to other services. However, these methods implicitly require services to have complete documentation and explanatory information. In practice, many services lack dedicated management and are often accompanied by non-standard and incomplete documentation, leading to the chatbot's weakness when using unintegrated services. Designing a new paradigm for the entire process of using unintegrated services provides a solution that can be easily applied to various application scenarios, laying the technological foundation for seamless service integration in the Web 3.0 era and a broader service ecosystem in the future. Summary of the Invention

[0003] The purpose of this invention is to solve the problems existing in the prior art, and to provide a self-monitoring method and system for promoting chatbots to utilize unintegrated services.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A self-monitoring system for facilitating chatbots to utilize unintegrated services, the system comprising: a service repository module, a service management module, a demand matching module, a generation-evaluation iterative optimization module, and an interaction control module;

[0006] The signal input terminals of the service repository module are connected to the signal output terminals of the service management module, the requirement matching module, and the generation-evaluation-iteration optimization module. The signal output terminals of the service repository module are connected to the signal input terminals of the service management module and the requirement matching module. The signal input and output terminals of the requirement matching module are connected to the signal input and output terminals of the generation-evaluation-iteration optimization module. The signal output terminal of the interaction control module is connected to the signal input terminal of the requirement matching module. The signal input and output terminals of the generation-evaluation-iteration optimization module are connected to the signal input and output terminals of the interaction control module.

[0007] The service repository module is used to match user requests to the chatbot and provide ready-to-use service support; it receives management signals from the service management module and provides service content to the service management module; it receives storage signals and service content from the generation-evaluation-iteration optimization module; and it receives interaction signals from the request matching module and provides service content to the request matching module.

[0008] The service management module supports lifecycle management of services already stored in the service repository module. It supports manual entry, automatic discovery, and soft deletion mechanisms; it supports version control interfaces to distinguish between new and old services and mark expired content; it supports an exception protection mechanism to prevent service removal from causing interruption of dependency processes; it sends management signals to the service repository module and accepts service content provided by the service repository module.

[0009] The demand matching module is used to understand user intent and accurately associate available services; it sends retrieval signals to the service repository module and receives service content from the service repository module; it receives demand signals from the interaction module; it sends service content and driving signals to the generation-evaluation iterative optimization module and receives feedback signals from the generation-evaluation module.

[0010] The generation-evaluation-iteration optimization module is used to generate standardized interface service documents and verify their compliance; it receives service content and driving signals from the demand matching module and sends feedback signals to the demand matching module; it sends storage signals and service content to the service storage module; it receives interaction signals from the interaction control module and sends interaction signals to the interaction control module.

[0011] The interactive control module is used to receive user feedback through a question-and-answer interface and control the number of matching services, verification conditions, and process jump logic; it sends interactive signals to the demand matching module; it sends interactive signals to the generation-evaluation iterative optimization module and receives interactive signals returned by the generation-evaluation module.

[0012] Preferably, the service repository module includes: an unintegrated service repository and an integrated service repository. Through a self-supervised workflow, services in the unintegrated service repository are continuously integrated into the integrated service repository.

[0013] The unintegrated service repository is used to store relevant unintegrated open services obtained from extensive web searches.

[0014] An integrated service library is used to store the basic content and service documentation of integrated services.

[0015] Preferably, the demand matching module includes: a semantic parsing unit, a multi-dimensional matching unit, and a service retrieval unit; the semantic parsing unit embeds user demands and service documents into semantic vectors, the multi-dimensional matching unit uses the semantic vectors parsed by the semantic parsing unit to calculate multi-dimensional similarity, and the service retrieval unit sorts services based on the similarity obtained by the multi-dimensional matching unit and coordinates the execution of subsequent steps;

[0016] Semantic parsing unit, used to extract deep user intent;

[0017] The multi-dimensional matching unit is used to comprehensively measure the degree of matching between user queries and service documents from different dimensions by employing multiple similarity calculation methods.

[0018] The service retrieval unit is used to retrieve services from the unintegrated service library based on the similarity in the multi-dimensional matching unit and activate the unintegrated service call process.

[0019] Preferably, the generation-evaluation iterative optimization module includes: a document generation unit, a multi-level verification unit, and a feedback iterative unit. The document generated by the document generation unit will be quality verified by the multi-level verification unit. If it passes the multi-level verification unit, the process ends; otherwise, feedback is provided through the feedback iterative unit, and the process returns to the document generation unit for generation.

[0020] The document generation unit is used to convert unstructured service descriptions into standardized interface documents;

[0021] Multi-level verification units are used to verify document compliance through a triple mechanism of rule constraints, logical verification, and manual intervention;

[0022] The feedback iteration unit is used to ensure reliable access and performance improvement for unintegrated services.

[0023] A self-supervised method based on a self-supervised system to facilitate chatbots' use of unintegrated services includes the following steps:

[0024] Step S1: Receive user input requirements, extract user input requirements through the semantic parsing unit in the requirement matching module, check whether there is a service interface matching the user input requirements in the integrated service library in the service repository module. If yes, proceed to step S7; otherwise, call the unintegrated service to proceed to step S2.

[0025] Step S2: Based on the user's input requirements, retrieve a set of services related to the user's input requirements from the unintegrated service library in the service repository module through the multi-dimensional matching unit in the requirements matching module;

[0026] Step S3: Through the service retrieval unit in the demand matching module, perform multi-level knowledge collection on the most relevant services in the service set retrieved in step S2;

[0027] Step S4: Through the document generation unit in the generation-evaluation-iteration optimization module, the most relevant service information is structured and parsed to generate standardized service documents; and the compliance of the standardized service documents is verified through the multi-level verification unit in the generation-evaluation-iteration optimization module.

[0028] Step S5: Based on the retrieval results of step S2, the knowledge collected in step S3, and the parsed information in step S4, the service document is optimized using the generation-evaluation iteration mechanism through the feedback iteration unit in the generation-evaluation iteration optimization module.

[0029] Step S6: Call the service interface corresponding to the optimized service document through the generation-evaluation iteration mechanism. If the call is successful, add the optimized service document to the integrated service library, fine-tune the chatbot, and then proceed to step S7; if the call fails, return to step S3.

[0030] Step S7: Execute the integrated service that matches the user's input requirements and return the result to the user.

[0031] Preferably, the method for checking whether there is a service interface matching the user's input requirements in the integrated service library of the service repository module in step S1 is as follows:

[0032] The interactive control module receives the user's input description of their needs, as well as the search preference parameters adjusted by the user through the interactive control module. It determines whether to invoke unintegrated services based on the degree of need matching. The degree of need matching is calculated by the cosine similarity between the user's need vector q and the integrated service interface description text vector t. The calculation formula is as follows:

[0033]

[0034] A match is considered to be found when the degree of matching exceeds the search preference parameter. The default search preference parameter is 0.5. Users can adjust the search preference parameter to make the chatbot select more services that are not integrated or integrated. The higher the search preference parameter, the higher the matching requirement and the more likely it is to select services that are not integrated.

[0035] The multi-dimensional similarity matching method described in step S2 includes the following steps:

[0036] Construct a service similarity function based on semantic embedding; and perform weighted ranking by combining the cosine similarity, keyword overlap rate, and contextual relevance between the semantics of user input requirements and service documents in the unintegrated service library.

[0037] The service similarity function is defined as follows:

[0038]

[0039] The service similarity function measures the degree of matching between user query q and service document d from three dimensions, including cosine similarity. To express semantic similarity, keyword overlap rate Jaccard similarity is used to calculate word-level similarity and contextual relevance. Co-occurrence of keywords in the current dialogue history is statistically analyzed, and the similarity of the context is calculated.

[0040] Preferably, the multi-level knowledge acquisition in step S3 includes: content knowledge, style knowledge, and logical knowledge; specifically as follows:

[0041] (1) Content knowledge: Supplement the missing fields in the service documents of the most relevant services in the service set retrieved in step S2. The content comes from the operation description of the operation with the same name, the parameter description of the parameter with the same name and the calling example in the integrated service library;

[0042] (2) Style knowledge: Analyze the consistency characteristics of naming conventions, commenting styles and implementation patterns of service groups with the same source. Service groups with the same source are services in the unintegrated service library that have the same source as the service.

[0043] (3) Logical knowledge: Construct a role profile through the most relevant service call chain, including preconditions and subsequent service dependencies.

[0044] Preferably, the specific steps of step S4 are as follows:

[0045] (1) Perform a cleanup operation on non-standard service documents and use an automated parser to extract the core elements of the service, including: service description, operation description, operation type, parameter name, and parameter type;

[0046] (2) Construct a multi-format conversion module to split the service document into a three-level field structure of service, interface and parameter, and convert the unstructured document into an intermediate representation of JSON format.

[0047] Preferably, the generation-evaluation iteration mechanism in step S5 includes the following steps:

[0048] (1) System prompt template: Defines the format specifications, parameter constraints and iterative optimization rules for service document generation;

[0049] (2) Feedback prompt template: Generate targeted correction instructions based on the verification results, including: feedback on missing fields, logical contradictions or insufficient confidence;

[0050] (3) Implement three-level quality verification, including: format verification, confidence verification and user-defined verification; if the three-level quality verification fails, return to step S5 and continue the generation-evaluation iteration mechanism process until the predefined maximum iteration limit is reached. At this time, the service is marked as unintegrable; if the three-level quality verification passes, proceed to step S6.

[0051] If the service call described in step S6 fails, the following steps will be executed automatically:

[0052] (1) Mark the current service as pending verification and record the reason for failure;

[0053] (2) Trigger alternative services to re-execute the multi-level knowledge collection and generation-evaluation iteration mechanism to optimize the service document process until all candidate services are traversed or the maximum number of iterations is reached.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] 1. This invention can break through the dependence of traditional chatbots on pre-integrated service libraries. By designing a self-supervised method for service retrieval, information extraction and document enhancement, it can achieve autonomous discovery, understanding and invocation of unintegrated services, which significantly reduces the cost of manual intervention.

[0056] 2. This invention can accurately understand user intent and match appropriate services. Based on multi-level knowledge acquisition and context association algorithms, it enhances the dynamic integration of unintegrated services and improves service call coverage to over 75%.

[0057] 3. This invention can identify potential defects in service documents, including unparseable format problems, lack of completeness and consistency problems, redundancy or ambiguity problems, and errors or misleading information, and automatically generate and optimize the documents, increasing the service call success rate to over 65%.

[0058] 4. This invention enables interactive control of the entire chatbot process. The service management module can manage and filter services, ensuring the availability and timeliness of the interface, allowing human intervention at key decision points, and avoiding the risk of blind execution of purely automated solutions. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating the process of a chatbot application using a self-supervised method to call unintegrated services, as described in this invention.

[0060] Figure 2 This is a field structure diagram of the service document of the present invention.

[0061] Figure 3 This is a flowchart of the generation-evaluation iterative optimization module of the present invention.

[0062] Figure 4 This is an example of a service document optimized by the generation-evaluation iterative optimization module of the present invention.

[0063] Figure 5 This is an interactive example and system interface of the chatbot of the present invention.

[0064] Figure 6 This is a diagram illustrating the composition of a self-supervised system for enabling chatbots to utilize unintegrated services, as described in this invention.

[0065] Figure 7 This is a block diagram of a self-supervised system module for promoting chatbots to utilize unintegrated services, as described in this invention. Detailed Implementation

[0066] The present invention will be further described in detail below with reference to the accompanying drawings: This embodiment is implemented under the premise of the technical solution of the present invention, and detailed implementation methods are given, but the protection scope of the present invention is not limited to the following embodiments. Specific implementation method one:

[0068] A self-supervised system for facilitating chatbots' utilization of unintegrated services includes: a service repository module, a service management module, a demand matching module, a generation-evaluation iterative optimization module, and an interaction control module. The service repository module, acting as the underlying database, is managed by the service management module and provides high-quality input data for the entire system and the self-supervised method. The interaction control module is responsible for the dialogue between the user and the chatbot, as well as the condition control within the self-supervised method. When conditions are met, it triggers the demand matching module, which then generates high-quality service documents through the generation-evaluation iterative optimization module to help the chatbot utilize unintegrated services. These service documents are then stored in the service repository module. The system module diagram is shown below. Figure 7 As shown, the signal input terminals of the service repository module are connected to the signal output terminals of the service management module, the demand matching module, and the generation-evaluation-iteration optimization module. The signal output terminals of the service repository module are connected to the signal input terminals of the service management module and the demand matching module. The signal input / output terminals of the demand matching module are connected to the signal input / output terminals of the generation-evaluation-iteration optimization module. The signal output terminal of the interaction control module is connected to the signal input terminal of the demand matching module. The signal input / output terminals of the generation-evaluation-iteration optimization module are connected to the signal input / output terminals of the interaction control module.

[0069] like Figure 1 As shown, a self-supervised method for facilitating chatbots to utilize unintegrated services includes the following steps:

[0070] Step S1: User inputs requirements. The interaction control module receives the user's input requirement description and search preference parameters that the user can adjust through the interaction control module. The semantic parsing unit in the requirement matching module embeds the user's requirement into a semantically informational vector. It then checks if any service interface in the integrated service library matches the current intent, i.e., the user's input requirement. The degree of requirement matching determines whether to call an unintegrated service. The degree of requirement matching is calculated by the cosine similarity between the user's requirement vector q and the integrated service interface description text vector t. The calculation formula is:

[0071]

[0072] A match is considered complete if the degree of matching exceeds the search preference parameter. The default search preference parameter is 0.5, which users can adjust to make the chatbot more likely to select either unintegrated or integrated services. The higher the search preference parameter, the higher the matching requirement, and the more likely it is to select unintegrated services. When an unintegrated service needs to be called, proceed to step S2 for multi-dimensional search; if an unintegrated service does not need to be called, proceed to the final service execution step S7.

[0073] Step S2: A self-supervised system that facilitates chatbots to utilize unintegrated services retrieves a set of relevant services from the unintegrated service library based on the user's request submitted in step S1, using multi-dimensional similarity matching. Specifically, the multi-dimensional matching unit uses multiple similarity calculation methods to comprehensively calculate the service similarity function, and the service retrieval unit selects the top-K closest services and sorts them, then selects the most relevant services in order to execute subsequent steps.

[0074] The service similarity function is defined as follows:

[0075]

[0076] The service similarity function measures the degree of matching between user query q and service document d from three dimensions, including cosine similarity. The calculation will assess semantic similarity and keyword overlap. Jaccard similarity is used to calculate word-level similarity and contextual relevance. Co-occurrence of keywords in the current dialogue history is statistically analyzed, and the similarity of the context is calculated.

[0077] Step S3: Based on the service set retrieved in S2, perform multi-level knowledge collection on the most relevant services. Multi-level knowledge collection includes: content knowledge, style knowledge, and logical knowledge, specifically including:

[0078] Content knowledge: Service entities usually come with documentation or comments containing basic descriptive data about the service, such as: interface function descriptions, parameter definitions, etc.

[0079] Style knowledge: Aggregate analysis on files from the same source or service groups created by the same developer to which the current search service belongs. Homogeneous services include consistent features such as naming conventions, comment styles, and implementation patterns.

[0080] Logical knowledge: By tracing the call chain of the current service and its related services, a service role profile is constructed, including: the chain of front-end services and subsequent services.

[0081] Step S4: Perform structured parsing, extract information, extract key information, and initially generate standardized service documents. The service document field diagram is shown below. Figure 2 As shown, this includes information at three different levels: service, interface, and parameters. The specific steps are as follows:

[0082] The service specifications are parsed in layers, the marked content is decomposed and extracted, illegal characters that cannot be processed by automated tools are removed by manual intervention, and the core elements of the service are extracted by an automated parser, including: service description, operation description, operation type, parameter name, and parameter type.

[0083] The service specifications of different versions and types are classified and standardized, and the referenced parts are embedded into the main content. According to the service type with different formats, a multi-format conversion module is built to convert unstructured documents into JSON intermediate representation. A cleanup process is performed to remove HTML tags, special symbols, escape characters, etc. to reduce noise.

[0084] Step S5: Based on the relevant service set retrieved in S2, the knowledge collected in S3, and the information parsed in S4, select the unintegrated service that best matches the user's needs, and optimize the service document through a generation-evaluation iteration mechanism. The specific steps are as follows:

[0085] (1) In the generation stage, a lightweight approach is adopted, based on prompting engineering and mind chain technology, to guide the chatbot to generate enhanced documents. The prompts include two parts:

[0086] System prompt: When executing a query, a service document needs to be generated based on the service information so that external services can be called to complete the user's query task. This includes definitions of chatbot format specifications, parameter specifications, special instructions, length limits, and iterative optimization.

[0087] Feedback prompts: Based on the quality issues identified in the assessment, prompts will be made to correct and rewrite the service documentation for the chatbot, including feedback on file format, field parsing, duplicates, missing items, errors, insufficient confidence, and user-defined conditions.

[0088] (2) During the evaluation phase, a three-level quality verification is implemented, including format verification, confidence verification, and user-defined verification. If the quality verification fails, the process reverts to step S5 and continues to generate and evaluate the iteration process until the predefined maximum iteration limit is reached. At this point, the service is marked as "unable to integrate." If the quality verification passes, the process proceeds to step S6, where the three-level quality verification includes:

[0089] Format validation: The file must be in JSON format. This step will parse and check the service document to identify issues such as non-standard formatting, missing fields, or duplicate fields.

[0090] Confidence verification: Taking into account the fluency, certainty, and logical rationality of the generated content, confidence quality verification is carried out on the spot based on features such as the negative logarithm generated during the chatbot generation process.

[0091] The confidence verification is based on the negative log-likelihood value of the generative model and the logical reasonableness score.

[0092] User-defined verification: Users can add customized evaluation conditions according to their specific needs to guide the chatbot in generating service documentation;

[0093] The interactive control module in S5 allows users to add custom conditions.

[0094] Step S6: Call the service interface based on the enhanced document. If the call is successful, inject the integrated service library for easy reuse in the next design. For open source chatbots, the parameters can be further fine-tuned on this high-quality data using a completion-based method, thereby improving their ability to understand and generate service documents. If the call fails, return to step S3 and try other candidate services until all retrieved service sets are traversed.

[0095] Step S7: Execute the integrated service that matches the user's needs and return it to the user. The interaction control module will collect user feedback for future improvements.

[0096] The service management module ensures the quality of service documents integrated into the service library in steps S1 and S6 of the self-supervised method, and is responsible for the format conversion required for workflow input and output.

[0097] Steps S1 to S7 above realize the entire process of enhancing the chatbot to use unintegrated services. The component resources required by this method include chatbot, unintegrated service library, prompt project template library, initial condition setting and integrated service library.

[0098] To facilitate efficient conversations using chatbots and provide timely feedback on service call results, an interactive chatbot system is presented. The system composition diagram is shown below. Figure 6As shown, the system includes: a service repository, a service management module, a requirement matching module, a generation-evaluation iterative optimization module, and an interaction control module. The composition and function of each part are described below:

[0099] The service repository, comprising both unintegrated and integrated service repositories, is used to match user requests to the chatbot and provide ready-to-use service support. The unintegrated service repository is a heterogeneous collection of relevant open services obtained from extensive web searches. The integrated service repository includes basic service content and service documentation, which conforms to the standard JSON document format generated by a self-supervised method. As the chatbot successfully calls unintegrated services, these services will be continuously added to the integrated service repository as a source for subsequent knowledge acquisition.

[0100] The service management module is responsible for maintaining, optimizing, and dynamically expanding the service library, ensuring the system can efficiently retrieve, filter, and update service resources, and providing complete service lifecycle management capabilities. This module not only needs to support basic service filtering, addition, deletion, and modification functions, but also needs to incorporate a self-supervised learning mechanism to dynamically adapt to the discovery and integration process of unintegrated services. The following are the core functions this module should possess: Service addition: Supports both manual entry and automatic discovery. For unintegrated services, the system can crawl open API documentation, parse unstructured descriptions provided by developers, or dynamically generate service metadata based on user interaction, and initially store it in the service library to be verified. Service deletion: When a service expires, fails, or does not meet quality standards, the system should support soft or hard deletion and provide exception protection to prevent the failure of dialogue processes dependent on the service due to service removal. Service modification: Allows manual updates to interface parameters, optimization of function descriptions, or adjustment of call constraints. Modified services must re-enter the verification process to ensure consistency between documentation and interfaces. The filtering includes service type, service version, and service time. Service type includes general category services and domain category services. Service version can distinguish between new and old interfaces or different iteration stages. Service time is used to filter out expired or soon-to-be-updated content. These conditions can control the content of the service library and optimize the service recommendation results, ensuring that the returned results not only meet the current needs but also have high availability.

[0101] The demand matching module is responsible for understanding user intent and accurately associating it with available services. This module works in collaboration with the semantic parsing unit and the service retrieval unit to ensure the system can flexibly respond to service demands in different scenarios and has the ability to continuously learn new services. The semantic parsing unit can identify implicit semantic relationships by combining current dialogue history, user preferences, and domain knowledge; the service retrieval unit will search for a set of candidate services in the unintegrated service library based on intent similarity, and select the closest service for knowledge acquisition according to real-time demand, triggering the invocation process of unintegrated services, thus achieving precise location and on-demand activation of service resources.

[0102] The generation-evaluation-iterative optimization module includes a document generation unit, a multi-level verification unit, and a feedback iteration unit, such as... Figure 3 The diagram illustrates the flowchart of the generation-evaluation-iterative optimization module. The document generation unit converts unstructured service descriptions into standardized interface documents based on chained reasoning technology. The multi-level verification unit verifies document compliance through a triple mechanism of rule constraints, logical checks, and manual intervention. The feedback-iterative unit dynamically optimizes the generation strategy based on the verification results to ensure reliable access to unintegrated services and improve performance. Figure 4 The image shows a service document instance optimized by the generation-evaluation iterative optimization module.

[0103] The interaction control module includes the chatbot's question-and-answer interaction and condition control. The question-and-answer interaction involves ending the process if the user is satisfied and re-entering the method if they are not. It analyzes user feedback to understand user needs and matches relevant services. Condition control includes managing the number of matched services, matching accuracy, and custom conditions for service document verification. For example... Figure 5 The image shows an interactive example and system interface of a chatbot. The interface of the interactive control module includes:

[0104] (1) The central message window adopts a mixed layout of bubbles and cards to alternately present text messages and service result cards;

[0105] (2) Sidebar, used to support conversation history management, providing pin, rename and delete operations;

[0106] (3) Navigation bar, used to support switching between interfaces and functions, including: Service Management, Document Optimization, Condition Configuration and Chatbot pages;

[0107] (4) Multimodal model, used to retain the input interface of the multimodal chatbot, allowing the upload of files or pictures.

[0108] The left side displays collapsed chat windows, where users can select different chatbots. Historical chat windows are stored as dialog boxes, which users can pin, rename, or delete. The right side is a navigation bar for switching pages to view other functions. Once in a chat, the user sends a request, and the system executes the aforementioned modules in the background. The central message window uses a hybrid "bubble + card" format, alternating between text messages and service result cards to demonstrate the application of backend methods. For multimodal models, uploading files or images is allowed to facilitate user interaction with the chatbot. When the user ends the chat, the system determines whether the issue is "resolved" or "unresolved."

[0109] The above description is merely a preferred embodiment of the present invention. These specific embodiments are different implementations based on the overall concept of the present invention, and the scope of protection of the present invention 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 the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A self-monitoring system for facilitating chatbots to utilize unintegrated services, characterized in that, The system includes: a service repository module, a service management module, a requirement matching module, a generation-evaluation iterative optimization module, and an interaction control module; The signal input terminals of the service repository module are connected to the signal output terminals of the service management module, the requirement matching module, and the generation-evaluation-iteration optimization module. The signal output terminals of the service repository module are connected to the signal input terminals of the service management module and the requirement matching module. The signal input and output terminals of the requirement matching module are connected to the signal input and output terminals of the generation-evaluation-iteration optimization module. The signal output terminal of the interaction control module is connected to the signal input terminal of the requirement matching module. The signal input and output terminals of the generation-evaluation-iteration optimization module are connected to the signal input and output terminals of the interaction control module.

2. The self-monitoring system for promoting chatbot utilization of unintegrated services according to claim 1, characterized in that, The service repository module is used to match user requests to the chatbot and provide ready-to-use service support; It receives management signals from the service management module and provides service content to the service management module; it also receives storage signals and service content from the generation-evaluation-iteration optimization module. Receive interaction signals from the demand matching module and provide service content to the demand matching module; The service management module supports lifecycle management of services already stored in the service repository module, supporting manual entry, automatic discovery, and soft deletion mechanisms; it supports version control interfaces to distinguish between new and old services and mark expired content; and it supports an exception protection mechanism to prevent service removal from causing interruptions to dependency processes. It sends management signals to the service repository module and accepts service content provided by the service repository module; The demand matching module is used to understand user intent and accurately associate available services; it sends retrieval signals to the service repository module and receives service content from the service repository module; it receives demand signals from the interaction module; it sends service content and driving signals to the generation-evaluation iterative optimization module and receives feedback signals from the generation-evaluation module. The generation-evaluation-iteration optimization module is used to generate standardized interface service documentation and verify its compliance. It receives service content and driving signals from the demand matching module and sends feedback signals to the demand matching module; it sends storage signals and service content to the service storage module; it receives interaction signals from the interaction control module and sends interaction signals to the interaction control module. The interactive control module is used to receive user feedback through a question-and-answer interface and control the number of matching services, verification conditions, and process jump logic; it sends interactive signals to the demand matching module; it sends interactive signals to the generation-evaluation iterative optimization module and receives interactive signals returned by the generation-evaluation module.

3. A self-monitoring system for promoting chatbot utilization of unintegrated services according to claim 1, characterized in that, The service repository module includes: an unintegrated service repository and an integrated service repository. Through a self-supervised workflow, services in the unintegrated service repository are continuously integrated into the integrated service repository. The unintegrated service repository is used to store relevant unintegrated open services obtained from extensive web searches. An integrated service library is used to store the basic content and service documentation of integrated services.

4. A self-monitoring system for promoting chatbot utilization of unintegrated services according to claim 3, characterized in that, The demand matching module includes: a semantic parsing unit, a multi-dimensional matching unit, and a service retrieval unit; the semantic parsing unit embeds user demands and service documents into semantic vectors; the multi-dimensional matching unit uses the semantic vectors parsed by the semantic parsing unit to calculate multi-dimensional similarity; and the service retrieval unit sorts services based on the similarity obtained by the multi-dimensional matching unit and coordinates the execution of subsequent steps. Semantic parsing unit, used to extract deep user intent; The multi-dimensional matching unit is used to comprehensively measure the degree of matching between user queries and service documents from different dimensions by employing multiple similarity calculation methods. The service retrieval unit is used to retrieve services from the unintegrated service library based on the similarity in the multi-dimensional matching unit and activate the unintegrated service call process.

5. A self-monitoring system for promoting chatbot utilization of unintegrated services according to claim 4, characterized in that, The generation-evaluation iterative optimization module includes: a document generation unit, a multi-level verification unit, and a feedback iterative unit. The document generated by the document generation unit will be quality verified by the multi-level verification unit. If it passes the multi-level verification unit, the process ends; otherwise, feedback is provided through the feedback iterative unit, and the process returns to the document generation unit for generation. The document generation unit is used to convert unstructured service descriptions into standardized interface documents; Multi-level verification units are used to verify document compliance through a triple mechanism of rule constraints, logical verification, and manual intervention; The feedback iteration unit is used to ensure reliable access and performance improvement for unintegrated services.

6. A self-supervised method for promoting chatbots to utilize unintegrated services based on the self-supervised system of claim 5, characterized in that, Includes the following steps: Step S1: Receive user input requirements, extract user input requirements through the semantic parsing unit in the requirement matching module, check whether there is a service interface matching the user input requirements in the integrated service library in the service repository module. If yes, proceed to step S7; otherwise, call the unintegrated service to proceed to step S2. Step S2: Based on the user's input requirements, retrieve a set of services related to the user's input requirements from the unintegrated service library in the service repository module through the multi-dimensional matching unit in the requirements matching module; Step S3: Through the service retrieval unit in the demand matching module, perform multi-level knowledge collection on the most relevant services in the service set retrieved in step S2; Step S4: Through the document generation unit in the generation-evaluation-iteration optimization module, the most relevant service information is structured and parsed to generate standardized service documents; and the compliance of the standardized service documents is verified through the multi-level verification unit in the generation-evaluation-iteration optimization module. Step S5: Based on the retrieval results of step S2, the knowledge collected in step S3, and the parsed information in step S4, the service document is optimized using the generation-evaluation iteration mechanism through the feedback iteration unit in the generation-evaluation iteration optimization module. Step S6: Call the service interface corresponding to the optimized service document through the generation-evaluation iteration mechanism. If the call is successful, add the optimized service document to the integrated service library, fine-tune the chatbot, and then proceed to step S7; if the call fails, return to step S3. Step S7: Execute the integrated service that matches the user's input requirements and return the result to the user.

7. A self-monitoring method for promoting chatbot utilization of unintegrated services according to claim 6, characterized in that, The method for checking whether a service interface matching the user's input requirements exists in the integrated service library of the service repository module in step S1 is as follows: The interactive control module receives the user's input description of their needs, as well as the search preference parameters adjusted by the user through the interactive control module. It determines whether to invoke unintegrated services based on the degree of need matching. The degree of need matching is calculated by the cosine similarity between the user's need vector q and the integrated service interface description text vector t. The calculation formula is as follows: A match is considered to be found when the degree of matching exceeds the search preference parameter. The default search preference parameter is 0.

5. Users can adjust the search preference parameter to make the chatbot select more services that are not integrated or integrated. The higher the search preference parameter, the higher the matching requirement and the more likely it is to select services that are not integrated. The multi-dimensional similarity matching method described in step S2 includes the following steps: Construct a service similarity function based on semantic embedding; and perform weighted ranking by combining the cosine similarity, keyword overlap rate, and contextual relevance between the semantics of user input requirements and service documents in the unintegrated service library. The service similarity function is defined as follows: The service similarity function measures the degree of matching between user query q and service document d from three dimensions, including cosine similarity. To express semantic similarity, keyword overlap rate Jaccard similarity is used to calculate word-level similarity and contextual relevance. Co-occurrence of keywords in the current dialogue history is statistically analyzed, and the similarity of the context is calculated.

8. A self-monitoring method for promoting chatbots to utilize unintegrated services according to claim 6, characterized in that, The multi-level knowledge acquisition mentioned in step S3 includes: content knowledge, style knowledge, and logical knowledge; specifically as follows: (1) Content knowledge: Supplement the missing fields in the service documents of the most relevant services in the service set retrieved in step S2. The content comes from the operation description of the operation with the same name, the parameter description of the parameter with the same name and the calling example in the integrated service library; (2) Style knowledge: Analyze the consistency characteristics of naming conventions, commenting styles and implementation patterns of service groups with the same source. Service groups with the same source are services in the unintegrated service library that have the same source as the service. (3) Logical knowledge: Construct a role profile through the most relevant service call chain, including preconditions and subsequent service dependencies.

9. A self-monitoring method for promoting chatbots to utilize unintegrated services according to claim 6, characterized in that, The specific steps of step S4 are as follows: (1) Perform a cleanup operation on non-standard service documents and use an automated parser to extract the core elements of the service, including: service description, operation description, operation type, parameter name, and parameter type; (2) Construct a multi-format conversion module to split the service document into a three-level field structure of service, interface and parameter, and convert the unstructured document into an intermediate representation of JSON format.

10. A self-monitoring method for promoting chatbot utilization of unintegrated services according to claim 6, characterized in that, The generation-evaluation iterative mechanism described in step S5 includes the following steps: (1) System prompt template: Defines the format specifications, parameter constraints, and iterative optimization rules for generating service documents; (2) Feedback prompt template: Generate targeted correction instructions based on the verification results, including feedback for missing fields, logical contradictions, or insufficient confidence. (3) Implement three-level quality verification, including: format verification, confidence verification and user-defined verification; if the three-level quality verification fails, return to step S5 and continue the generation-evaluation iteration mechanism process until the predefined maximum iteration limit is reached. At this time, the service is marked as unintegrable; if the three-level quality verification passes, proceed to step S6. If the service call described in step S6 fails, the following steps will be executed automatically: (1) Mark the current service as pending verification and record the reason for failure; (2) Trigger alternative services to re-execute the multi-level knowledge collection and generation-evaluation iteration mechanism to optimize the service document process until all candidate services are traversed or the maximum number of iterations is reached.

Images