Question answering method and system for aggregated answering

Abstract

The invention discloses a question-answering method and system for aggregating answers, and the method comprises the steps: receiving interaction content inputted by a user in a question-answering interface in a front-end page, generating a question-answering request according to the interaction content, and transmitting the question-answering request to a rear-end interface service, the server is used for receiving an aggregation answer result returned by the interface service, displaying the aggregation answer result when receiving the aggregation answer result returned by the interface service, executing front-end search on interactive content based on a pre-established local index, and displaying the front-end search result as alternative information and the aggregation answer result to a user together; in the interface service process, a plurality of candidate answers are obtained in at least two modes of keyword matching, retrieval based on a knowledge base and model reasoning conducted through the model service, aggregation processing is conducted on the candidate answers, and an aggregation answer result including main answer information and alternative answer information is generated. According to the invention, the accuracy and stability of the answer content can be improved, and the use experience of the user can be improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent question-answering technology, and more specifically, to a question-answering method and system for aggregating answers. Background Technology

[0002] As intelligent services and online business scenarios become increasingly diverse, users are relying more and more on question-and-answer methods to obtain information, perform operations, or receive business guidance. Currently, most traditional question-and-answer systems depend on a single answer generation method. When the front-end cannot provide auxiliary capabilities and lacks flexible utilization of information from multiple sources, the accuracy of the system's output answers is often limited, failing to maintain consistent answer quality across different user input scenarios. These problems become particularly prominent as business systems expand and user demands increase.

[0003] Therefore, existing technologies urgently need a question-and-answer processing method that can improve the accuracy of answers and the efficiency of interaction, so that the system can provide valuable answers faster and more stably after the user inputs, thereby improving the overall user experience. Summary of the Invention

[0004] In order to solve at least one of the technical problems in the background art described above, the present invention proposes a question-and-answer method and system for aggregated answers.

[0005] One aspect of the present invention provides a question-and-answer method for aggregating answers, the method comprising:

[0006] In the front-end page, a question-and-answer interface is rendered according to the configuration parameters, the interactive content input by the user in the question-and-answer interface is received, a question-and-answer request is generated according to the interactive content, the question-and-answer request is sent to the back-end interface service, and the aggregated answer result is displayed when the aggregated answer result returned by the interface service is received.

[0007] On the front-end page, a front-end search is performed on the interactive content based on a pre-established local index to obtain front-end search results, and the front-end search results are displayed to the user as alternative information along with the aggregated answer results;

[0008] During the data management process, instance data corresponding to question-and-answer instances are managed for different access parties. The instance data includes instance configuration data, knowledge base data, common question corpus, and user question-and-answer records and feedback information. Based on the instance data, knowledge base data and corpus information for question-and-answer processing are provided to the interface service, and corpus information for model reasoning is provided to the model service.

[0009] During the interface service process, the question and answer request is received. Based on the knowledge base data and corpus information, multiple candidate answers are obtained using at least two of the following methods: keyword matching, knowledge base retrieval, and model inference performed through the model service. The multiple candidate answers are aggregated to determine the main answer and alternative answer information for supplementary explanation. The aggregated answer result, including the main answer and the alternative answer information, is generated.

[0010] During the model service process, at least one language model is invoked to perform inference based on the question-and-answer request and corpus information, generating candidate answers for the question-and-answer request and returning them to the interface service.

[0011] Optionally, generating a question-and-answer request based on the interaction content includes:

[0012] The user-input interactive content is type-determined. When the interactive content conforms to a preset instruction format, the interactive content is parsed into a front-end instruction, and the operation corresponding to the front-end instruction is executed locally on the front-end.

[0013] When the interactive content does not conform to the preset instruction format, the question-and-answer request is generated based on the interactive content and sent to the interface service, while a front-end search is performed on the interactive content.

[0014] Optionally, performing a front-end search on the interactive content based on a pre-established local index includes:

[0015] Receive index data sent by the interface service, and construct the local index based on the index data. The index data includes menu items, page links, frequently asked questions entries, and corresponding target addresses or operation identifiers.

[0016] When performing a front-end search, several candidate entries are selected from the local index based on keyword matching and relevance scores, and then the candidate entries are sorted according to the relevance scores and displayed to the user.

[0017] Optionally, the aggregation process for the plurality of candidate answers includes:

[0018] Calculate the matching scores for candidate answers obtained through keyword matching, candidate answers obtained through knowledge base retrieval, and candidate answers obtained through model inference, respectively.

[0019] The main answer is determined based on the preset priority rules and the matching score. Specifically, at least when the matching score of keyword matching exceeds a preset threshold, the candidate answer obtained through keyword matching is determined as the main answer. When the preset threshold is not met, the candidate answer obtained based on model reasoning is determined as the main answer.

[0020] Candidate answers that were not selected as the main answer are added to the aggregated answer results as alternative answer information.

[0021] Optionally, generating a question-and-answer request based on the interaction content and sending the question-and-answer request to the backend interface service includes:

[0022] When generating the question-and-answer request on the front end, an access party identifier is carried to identify the access party;

[0023] During the interface service process, a target question-answer instance is selected from multiple pre-maintained question-answer instances based on the access party identifier, and candidate answer acquisition and the generation of the aggregated answer result are performed based on the instance configuration data, knowledge base data and corpus information associated with the target question-answer instance.

[0024] Optionally, model inference can be performed via model services, including:

[0025] During the model service process, multiple model instances are maintained, and each model instance is used to implement a language model and / or a model version.

[0026] During the interface service process, the target model instance is determined based on the model identifier recorded in the instance configuration data, and the target model instance is invoked to perform model inference on the question and answer request to generate candidate answers for the corresponding access party question and answer instance.

[0027] Optionally, the question-and-answer method for aggregating answers further includes:

[0028] During the model service process, upon receiving a model update instruction based on the training corpus, the target model instance is trained based on the training corpus to obtain the updated model instance.

[0029] Assign an independent service port to the updated model instance and complete a health check, and notify the interface service to switch the call to the target model instance to the updated model instance;

[0030] After confirming the call switch is completed, shut down the service port of the original target model instance to complete the model update without interrupting the external question and answer service.

[0031] In another aspect, the present invention provides a question-and-answer system for aggregating answers, the system comprising: a front-end interaction module disposed on the front end, and a data management module, an interface service module, and a model service module disposed on the back end;

[0032] The front-end interaction module is used to be embedded into the front-end page of the access party, render the question and answer interface according to the configuration parameters, receive the interactive content input by the user in the question and answer interface, generate a question and answer request according to the interactive content and send it to the interface service module, and display the aggregated answer result when the interface service module returns the aggregated answer result. At the same time, it performs a front-end search on the interactive content based on a pre-established local index and displays the search results obtained from the front-end search as alternative information to the user.

[0033] The data management module is used to manage instance data of corresponding question-and-answer instances for different access parties. The instance data includes instance configuration data, knowledge base data, common question corpus, user question-and-answer records and feedback information. Based on the instance data, the module provides the interface service module with knowledge base data and corpus information for question-and-answer processing, and provides the model service module with corpus information for model inference.

[0034] The interface service module is used to receive the question-and-answer request sent by the front-end interaction module, and based on the knowledge base data and corpus information provided by the data management module, obtain multiple candidate answers using at least two of the following methods: keyword matching, knowledge base retrieval, and model inference performed by the model service module. The multiple candidate answers are aggregated to determine the main answer and alternative answer information for supplementary explanation, generate an aggregated answer result including the main answer and the alternative answer information, and return the aggregated answer result to the front-end interaction module.

[0035] The model service module is used to invoke at least one language model to perform inference based on the question-and-answer request and corpus information forwarded by the interface service module, generate candidate answers for the question-and-answer request, and return them to the interface service module.

[0036] Optionally, the front-end interaction module is further configured to determine the type of the interaction content; when the interaction content conforms to a preset instruction format, the interaction content is parsed into a front-end instruction and the operation corresponding to the front-end instruction is executed locally; when the interaction content does not conform to the preset instruction format, a question-and-answer request is generated based on the interaction content and sent to the interface service module, while a front-end search is performed on the interaction content.

[0037] Optionally, the front-end interaction module is configured to support at least two working modes: in the first working mode, a front-end search is performed on the interactive content based on the local index and the search results are displayed without sending the question-and-answer request to the interface service module; in the second working mode, the question-and-answer request is sent to the interface service module and the aggregated answer results are received, while a front-end search is performed and the front-end search results are displayed as supplementary information to the aggregated answer results.

[0038] Optionally, the local index in the front-end interaction module is constructed from the index data issued by the interface service module. The index data includes menu items, page links, frequently asked questions, and corresponding target addresses or operation identifiers. When performing a front-end search, the front-end interaction module selects several candidate entries from the local index based on keyword matching and relevance scores, sorts the candidate entries according to the relevance scores, and displays them to the user.

[0039] Optionally, the interface service module is specifically used to calculate the corresponding matching scores for candidate answers obtained through keyword matching, candidate answers obtained through knowledge base retrieval, and candidate answers obtained through model inference when aggregating multiple candidate answers; determine the main answer based on preset priority rules and the matching scores, wherein at least when the matching score of keyword matching exceeds a preset threshold, the candidate answer obtained through keyword matching is determined as the main answer, and when the preset threshold is not met, the candidate answer obtained through model inference is determined as the main answer; and add the candidate answers that are not selected as the main answer as the alternative answer information to the aggregated answer result.

[0040] Optionally, the interface service module maintains multiple question-and-answer instances for providing question-and-answer services to different access parties; the front-end interaction module carries an access party identifier to identify the access party when generating the question-and-answer request; the interface service module selects a target question-and-answer instance based on the access party identifier, and performs candidate answer acquisition and aggregation processing based on the instance configuration data, knowledge base data and corpus information associated with the target question-and-answer instance.

[0041] Optionally, the model service module maintains multiple model instances, each model instance being used to implement a language model and / or a model version; the interface service module determines the target model instance based on the model identifier recorded in the instance configuration data, and calls the target model instance to perform model inference on the question-and-answer request, generating candidate answers for the corresponding access party question-and-answer instance.

[0042] Optionally, the model service module is further configured to, upon receiving a model update instruction based on a training corpus, train the target model instance based on the training corpus to obtain an updated model instance; after allocating an independent service port to the updated model instance and completing a health check, notify the interface service module to switch the call to the target model instance to the updated model instance; and after confirming the call switch is completed, close the service port of the original target model instance to complete the model update without interrupting the external question-and-answer service.

[0043] To achieve the above objectives, according to another aspect of the present invention, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described question-and-answer method for aggregated answers.

[0044] To achieve the above objectives, according to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program / instructions are stored, which, when executed by a processor, implement the steps of the question-and-answer method for aggregated answers described above.

[0045] To achieve the above objectives, according to another aspect of the present invention, a computer program product is also provided, comprising a computer program / instructions that, when executed by a processor, implement the steps of the above-described question-and-answer method for aggregated answers.

[0046] The beneficial effects of this invention are as follows:

[0047] This invention enables the system to simultaneously obtain real-time candidate information from the front end and a higher-quality aggregated answer from the back end after user input by performing a fast search based on a local index on the front end and combining multiple answer generation methods in the back end for aggregation. This maintains the smoothness of front-end interaction while improving the accuracy and stability of answer content, thereby addressing the quality deficiencies caused by the single source of answers in traditional question-answering systems. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0049] Figure 1 This is a flowchart of the question-and-answer method for aggregating answers according to an embodiment of the present invention;

[0050] Figure 2This is a schematic diagram of a question-and-answer system that aggregates answers according to an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the user question-and-answer process according to an embodiment of the present invention;

[0052] Figure 4 This is a schematic diagram of the data flow process according to an embodiment of the present invention;

[0053] Figure 5 This is a schematic diagram of the system architecture of an embodiment of the present invention;

[0054] Figure 6 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0055] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0056] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0058] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0059] It should be noted that the information collected in the technical solution of this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0060] It should be noted that the technical solution of this application provides users with corresponding operation entry points for users to choose to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0061] It should be noted that the question-and-answer method and system for aggregated answers of the present invention can be used in the financial field, or in any field other than the financial field. The application field of the question-and-answer method and device for aggregated answers of the present invention is not limited.

[0062] This invention proposes a question-and-answer method and system that aggregates answers. By comprehensively utilizing various suitable answer generation methods such as front-end search, back-end retrieval, and model inference in the question-and-answer process, and supporting the display of multiple data formats, the system can provide users with more comprehensive, accurate, and easy-to-understand answer content. Simultaneously, this invention encapsulates and parameterizes the front-end components and front-end / back-end configuration methods, enabling the question-and-answer system to be quickly integrated into different websites or business systems via embedding, and allowing for flexible and personalized configuration according to the needs of the access parties, thus facilitating deployment and maintenance.

[0063] Figure 2 This is a schematic diagram of a question-and-answer system that aggregates answers according to an embodiment of the present invention, such as... Figure 2 As shown, in one embodiment of the present invention, the question-and-answer system for aggregated answers of the present invention includes: a front-end interaction module set at the front end, and a data management module, an interface service module, and a model service module set at the back end;

[0064] The front-end interaction module is used to be embedded into the front-end page of the access party, render the question and answer interface according to the configuration parameters, receive the interactive content input by the user in the question and answer interface, generate a question and answer request according to the interactive content and send it to the interface service module, and display the aggregated answer result when the interface service module returns the aggregated answer result. At the same time, it performs a front-end search on the interactive content based on a pre-established local index and displays the search results obtained from the front-end search as alternative information to the user.

[0065] The data management module is used to manage instance data of corresponding question-and-answer instances for different access parties. The instance data includes instance configuration data, knowledge base data, common question corpus, user question-and-answer records and feedback information. Based on the instance data, the module provides the interface service module with knowledge base data and corpus information for question-and-answer processing, and provides the model service module with corpus information for model inference.

[0066] The interface service module is used to receive the question-and-answer request sent by the front-end interaction module, and based on the knowledge base data and corpus information provided by the data management module, obtain multiple candidate answers using at least two of the following methods: keyword matching, knowledge base retrieval, and model inference performed by the model service module. The multiple candidate answers are aggregated to determine the main answer and alternative answer information for supplementary explanation, generate an aggregated answer result including the main answer and the alternative answer information, and return the aggregated answer result to the front-end interaction module.

[0067] The model service module is used to invoke at least one language model to perform inference based on the question-and-answer request and corpus information forwarded by the interface service module, generate candidate answers for the question-and-answer request, and return them to the interface service module.

[0068] In one embodiment of the present invention, the front-end interaction module can be integrated by embedding script code or front-end components in the access party's page. It appears as a floating entry icon or a floating button on the page, which, when triggered by the user, pops up a dialog window to display the question-and-answer interface. The front-end interaction module can dynamically adjust the layout and display format of the question-and-answer interface according to the theme style, display text, and interaction parameters configured by the access party. When receiving user input, it can simultaneously collect the context information or location identifier of the current page, so that the backend can respond to the question-and-answer request based on the page context. When performing a front-end search, the front-end interaction module can perform keyword matching on the user input based on pre-cached links, menu items, or frequently asked questions in the local index, and sort them by relevance. It can then display the most relevant entries as candidate information in the dialog window, allowing the user to view the aggregated answers returned by the backend while simultaneously using the front-end search results for quick navigation or further operations.

[0069] In one embodiment of the present invention, the data management module can maintain an independent question-and-answer instance space for each access party. Through a configuration interface, access party administrators can personalize configurations such as instance name, topic information, display text, and model selection strategies, and save the configuration results as instance configuration data. The data management module can provide knowledge base maintenance functions for inputting and editing structured business knowledge, frequently asked questions, and corresponding answers, and supports categorized management by topic, business line, or tag. Furthermore, the data management module can record users' question-and-answer history, including question content, system answers, user evaluations, and manual feedback, allowing access parties to conduct operational analysis of question-and-answer effectiveness. The module also provides the processed and labeled corpus information to the interface service module and model service module for optimizing subsequent question-and-answer processing and model inference.

[0070] In one embodiment of the present invention, after receiving a question-and-answer request from the front-end interaction module, the interface service module can first determine the question-and-answer instance corresponding to the current request based on the access party identifier and instance configuration data. If the permission verification is successful, it can then obtain the knowledge base data and corpus information corresponding to that instance from the data management module. The interface service module can perform keyword matching based on the user input content in the request to quickly find entries highly relevant to the question. Simultaneously, it can perform a search on the knowledge base to obtain candidate answers that are semantically similar to the question or business-related. The question-and-answer request and corpus information are then forwarded to the model service module, where the language model generates one or more semantically relevant candidate answers. Afterward, the interface service module can score and rank the candidate answers from different sources according to a preset strategy, selecting the most relevant and clearly expressed answer as the main answer. The remaining candidate answers with reference value are compiled into alternative answer information, and all are packaged into an aggregated answer result and returned to the front-end interaction module for comprehensive display.

[0071] In one embodiment of the present invention, the model service module can preload one or more language models and configure corresponding model instances for different access parties or different business scenarios. During operation, the model service module receives question-and-answer requests and corpus information forwarded from the interface service module, organizes user input, context information, and related corpus into an input format acceptable to the model, calls the target language model to perform inference operations, and obtains text results or structured results representing the answer content. The model service module can perform basic formatting processing on the model output, such as processing paragraph structure, list structure, or highlighting information, and returns the processed candidate answers to the interface service module for subsequent aggregation processing. By centralizing the model inference capabilities in the model service module, it is convenient to manage and extend the model uniformly, and to upgrade and replace the model capabilities without affecting the external interface of the interface service module.

[0072] In one embodiment of the present invention, the front-end interaction module serves as the entry point for users to interact with the question-and-answer system and is embedded into the website page of the access party. The access party only needs to add the corresponding script file to the page to display a floating entry icon on the front end. Users can click on this icon to open a dialogue window to interact with the question-and-answer robot. This module is responsible for receiving user-inputted interactive content and performing instruction judgment locally to identify whether the user input belongs to a preset front-end instruction. When it is determined to be a front-end instruction, the corresponding operation can be executed directly on the front end; when it is determined to be non-instruction content, a question-and-answer request is generated and sent to the interface service module.

[0073] After receiving the formatted answer from the API service module, the front-end interaction module can perform a front-end search based on a preset local index, retrieving links, menu items, or other front-end displayable data content related to the user's input. These search results are then displayed along with the back-end answer as alternative information. To improve display quality, the front-end interaction module supports parsing the formatted answer, constructing it into HTML content, and providing various interaction methods such as click-to-jump, query triggering, and chart display.

[0074] Furthermore, the front-end interaction module reserves multiple configuration parameters and extension methods, allowing users to customize interface styles, interactive behaviors, search logic, and display modes according to their own business needs, thus achieving flexible and personalized configuration. This module is independent of the back-end service and can run independently when only front-end search is used, or it can work in conjunction with the back-end question-and-answer service to meet the usage requirements of different scenarios.

[0075] In one embodiment of the present invention, the interface service module, as the core data processing module of the question-answering system, provides a unified question-answering service interface for the front-end interaction module. This module can obtain instance configuration data, knowledge base data, and corpus information corresponding to the access party from the data management module, and parse, index, and verify the relevant data according to actual needs to generate client configuration, keyword index, and corpus index required by the front-end. Simultaneously, the interface service module can also send model configuration and corpus data to the model service module to support model training, updates, and management.

[0076] During the question-and-answer process, the interface service module receives question-and-answer requests from the front-end interaction module and executes various answer generation methods based on the acquired data. Specifically, the interface service module can obtain multiple candidate answers using keyword matching, model reasoning, and knowledge base retrieval. Keyword matching is used to quickly identify user intent, model reasoning is used to generate semantic-level answer content, and knowledge base retrieval is used to supplement structured or recorded information. When model reasoning cannot provide a valid answer, the interface service module can select the highest-scoring result from the retrieved candidate information as the answer. Finally, the interface service module aggregates the candidate answers from the above multiple sources, determines the main answer and alternative answer information, and returns the aggregated answer result to the front-end interaction module. Simultaneously, the interface service module provides necessary service interfaces to the front-end and data management modules to support the operation and maintenance of the overall question-and-answer system.

[0077] In one embodiment of the present invention, a data management module is used to centrally manage question-and-answer instances corresponding to different access parties and provide various operational and configuration functions related to the question-and-answer service. This module can obtain operational data from the interface service module and classify, store, and maintain instance data based on the business needs of the access parties. The data management module typically includes operational data display, instance configuration management, FAQ and event management, and question-and-answer record management.

[0078] The operational data display presents the overall operational status of the Q&A instance in charts, including question and answer volume, response rate, trending questions, and user feedback, allowing the access party to understand the robot's effectiveness in a timely manner. Instance configuration management, performed by the access party's website administrator, is used to personalize the Q&A instance's topic data, model pipeline, and other business parameters, enabling the same system to adapt to the business scenarios of different access parties. FAQ and issue management support the joint maintenance of frequently asked questions and related issues by administrators, business specialists, and ordinary users, ensuring content quality and consistency through an approval mechanism. Q&A record management displays users' question and answer history, allowing access party administrators to provide manual responses or further processing as needed.

[0079] Through the above functions, the data management module can effectively manage instance data, knowledge content and user feedback related to the question-and-answer service, and provide the interface service module and model service module with the necessary configuration data and corpus information, thereby supporting the continuous operation and optimization of the entire question-and-answer system.

[0080] In one embodiment of the present invention, the model service module manages multiple language models and provides a unified model inference capability to the interface service module. This module can maintain multiple model instances, each corresponding to a different type of language model or a different version of the model configuration, to adapt to the question-answering quality requirements of different access parties or different business scenarios. The model service module is responsible for executing the model inference process, receiving question-answering requests and corpus information forwarded by the interface service module, generating candidate answers using the corresponding model instances, and returning the inference results to the interface service module.

[0081] To support continuous model capability updates, the model service module provides a seamless model update mechanism. This module receives model configurations or data from the interface service module, internally converts them to generate model data files suitable for model execution, and switches service access to the updated model instance after a new model instance is deployed. After a successful switch, the model service module can take the original model instance offline, thus completing the model update without interrupting the question-and-answer service. Furthermore, the model service module uniformly encapsulates model access methods, ensuring that access logic for different models is provided externally through a consistent interface. The module internally manages the specific port or access method of the model instance, simplifying the API service module's call process.

[0082] Since the model service module undertakes computationally intensive tasks such as model running, model loading, and inference, its deployment and restart costs are relatively high. Therefore, deploying it independently from the interface service module can improve the overall system stability and maintenance efficiency.

[0083] In one embodiment of the present invention, both the interface service module and the model service module support horizontal and vertical scalability. Each module can maintain multiple independent instances, with different instances corresponding to different access parties or different business configurations, to achieve isolated and customized deployment of the question-and-answer service. In actual operation, the system can select the corresponding interface service module instance based on the access party's identifier, providing an independent question-and-answer processing flow for that access party; simultaneously, the interface service module can also select and call a matching model instance from the model service module based on instance configuration data to ensure the correspondence between model inference and question-and-answer instances.

[0084] Furthermore, both the interface service module and the model service module support the deployment of multiple service nodes. Accessing parties can specify the robot instance to access within the target interface service module, while the interface service module can further specify the model instance to access within the target model service module. Through this design, the system can maintain module independence while achieving flexible scalability, thereby adapting to the question-and-answer service needs of multiple accessing parties and multiple business scenarios, and improving overall maintainability and scalability.

[0085] In one embodiment of the present invention, the front-end interaction module is further configured to determine the type of the interaction content; when the interaction content conforms to a preset instruction format, the interaction content is parsed into a front-end instruction and the operation corresponding to the front-end instruction is executed locally; when the interaction content does not conform to the preset instruction format, a question-and-answer request is generated based on the interaction content and sent to the interface service module, while a front-end search is performed on the interaction content.

[0086] In one embodiment of the present invention, the front-end interaction module is configured to support at least two working modes: in the first working mode, a front-end search is performed on the interactive content based on the local index and the search results are displayed without sending the question-and-answer request to the interface service module; in the second working mode, the question-and-answer request is sent to the interface service module and the aggregated answer results are received, while a front-end search is performed and the front-end search results are displayed as supplementary information to the aggregated answer results.

[0087] In one embodiment of the present invention, the local index in the front-end interaction module is constructed from the index data issued by the interface service module. The index data includes menu items, page links, frequently asked questions entries, and corresponding target addresses or operation identifiers. When performing a front-end search, the front-end interaction module selects several candidate entries from the local index based on keyword matching and relevance scores, sorts the candidate entries according to the relevance scores, and displays them to the user.

[0088] In one embodiment of the present invention, the front-end interaction module can construct a local index for local search based on the index data issued by the interface service module. Specifically, the interface service module can issue index data periodically or when the instance configuration is updated. The index data can adopt a structured data format, such as JSON or other suitable key-value structures. The index content includes menu items, page links, frequently asked questions (FAQ) entries, and the target addresses or operation identifiers corresponding to these entries in the access system. After receiving the index data, the front-end interaction module can parse it and store it in a local memory data structure, such as an array, a key-value table, or a tag-based index mapping, to facilitate subsequent fast searches.

[0089] After a user inputs interactive content on the Q&A interface, the front-end interaction module can initiate the front-end search process. Specifically, the module can extract keywords from the user's input and filter relevant entries in the local index based on keyword matching, such as matching menu names, page titles, question titles, or description fields. After identifying relevant candidate entries, the front-end interaction module can also calculate the relevance score for each candidate entry based on preset relevance scoring rules, such as keyword hit location, hit frequency, or field weight. Subsequently, the module sorts the candidate entries according to their relevance scores and displays the sorted candidate entries as the front-end search results to the user. In this way, the front-end interaction module can provide users with prompting local search results in a timely manner before the back-end returns an answer, thereby improving the interaction efficiency and user experience of the Q&A system.

[0090] In one embodiment of the present invention, the interface service module is specifically used to calculate the corresponding matching scores for candidate answers obtained by keyword matching, candidate answers obtained by knowledge base retrieval, and candidate answers obtained by model inference when aggregating multiple candidate answers; determine the main answer based on preset priority rules and the matching scores, wherein, at least when the matching score of keyword matching exceeds a preset threshold, the candidate answer obtained by keyword matching is determined as the main answer, and when the preset threshold is not met, the candidate answer obtained by model inference is determined as the main answer; and add the candidate answers that are not selected as the main answer as the alternative answer information to the aggregated answer result.

[0091] In one embodiment of the present invention, the interface service module can aggregate candidate answers from different answer sources to determine the main answer and alternative answer information to be returned to the user. Specifically, the interface service module can obtain corresponding candidate answers from the keyword matching module, the knowledge base-based retrieval module, and the model service module, and calculate a corresponding matching score for each type of candidate answer. The matching score can be calculated based on keyword hit rate, relevance score of retrieval results, or confidence information contained in model inference results, to reflect the degree of matching between the answer and the user input.

[0092] After obtaining matching scores from multiple candidate answers, the interface service module can determine the primary answer based on preset priority rules. For example, if the matching score of a candidate answer obtained through keyword matching exceeds a preset threshold, the interface service module can directly determine that candidate answer as the primary answer; if the keyword matching score does not reach the preset threshold, the module can select the model inference candidate answer with the highest matching score as the primary answer. Furthermore, other candidate answers from knowledge base retrieval or model inference can be added to the alternative answer list as supplementary information to enhance the completeness and reference value of the answer content. Through this method, the interface service module can effectively filter and integrate multiple candidate answers, providing users with structured and hierarchical answer content while ensuring answer accuracy, thereby improving the usability and user experience of the question-and-answer system.

[0093] In one embodiment of the present invention, the interface service module maintains multiple question-and-answer instances for providing question-and-answer services to different access parties respectively; the front-end interaction module carries an access party identifier for identifying the access party when generating the question-and-answer request; the interface service module selects a target question-and-answer instance according to the access party identifier, and performs candidate answer acquisition and aggregation processing based on the instance configuration data, knowledge base data and corpus information associated with the target question-and-answer instance.

[0094] In one embodiment of the present invention, the model service module maintains multiple model instances, each model instance being used to implement a language model and / or a model version; the interface service module determines the target model instance based on the model identifier recorded in the instance configuration data, and calls the target model instance to perform model inference on the question-and-answer request to generate candidate answers for the corresponding access party question-and-answer instance.

[0095] In one embodiment of the present invention, the model service module is further configured to, upon receiving a model update instruction based on a training corpus, train the target model instance based on the training corpus to obtain an updated model instance; after allocating an independent service port to the updated model instance and completing a health check, notify the interface service module to switch the call to the target model instance to the updated model instance; and after confirming that the call switch is completed, close the service port of the original target model instance to complete the model update without interrupting the external question-and-answer service.

[0096] Figure 3 This is a schematic diagram of the user question-and-answer process according to an embodiment of the present invention, such as... Figure 3 As shown, in one embodiment of the present invention, the user question-and-answer process may include the following steps:

[0097] Step 1: Users ask questions in the front-end window, including by typing or clicking interactive buttons;

[0098] Step 2: The front-end interaction module determines whether this is a command, such as clear, calculator, etc.

[0099] Step 3: If yes, execute the instruction and jump directly to step 10; otherwise, call the interface.

[0100] Step 4: The interface service module determines whether the user input matches keywords related to the question, issue, etc.

[0101] Step 5: If yes, use the keyword as the answer and proceed to Step 7;

[0102] Step 6: Call the interface of the model service module to perform answer inference;

[0103] Step 7: Use the background search function to search for answers as alternatives;

[0104] Step 8: If the model does not return a value, use the search result with the highest score as the answer;

[0105] Step 9: Search for links and menus in the front-end interaction module as alternatives;

[0106] Step 10: The front-end parses the answer and displays it as HTML elements to the user.

[0107] Figure 4 This is a schematic diagram of the data flow process according to an embodiment of the present invention, such as... Figure 4 As shown, in one embodiment of the present invention, the data flow process of the system may include the following steps:

[0108] Step 1: Personnel with appropriate permissions enter and maintain relevant data through the data management module;

[0109] Step 2: The instance administrator triggers instance updates or training through the data management module frontend;

[0110] Step 3: The interface service module obtains relevant data and generates keywords, indexes, configurations, etc.

[0111] Step 4: If it is training, access the update interface of the aforementioned model service module;

[0112] in,

[0113] Step 1 includes the following sub-steps:

[0114] Step 1.1: Data entered by non-global administrators is in a pending review state;

[0115] Step 1.2: The global administrator reviews the relevant data;

[0116] Step 4 includes the following sub-steps:

[0117] Step 4.1: The model service module accepts model configuration and training corpus;

[0118] Step 4.2: Compare with the previous files and save the incremental data to the model configuration and corpus files;

[0119] Step 4.3: Call the model pipeline to train the model;

[0120] Step 4.4: Start the new model process and port;

[0121] Step 4.5: After the new model process starts, switch the port it points to and shut down the old model process.

[0122] Figure 5 This is a schematic diagram of the system architecture of an embodiment of the present invention, as shown below. Figure 5 As shown, the aggregated answer question-and-answer system of the present invention can be deployed on a digital operations platform (DOS) and an access party system. The digital operations platform includes an operations management module for data statistics display and configuration management, and an interface service module for providing question-and-answer related management capabilities. The interface service module and the model service module are deployed in independent service environments, and they communicate through an HTTP interface. The interface service module is used to process question-and-answer requests, aggregate candidate answers, and forward corpus data and model management instructions to the model service module. The model service module is used to perform operations such as model inference or model updates.

[0123] On the access party's system side, a front-end interaction module is embedded in the front-end page and interacts with the interface service module via HTTP to submit question-and-answer requests and display aggregated answer results. The access party's back-end system can perform data management, model configuration, and instance management operations with the digital operation platform by configuring the access interface. Through this architecture, different access parties can access the system separately and obtain mutually isolated question-and-answer service instances. The interface service module and model service module can also be expanded according to the actual deployment, thereby supporting question-and-answer processing and model operation of multiple instances.

[0124] As can be seen from the above embodiments, this invention, by comprehensively utilizing various suitable answer generation methods such as front-end search, back-end retrieval, and model inference in the question-and-answer process, enables the system to obtain candidate information from different data sources and aggregate and display answer content, thereby achieving effective search and presentation of multiple types of data and improving the richness and quality of answer content. Through encapsulation and parameterization of front-end components and front-end / back-end configuration methods, the access party can quickly integrate the question-and-answer system without complex development and flexibly customize it according to its own business needs, thus reducing the difficulty of access and use and improving the system's customizability. Furthermore, this invention separates the interface service module and model service module and uniformly encapsulates the model access interface, enabling the system to flexibly expand question-and-answer instances and model instances, which is beneficial for improving the efficiency and update capability of model usage, thereby improving the system's maintainability and scalability.

[0125] Based on the same inventive concept, embodiments of the present invention also provide a question-and-answer method for aggregated answers, as described in the following embodiments. Since the principle of the question-and-answer method for aggregated answers is similar to that of the question-and-answer system for aggregated answers, embodiments of the question-and-answer method for aggregated answers can be found in embodiments of the question-and-answer system for aggregated answers, and repeated details will not be elaborated further.

[0126] Figure 1 This is a flowchart of the question-and-answer method for aggregating answers according to an embodiment of the present invention, as shown below. Figure 1 As shown, in one embodiment of the present invention, the question-and-answer method for aggregated answers of the present invention includes steps S101 to S105.

[0127] Step S101: On the front-end page, render the question and answer interface according to the configuration parameters, receive the interactive content input by the user on the question and answer interface, generate a question and answer request according to the interactive content, send the question and answer request to the back-end interface service, and display the aggregated answer result when the aggregated answer result returned by the interface service is received.

[0128] Step S102: On the front-end page, perform a front-end search on the interactive content based on the pre-established local index to obtain front-end search results, and display the front-end search results as alternative information together with the aggregated answer results to the user;

[0129] Step S103: During the data management process, instance data for corresponding question-and-answer instances is managed for different access parties. The instance data includes instance configuration data, knowledge base data, common question corpus, and user question-and-answer records and feedback information. Based on the instance data, knowledge base data and corpus information for question-and-answer processing are provided to the interface service, and corpus information for model reasoning is provided to the model service.

[0130] Step S104: During the interface service process, the question and answer request is received. Based on the knowledge base data and corpus information, multiple candidate answers are obtained by using at least two of the following methods: keyword matching, knowledge base retrieval, and model reasoning performed through the model service. The multiple candidate answers are aggregated to determine the main answer and alternative answer information for supplementary explanation. The aggregated answer result, including the main answer and the alternative answer information, is generated.

[0131] Step S105: During the model service process, at least one language model is invoked to perform inference based on the question-and-answer request and corpus information, generating candidate answers for the question-and-answer request and returning them to the interface service.

[0132] In one embodiment of the present invention, generating a question-and-answer request based on the interaction content includes:

[0133] The user-input interactive content is type-determined. When the interactive content conforms to a preset instruction format, the interactive content is parsed into a front-end instruction, and the operation corresponding to the front-end instruction is executed locally on the front-end.

[0134] When the interactive content does not conform to the preset instruction format, the question-and-answer request is generated based on the interactive content and sent to the interface service, while a front-end search is performed on the interactive content.

[0135] In one embodiment of the present invention, performing a front-end search on the interactive content based on a pre-established local index includes:

[0136] Receive index data sent by the interface service, and construct the local index based on the index data. The index data includes menu items, page links, frequently asked questions entries, and corresponding target addresses or operation identifiers.

[0137] When performing a front-end search, several candidate entries are selected from the local index based on keyword matching and relevance scores, and then the candidate entries are sorted according to the relevance scores and displayed to the user.

[0138] In one embodiment of the present invention, the aggregation process of the plurality of candidate answers includes:

[0139] Calculate the matching scores for candidate answers obtained through keyword matching, candidate answers obtained through knowledge base retrieval, and candidate answers obtained through model inference, respectively.

[0140] The main answer is determined based on the preset priority rules and the matching score. Specifically, at least when the matching score of keyword matching exceeds a preset threshold, the candidate answer obtained through keyword matching is determined as the main answer. When the preset threshold is not met, the candidate answer obtained based on model reasoning is determined as the main answer.

[0141] Candidate answers that were not selected as the main answer are added to the aggregated answer results as alternative answer information.

[0142] In one embodiment of the present invention, the step of generating a question-and-answer request based on the interaction content and sending the question-and-answer request to a backend interface service includes:

[0143] When generating the question-and-answer request on the front end, an access party identifier is carried to identify the access party;

[0144] During the interface service process, a target question-answer instance is selected from multiple pre-maintained question-answer instances based on the access party identifier, and candidate answer acquisition and the generation of the aggregated answer result are performed based on the instance configuration data, knowledge base data and corpus information associated with the target question-answer instance.

[0145] In one embodiment of the present invention, model inference is performed through a model service, including:

[0146] During the model service process, multiple model instances are maintained, and each model instance is used to implement a language model and / or a model version.

[0147] During the interface service process, the target model instance is determined based on the model identifier recorded in the instance configuration data, and the target model instance is invoked to perform model inference on the question and answer request to generate candidate answers for the corresponding access party question and answer instance.

[0148] In one embodiment of the present invention, the question-and-answer method for aggregating answers further includes:

[0149] During the model service process, upon receiving a model update instruction based on the training corpus, the target model instance is trained based on the training corpus to obtain the updated model instance.

[0150] Assign an independent service port to the updated model instance and complete a health check, and notify the interface service to switch the call to the target model instance to the updated model instance;

[0151] After confirming the call switch is completed, shut down the service port of the original target model instance to complete the model update without interrupting the external question and answer service.

[0152] In one embodiment of the present invention, the data management module is further configured to: select question-answer pairs that meet preset quality conditions based on the user question-answer records and feedback information, combine them with the corresponding knowledge base data and common question corpus to generate a training corpus set, and provide the training corpus set to the model service module for updating the language model.

[0153] In one embodiment of the present invention, the model service module is configured to perform model inference according to a preset model pipeline, the model pipeline including at least: a first model for classifying intent or identifying domain for question-and-answer requests; a second model for selecting or adjusting the scope of knowledge base data retrieval based on the classification results; and a third model for generating natural language candidate answers; the model service module uses the output results of the first model and the second model to constrain the generation process of the third model.

[0154] In one embodiment of the present invention, the data management module is further configured to statistically analyze indicators such as total number of questions and answers, answer success rate, and user satisfaction by question and answer instance dimension, and generate visualized data. The interface service module provides question and answer process logs to the data management module, and the data management module updates the statistical indicators of each instance based on the question and answer process logs to support the effect evaluation and configuration optimization of question and answer instances from different access parties.

[0155] In one embodiment of the present invention, the front-end interaction module and the interface service module jointly support fault degradation processing. When no aggregated answer result is received from the interface service module within a preset time window, or when the interface service module is detected to be unavailable, the front-end interaction module performs a front-end search and displays the search results based solely on the local index. When the interface service module becomes available again, it sends a recovery notification to the front-end interaction module, causing the front-end interaction module to switch back to the working mode of displaying the aggregated answer result.

[0156] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0157] To achieve the above objectives, according to another aspect of this application, a computer device is also provided. For example... Figure 6 As shown, the computer device includes a memory, a processor, a communication interface, and a communication bus. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps in the method of the above embodiments.

[0158] The processor can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0159] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and units, such as the program units corresponding to the above-described method embodiments of the present invention. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the methods described in the above-described method embodiments.

[0160] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0161] The one or more units are stored in the memory and, when executed by the processor, perform the methods described in the above embodiments.

[0162] The specific details of the aforementioned computer equipment can be understood by referring to the relevant descriptions and effects in the above embodiments, and will not be repeated here.

[0163] To achieve the above objectives, according to another aspect of this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed in a computer processor, implements the steps of the above-described aggregated answer question-and-answer method. Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0164] To achieve the above objectives, according to another aspect of this application, a computer program product is also provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described question-and-answer method for aggregated answers.

[0165] Obviously, those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, the present invention is not limited to any particular hardware and software combination.

[0166] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A question-and-answer method for aggregating answers, characterized in that, include: In the front-end page, the Q&A interface is rendered according to the configuration parameters, the interactive content entered by the user in the Q&A interface is received, a Q&A request is generated according to the interactive content, the Q&A request is sent to the back-end interface service, and the aggregated answer result is displayed when the aggregated answer result returned by the interface service is received. On the front-end page, a front-end search is performed on the interactive content based on a pre-established local index to obtain front-end search results, and the front-end search results are displayed to the user as alternative information along with the aggregated answer results; During the data management process, instance data corresponding to question-and-answer instances are managed for different access parties. The instance data includes instance configuration data, knowledge base data, common question corpus, and user question-and-answer records and feedback information. Based on the instance data, knowledge base data and corpus information for question-and-answer processing are provided to the interface service, and corpus information for model reasoning is provided to the model service. During the interface service process, the question and answer request is received. Based on the knowledge base data and corpus information, multiple candidate answers are obtained using at least two of the following methods: keyword matching, knowledge base retrieval, and model inference performed through the model service. The multiple candidate answers are aggregated to determine the main answer and alternative answer information for supplementary explanation. The aggregated answer result, including the main answer and the alternative answer information, is generated. During the model service process, at least one language model is invoked to perform inference based on the question-and-answer request and corpus information, generating candidate answers for the question-and-answer request and returning them to the interface service.

2. The question-and-answer method for aggregating answers according to claim 1, characterized in that, The step of generating a question-and-answer request based on the interaction content includes: The user-input interactive content is type-determined. When the interactive content conforms to a preset instruction format, the interactive content is parsed into a front-end instruction, and the operation corresponding to the front-end instruction is executed locally on the front-end. When the interactive content does not conform to the preset instruction format, the question-and-answer request is generated based on the interactive content and sent to the interface service, while a front-end search is performed on the interactive content.

3. The question-and-answer method for aggregating answers according to claim 1, characterized in that, The step of performing a front-end search on the interactive content based on a pre-established local index includes: Receive index data sent by the interface service, and construct the local index based on the index data. The index data includes menu items, page links, frequently asked questions entries, and corresponding target addresses or operation identifiers. When performing a front-end search, several candidate entries are selected from the local index based on keyword matching and relevance scores, and then the candidate entries are sorted according to the relevance scores and displayed to the user.

4. The question-and-answer method for aggregating answers according to claim 1, characterized in that, The aggregation process for the multiple candidate answers includes: Calculate the matching scores for candidate answers obtained through keyword matching, candidate answers obtained through knowledge base retrieval, and candidate answers obtained through model inference, respectively. The main answer is determined based on the preset priority rules and the matching score. Specifically, at least when the matching score of keyword matching exceeds a preset threshold, the candidate answer obtained through keyword matching is determined as the main answer. When the preset threshold is not met, the candidate answer obtained based on model reasoning is determined as the main answer. Candidate answers that were not selected as the main answer are added to the aggregated answer results as alternative answer information.

5. The question-and-answer method for aggregating answers according to claim 1, characterized in that, The step of generating a question-and-answer request based on the interaction content and sending the question-and-answer request to the backend interface service includes: When generating the question-and-answer request on the front end, an access party identifier is carried to identify the access party; During the interface service process, a target question-answer instance is selected from multiple pre-maintained question-answer instances based on the access party identifier, and candidate answer acquisition and the generation of the aggregated answer result are performed based on the instance configuration data, knowledge base data and corpus information associated with the target question-answer instance.

6. The question-and-answer method for aggregating answers according to claim 1, characterized in that, Model inference is performed through model services, including: During the model service process, multiple model instances are maintained, and each model instance is used to implement a language model and / or a model version. During the interface service process, the target model instance is determined based on the model identifier recorded in the instance configuration data, and the target model instance is invoked to perform model inference on the question and answer request to generate candidate answers for the corresponding access party question and answer instance.

7. The question-and-answer method for aggregating answers according to claim 1, characterized in that, Also includes: During the model service process, upon receiving a model update instruction based on the training corpus, the target model instance is trained based on the training corpus to obtain the updated model instance. Assign an independent service port to the updated model instance and complete a health check, and notify the interface service to switch the call to the target model instance to the updated model instance; After confirming the call switch is completed, shut down the service port of the original target model instance to complete the model update without interrupting the external question and answer service.

8. A question-and-answer system that aggregates answers, characterized in that, include: The front-end interaction module is set up on the front end, and the data management module, interface service module, and model service module are set up on the back end; The front-end interaction module is used to be embedded into the front-end page of the access party, render the question and answer interface according to the configuration parameters, receive the interactive content input by the user in the question and answer interface, generate a question and answer request according to the interactive content and send it to the interface service module, and display the aggregated answer result when the interface service module returns the aggregated answer result. At the same time, it performs a front-end search on the interactive content based on a pre-established local index and displays the search results obtained from the front-end search as alternative information to the user. The data management module is used to manage instance data of corresponding question-and-answer instances for different access parties. The instance data includes instance configuration data, knowledge base data, common question corpus, user question-and-answer records and feedback information. Based on the instance data, the module provides the interface service module with knowledge base data and corpus information for question-and-answer processing, and provides the model service module with corpus information for model inference. The interface service module is used to receive the question-and-answer request sent by the front-end interaction module, and based on the knowledge base data and corpus information provided by the data management module, obtain multiple candidate answers using at least two of the following methods: keyword matching, knowledge base retrieval, and model inference performed by the model service module. The multiple candidate answers are aggregated to determine the main answer and alternative answer information for supplementary explanation, generate an aggregated answer result including the main answer and the alternative answer information, and return the aggregated answer result to the front-end interaction module. The model service module is used to invoke at least one language model to perform inference based on the question-and-answer request and corpus information forwarded by the interface service module, generate candidate answers for the question-and-answer request, and return them to the interface service module.

9. The question-and-answer system with aggregated answers according to claim 8, characterized in that, The front-end interaction module is also used to determine the type of the interaction content. When the interaction content conforms to the preset instruction format, the interaction content is parsed into a front-end instruction and the operation corresponding to the front-end instruction is executed locally. When the interactive content does not conform to the preset instruction format, a question-and-answer request is generated based on the interactive content and sent to the interface service module, while a front-end search is performed on the interactive content.

10. The question-and-answer system with aggregated answers according to claim 8, characterized in that, The front-end interaction module is configured to support at least two working modes: in the first working mode, the front-end search is performed on the interactive content based on the local index and the search results are displayed, without sending the question-and-answer request to the interface service module; In the second working mode, the question-and-answer request is sent to the interface service module and the aggregated answer result is received. At the same time, a front-end search is performed and the front-end search results are displayed as supplementary information to the aggregated answer result.

11. The question-and-answer system with aggregated answers according to claim 8, characterized in that, The local index in the front-end interaction module is constructed from the index data issued by the interface service module. The index data includes menu items, page links, frequently asked questions, and corresponding target addresses or operation identifiers. When performing a front-end search, the front-end interaction module selects several candidate items from the local index based on keyword matching and relevance scores, sorts the candidate items according to the relevance scores, and then displays them to the user.

12. The question-and-answer system with aggregated answers according to claim 8, characterized in that, The interface service module is specifically used to calculate the corresponding matching scores for candidate answers obtained by keyword matching, candidate answers obtained by knowledge base retrieval, and candidate answers obtained by model reasoning when aggregating multiple candidate answers. The main answer is determined based on the preset priority rules and the matching score. At least when the matching score of keyword matching exceeds a preset threshold, the candidate answer obtained by keyword matching is determined as the main answer. When the preset threshold is not met, the candidate answer obtained by model reasoning is determined as the main answer. The candidate answers that were not selected as the main answer are then added to the aggregated answer results as alternative answer information.

13. The question-and-answer system with aggregated answers according to claim 8, characterized in that, The interface service module maintains multiple question-and-answer instances for providing question-and-answer services to different access parties; the front-end interaction module carries an access party identifier when generating the question-and-answer request; the interface service module selects a target question-and-answer instance based on the access party identifier, and performs candidate answer acquisition and aggregation processing based on the instance configuration data, knowledge base data and corpus information associated with the target question-and-answer instance.

14. The question-and-answer system with aggregated answers according to claim 8, characterized in that, The model service module maintains multiple model instances, each of which is used to implement a language model and / or a model version. The interface service module determines the target model instance based on the model identifier recorded in the instance configuration data, and calls the target model instance to perform model inference on the question-and-answer request to generate candidate answers for the corresponding access party question-and-answer instance.

15. The question-and-answer system with aggregated answers according to claim 8, characterized in that, The model service module is further configured to, upon receiving a model update instruction based on a training corpus, train the target model instance based on the training corpus to obtain an updated model instance; and after allocating an independent service port to the updated model instance and completing a health check, notify the interface service module to switch the call to the target model instance to the updated model instance. After confirming the call switch is completed, shut down the service port of the original target model instance to complete the model update without interrupting the external question and answer service.

16. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

17. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

18. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

Images