A method and system for aggregating multiple large models for user autonomous question answering

Abstract

The application provides a user self-help question and answer method and system for aggregating multiple large models, comprising: starting question and answer interaction through multiple large model selection units when the user is interacting with text-based question and answer; requesting multiple large model processing in parallel, and displaying the results using multiple large model result display units; in multiple rounds of dialogue, the user uses a question generation unit to select part of the content from the output answer of the large model and combines it with the new input content; the intelligent input optimization module outputs the original user and the system optimized double-path question to the multiple large models that have been reserved or reselected; the multiple round dialogue result output unit and the result sorting unit display the output result; the multiple model cooperation mechanism supports selecting a large model that is completely inconsistent with the previous round to answer the question. Through the cooperation and selection mechanism of multiple models, the flexibility and adaptability of the question and answer system are improved, ensuring that the user can get more comprehensive, accurate and personalized answers, and improving the practicality and response efficiency.

Description

Technical Field

[0001] This invention discloses a user-initiated question-answering method and system that aggregates multiple large models, relating to the fields of artificial intelligence, natural language processing, and human-computer interaction, and particularly to a user-initiated question-answering system that aggregates multiple large language models (LLMs), aiming to improve question-answering quality and user experience. Background Technology

[0002] Current large-scale language models perform well in question-answering tasks, but existing single large models may have limitations in answering specific questions or domains, offering limited content or lacking in professional depth. Users often struggle to assess the differences in capabilities between models and cannot choose the most suitable model based on their needs, frequently failing to obtain comprehensive and multi-faceted answers when seeking information. Therefore, there is an urgent need for a question-answering system that can integrate multiple large models and allows users to make their own choices and provide feedback, in order to improve the quality of answers and the user experience. Summary of the Invention

[0003] Currently, existing large language model question-answering systems generally suffer from the limitation of single-model interaction. Users can typically only select one large model for question-answering, receiving only a limited number of answers from a single model each time, making it difficult to compare and evaluate responses from different models in real time. This invention, by introducing collaboration among multiple large models, user-selected responses, and evaluation scoring, aims to significantly improve the answer quality and user experience of question-answering systems. It achieves parallel interaction of multiple large models and an intelligent recommendation mechanism, breaking through the technical bottleneck of traditional single-model question-answering.

[0004] This application provides a user-initiated question-answering method and system that aggregates multiple large models, including:

[0005] When users engage in text-based question-and-answer interactions, the interaction begins through multiple large model selection units, including a manually selected model module and an intelligent recommendation model module.

[0006] Based on user input, parallel request processing is performed, multiple large models are invoked simultaneously and corresponding results are generated, and the results are displayed using multiple large model result display units.

[0007] In the second and subsequent rounds of dialogue, users can use the question generation unit to flexibly select parts of the output answers from the previous round of the large model and combine them with their own new input needs. Through the intelligent input optimization module, dual-path questions are generated to complete multiple rounds of dialogue.

[0008] The intelligent input optimization module outputs dual-path questions, such as user-native questions and system optimization questions, and then inputs them to the user to retain or reselect multiple large models.

[0009] The output results are displayed through a multi-turn dialogue result output unit and a result sorting unit.

[0010] The multi-model collaboration mechanism allows for continuous and progressive interaction between models, enabling users to select a large model that is completely different from the previous one to answer subsequent questions.

[0011] The system supports default mechanisms where users do not select user rating areas, dynamic sorting mechanisms, or text tag content. In some embodiments, during user text-based question-and-answer interactions, the interaction begins through multiple large model selection units, including a manually selected model module and an intelligent recommendation model module. It supports both manual selection of n dialogue models by the user and automatic recommendation of domain-relevant models based on user input by the intelligent recommendation module.

[0012] The manual model selection module provides a large list of models for users to manually select one or more. Users can browse and choose models independently. Based on their expertise and experience, users can filter and compare models from the list to ultimately select the one that meets their needs. This unit typically provides model information such as model name, brief description, version information, release and update time, provider, areas of expertise, applicable scenarios, known limitations, sample output, and maximum token count, and offers filtering functionality to help users accurately find their target model.

[0013] The intelligent recommendation model module analyzes the semantics and domain of the user's input question, recommending relevant large-scale models for the user to choose from. Based on the user's input data and preset conditions, it analyzes the user's input question, identifies keywords, determines the domain of the question, and intelligently recommends suitable models by combining user question sentiment, historical dialogue interaction records, etc. This unit comprehensively evaluates and provides recommendation results based on the characteristics of user data, such as data type, scale, and distribution, combined with the performance indicators and applicable scenarios of pre-trained models. Simultaneously, the system provides detailed reasons for the recommendations, explaining why these models are recommended, helping users understand the logic behind the recommendations. This unit typically provides the model name and the reason for the recommendation.

[0014] In some embodiments, parallel request processing is performed based on user input, multiple large models are invoked simultaneously and corresponding results are generated, and the results are displayed using multiple large model result display units.

[0015] The system employs differentiated display strategies for multiple large model results across different devices: a 3x3 format for desktop computers and an nx3 format for mobile devices. Parallel request processing ensures that when a user selects multiple large models, the system sends the user's question to all selected models in parallel.

[0016] The results display unit for multiple large models includes a basic display module and a dual-path display module. The basic display module includes the large model name, the model's answer, and a user rating area. The dual-path display module is a further subdivision of the basic display module, including the large model name, the model's answer (subdivided into Path 1 and Path 2), and a user rating area (subdivided into Path 1 and Path 2). The first round of dialogue uses the basic display module to show the results.

[0017] The user rating area also includes a user preference recording section, used to record users' rating preferences. A five-star rating system is used.

[0018] In some embodiments, during the second and subsequent rounds of dialogue, users can utilize the question generation unit to flexibly select parts of the output answers from the previous large model and combine them with their own new input needs. Through the intelligent input optimization module, dual-path questions are generated to complete multiple rounds of dialogue.

[0019] The question generation unit intelligently transforms and integrates user interaction information to form the foundation for a new round of dialogue. The question generation unit consists of six core components: a previous round answer module, an answer improvement direction module, a user-inputted content module, user-provided examples, a model selection module, and an intelligent input optimization module. Its aim is to achieve dynamic optimization and deep customization of dialogue content.

[0020] The first-round answer module accurately captures key interaction information from the user's previous conversation: recording the main model name, the content selected by the user, and its proportion in the original answer. The first-round answer module provides both an automatic semantic segmentation module and user-selected text.

[0021] The user-provided sample is a text that the user finds particularly appealing during a non-question-and-answer process.

[0022] The answer improvement direction module completes the answer using text tags. These tags include: writing style, stylistic features, text content, word count limit (numerical), writing tone, structural logic, emotional attitude, imagery, detail description, data support, explanatory functions, and conciseness and comprehensiveness. In some embodiments, the intelligent input optimization module outputs dual-path questions, such as the user's original question and the system optimization question, which are then input to the user to save or reselect multiple large models.

[0023] The user-generated question is derived from the output content of the large model and its percentage; the user evaluation model's total performance score is obtained by directly concatenating the scores of various user evaluation indicators, answer improvement direction labels, user-provided examples, and user input content modules; the system optimization question is to refine the text based on the user-generated question by multi-weighted fusion and contextual reasoning while preserving the original semantics and logic.

[0024] Selecting the model module will handle the dual-path problem of the intelligent output optimization module.

[0025] The user review scores are tagged with accuracy, relevance, and readability.

[0026] The system optimization problem reconstructs the user's input, eliminates ambiguity and vagueness, and ensures that the generated problem is more targeted and actionable. During the optimization process, the system provides expressions that are suitable for the user's understanding, taking into account the user's knowledge level and background.

[0027] In some embodiments, the system displays the output results through a multi-turn dialogue result output unit and a result sorting unit.

[0028] The multi-turn dialogue result output unit also includes a raw multi-model result display unit and an integrated multi-model result display unit. The raw multi-model result display unit uses a dual-path display module for displaying the results. User's original question corresponds to path 1, and system optimization question corresponds to path 2.

[0029] The results sorting unit adjusts the display order of answers for different large models in the results display unit of multiple large models. That is, it adopts a dynamic sorting mechanism based on user evaluation weights, dynamically adjusts and optimizes the path according to the speed, accuracy and emotional fluctuations of user responses, and intelligently reorganizes the output results of multiple models. The results presentation follows the principle of prioritizing those with higher evaluations and outputs results according to the multiple large model results display units.

[0030] The integrated large model results display unit follows the outputs of multiple large model results display units. This unit weights user ratings from previous rounds of dialogue, trust scores for each model, and the outputs of each model, then weights and fuses them to output an integrated answer. It also provides a user-controllable feedback learning mode, allowing users to actively fine-tune the model's responses during the dialogue, or automatically adjust subsequent dialogue strategies through reinforcement learning if user satisfaction is low, ensuring that the output of each question matches user expectations.

[0031] The trust score for each model is calculated based on the user rating weights for each model, and historical performance, domain knowledge, and output consistency to assess the trustworthiness of the model.

[0032] In some embodiments, the multi-model collaboration mechanism allows for continuous, progressive interaction between models, enabling users to select a large model that is completely different from the previous one to answer subsequent questions.

[0033] The multi-model collaboration mechanism involves Model A generating an initial answer, Model B providing supplementary or improvement suggestions based on this, and finally Model C performing comprehensive integration to output a highly optimized integrated answer.

[0034] The model selection and switching unit also includes the ability for users to easily choose to continue interacting with all or some of the previous large models, or to add new models for interaction. The system provides an intuitive interface that allows users to quickly view the list of selected models and choose which model to continue the interaction with or add a new model.

[0035] When multiple models are selected, the answers from all models are displayed on the entire page in different colors.

[0036] In some embodiments, the default mechanism allows users to choose not to select the user rating area, dynamic sorting mechanism, or text tag content.

[0037] The default mechanism also includes a system default configuration requiring a maximum of 2 large models (n ≥ 2) to ensure the basic need for multi-model collaboration. The user rating area can be left unselected by the user; this weight value is equal for all models, and the subsequent dynamic ranking mechanism remains unchanged by default. Text tag content can be left unselected by the user and defaults to be consistent with the output. User conversation history is retained until manually cleared by the user.

[0038] This application also provides a user-initiated question-answering method that aggregates multiple large models, including the following steps:

[0039] S101. During user text-based question-and-answer interaction, the question-and-answer interaction begins through multiple large model selection units, including a manually selected model module and an intelligent recommendation model module.

[0040] S102. Based on the user input, perform parallel request processing, simultaneously call multiple large models and generate corresponding results, and display them using multiple large model result display units.

[0041] In S103, the second and subsequent rounds of dialogue, users can use the question generation unit to flexibly select parts of the output answers from the previous round of the large model and combine them with their own new input needs. Through the intelligent input optimization module, dual-path questions are generated to complete multiple rounds of dialogue.

[0042] S104, the intelligent input optimization module outputs dual-path problems such as user-native problems and system optimization problems, and then inputs them to the user to retain or reselect multiple large models.

[0043] S105. Display the output results through the multi-turn dialogue result output unit and the result sorting unit.

[0044] S106. The multi-model collaboration mechanism allows for continuous and progressive interaction between models, and supports users in selecting a large model that is completely different from the previous one to answer subsequent questions.

[0045] S107. Supports users not selecting user rating areas, dynamic sorting mechanisms, and default mechanisms for text tag content.

[0046] This application also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the user-initiated question-answering method that aggregates multiple large models as described above.

[0047] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the user-initiated question-answering method that aggregates multiple large models as described above.

[0048] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the user-initiated question-answering method for aggregating multiple large models as described above.

[0049] This application provides a user-initiated question-answering system and method that aggregates multiple large models. By allowing multiple large models to participate in question-answering, enabling user selection and rating feedback, and establishing a collaboration mechanism between models, it effectively solves the problem of unsatisfactory output from a single model, improving the answer quality, flexibility, and user experience of the question-answering system. Users can choose suitable combinations of large models according to their needs and personalize the answers to obtain more comprehensive, accurate, and personalized responses. Furthermore, the model collaboration mechanism in this application can fully utilize the advantages of different large models, further enhancing the quality and depth of the answers through complementarity and optimization.

[0050] This invention is not limited to the specific configurations and processes described above. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method of this invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications, and additions after understanding the spirit of this invention.

[0051] The beneficial effects of the present invention are at least as follows:

[0052] This invention presents a user-initiated question-answering system that aggregates multiple large models. By allowing users to select and interact with multiple large language models, it overcomes the limitations of traditional single-model question-answering systems. Through the introduction of multiple large model selection units, multiple large model result display units, multi-turn dialogue user question generation units, intelligent input optimization modules, multi-turn dialogue result output and result ranking units, and multi-model collaboration mechanisms, the system's flexibility and user experience are significantly improved. This invention covers various technical requirements such as multi-model interaction, intelligent recommendation, personalized scoring, and answer fusion. Through precise model collaboration and user feedback mechanisms, it effectively manages and optimizes the large-model question-answering process, reduces the limitations of single-model responses, ensures that users receive comprehensive, in-depth, and personalized answers, and shortens the path for users to obtain accurate information.

[0053] Additional advantages, objects, and features of the invention will be set forth in part in the foregoing description, and in part will become apparent to those skilled in the art upon further study, or may be learned by practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures specifically pointed out in the specification.

[0054] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description.

[0055] For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. 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. Attached Figure Description

[0056] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. In the drawings:

[0057] Figure 1 This is a schematic diagram of a user-initiated question-and-answer process that aggregates multiple large models, as described in an embodiment of the present invention.

[0058] Figure 2 This is a schematic diagram illustrating the composition of multiple large model result display units according to an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below with reference to specific embodiments and accompanying drawings. It should be emphasized that the embodiments and descriptions of this invention are intended to illustrate the technical solutions of this invention, and not to limit this invention.

[0060] In this specification, for the sake of brevity, the accompanying drawings only show the structures and / or processing steps closely related to the present invention, and other details not directly related to the present invention are omitted.

[0061] It should be noted that the following description is only a part of the specific embodiments of the present invention, and not all of the embodiments. Other embodiments that can be derived by those skilled in the art based on the content of this application without creative effort are also within the protection scope of the present invention.

[0062] In the accompanying drawings, the same reference numerals represent the same or similar parts or steps. The embodiments of this application will now be analyzed based on the accompanying drawings to further clarify the technical solutions, objectives, and advantages of the present invention.

[0063] The term "large model" in this specification refers to an artificial intelligence model based on deep learning technology, possessing massive parameters (typically billions to hundreds of billions). Through self-supervised learning on massive amounts of text data from the Internet, it can understand, generate, and process natural language, exhibiting human-like intelligence in multiple cognitive tasks. Relying on the transformer architecture, it has powerful language understanding and generation capabilities, and can perform complex dialogue interactions, knowledge question answering, text generation, and multi-domain intelligent reasoning, representing the highest level of current artificial intelligence technology.

[0064] Currently, when users engage in question-and-answer sessions using large language models, they typically can only choose a single model to communicate with, resulting in answers that lack diversity, depth, and comprehensiveness. Furthermore, users find it difficult to assess the differences in capabilities between different models. Therefore, there is an urgent need for a user-driven question-and-answer system that can integrate multiple large models to improve question-and-answer quality and user experience.

[0065] This invention provides a user-initiated question-answering method that aggregates multiple large models. The method runs on both the user end and the model end, such as... Figure 1 As shown, the method includes the following steps S101 to S107:

[0066] Step S101: During user text-based question-and-answer interaction, the question-and-answer interaction begins through multiple large model selection units, including a manually selected model module and an intelligent recommendation model module.

[0067] Step S102: Based on the user input, perform parallel request processing, simultaneously call multiple large models and generate corresponding results, and display them using multiple large model result display units.

[0068] Step S103: In the second and subsequent rounds of dialogue, users can use the question generation unit to flexibly select parts of the output answer from the previous round of the large model and combine them with their own new input needs. Through the intelligent input optimization module, dual-path questions are generated to complete multiple rounds of dialogue.

[0069] Step S104: The intelligent input optimization module outputs dual-path problems, such as user-native problems and system optimization problems, and then inputs them to the user to retain or reselect multiple large models.

[0070] Step S105: Display the output results through the multi-turn dialogue result output unit and the result sorting unit.

[0071] Step S106: The multi-model collaboration mechanism allows for continuous and progressive interaction between models, and supports users in selecting a large model that is completely different from the previous one to answer subsequent questions.

[0072] Step S107: Support the default mechanism for users not selecting user rating area, dynamic sorting mechanism, and text tag content.

[0073] The user-initiated question-answering method of the present invention, which aggregates multiple large models, effectively solves the problem of the limitations of existing single models by allowing multiple large models to participate in question-answering in parallel, user-initiated selection, and scoring feedback mechanism. It significantly improves the quality and efficiency of question-answering and ensures that users can obtain more comprehensive, accurate and personalized answers.

[0074] In step S101, during user text-based question-and-answer interaction, the question-and-answer interaction begins through multiple large model selection units, which include a manually selected model module and an intelligent recommendation model module.

[0075] In a specific implementation, multiple large model selection units achieve a flexible mechanism for multi-model selection through a manual model selection module and an intelligent recommendation model module.

[0076] In some embodiments, multiple large model selection units allow users to manually select n dialogue models, while an intelligent recommendation module automatically recommends domain-related models based on user input.

[0077] Specifically, the manual model selection module provides a large list of models, allowing users to browse and select one or more. This module offers users comprehensive model information, including model name, brief description, version information, release and update time, model provider, areas of expertise, applicable scenarios, known limitations, sample output, and maximum token count. Users can filter and compare models from the list based on their expertise and experience to ultimately select the model that meets their needs.

[0078] In some embodiments, the intelligent recommendation model module intelligently recommends relevant large-scale models to users by deeply analyzing the semantics and domain of the user's input question. Based on the user-provided input data and preset conditions, this module accurately identifies keywords and determines the domain to which the question belongs, thereby recommending the most suitable model for the user. The module's recommendation mechanism relies on a comprehensive evaluation of user data characteristics, including but not limited to data type, data size, and data distribution. Simultaneously, the module also comprehensively considers the performance metrics and applicable scenarios of the pre-trained model, ultimately providing the optimal model recommendation result through multi-dimensional and precise evaluation methods.

[0079] To enhance user experience, the system not only provides model recommendations but also elaborates on the logic and rationale behind them. The dynamic adjustment of the recommendation algorithm is based on the following key elements: the semantic relevance of the input question, the model's historical performance in a specific domain, the model's professional coverage, the model's latest performance metrics, and user sentiment.

[0080] For example, for technical problems, the system may be more inclined to recommend specialized models that excel in the technical field; for literary creation problems, it may recommend models that are better at language generation and creative writing.

[0081] Through this multi-dimensional and intelligent model selection mechanism, the system provides users with a more accurate and efficient dialogue interaction experience.

[0082] In step S102, the system performs parallel request processing based on the user input, simultaneously calling multiple large models and generating corresponding results, which are then displayed using multiple large model result display units.

[0083] In a specific implementation, multiple large model result display units achieve efficient presentation of multi-model results through parallel request processing and differentiated display strategies.

[0084] In some embodiments, the system simultaneously invokes multiple large models to generate corresponding results based on user input. The display units for the results of these multiple large models employ differentiated display strategies for different terminals: a 3x3 format for desktop computers and an nx3 format for mobile devices, in order to optimize the user's browsing experience on different devices.

[0085] Specifically, when a user selects multiple large models, the system will send the user's question to all selected models in parallel. This parallel request processing mechanism can significantly reduce user waiting time and improve interaction efficiency.

[0086] Multiple large model result display units, such as Figure 2As shown, the S201 unit for displaying results of multiple large models includes the S202 basic display module and the S203 dual-path display module. The S202 basic display module provides a standard display method for the large model name, model answer, and user evaluation scoring area. In the first round of dialogue, the system defaults to displaying the results using the S202 basic display module.

[0087] In some embodiments, the S203 dual-path display module expands upon the basic display module in a more detailed way. This module not only includes the main model name but also subdivides the model answer and user rating areas into two paths, providing users with a richer way of presenting information.

[0088] Specifically, the user rating area also integrates a user preference recording unit, used to record and analyze user rating preferences. This unit dynamically tracks and learns through the following elements: the distribution of user ratings for different models, user model preferences in different domains, the time trend of rating changes, and the consistency and stability of user ratings.

[0089] For example, if a user consistently gives a particular model a high score on technical questions, the system may prioritize those models in subsequent recommendations.

[0090] Through this intelligent result display and user feedback recording mechanism, the system can continuously optimize the user's dialogue experience and provide a more personalized model interaction method.

[0091] In step S103, during the second and subsequent rounds of dialogue, the user can use the question generation unit to flexibly select parts of the output answer from the previous round of the large model and combine them with their own new input needs. The intelligent input optimization module generates dual-path questions to complete multiple rounds of dialogue.

[0092] In a specific implementation, the multi-turn dialogue user question generation unit achieves dynamic optimization and in-depth customization of dialogue content through an intelligent information transformation and integration mechanism.

[0093] In some embodiments, the unit allows users to flexibly select parts of the output answer from the previous large model in the second and subsequent rounds of dialogue, and combine them with their new input needs to generate new questions, thereby completing multiple rounds of dialogue.

[0094] Specifically, the question generation unit consists of six core components: a preliminary answer module, an answer improvement direction module, a user-inputted content module, a user-provided sample module, a model selection module, and an intelligent input optimization module. These components work together to achieve intelligent transformation and deep customization of dialogue content.

[0095] In some embodiments, the previous round answer module accurately captures key interaction information from the user's previous dialogue. This module has the ability to record the main model name, the content selected by the user, and its proportion in the original answer. Through an automatic semantic segmentation module and user-selected text, the module can flexibly extract and process dialogue content.

[0096] Specifically, the answer improvement direction module completes the content through text tags. These text tags include multi-dimensional content characteristics: (1) writing style: academic, colloquial, literary, etc.; (2) stylistic features: argumentative, lyrical, explanatory, etc.; (3) text content: theme, core viewpoint; (4) word limit: precise numerical constraints; (5) writing tone: formal, casual, humorous, etc.; (6) structural logic: argumentation method, logical chain; (7) emotional attitude: neutral, positive, critical, etc.; (8) artistic conception and imagery: abstract, concrete, metaphor, etc.; (9) detail description: accuracy, vividness; (10) data argumentation: quantitative analysis, empirical research; (11) explanatory function: explanation, elaboration, clarification; (12) content conciseness and detail: generality, depth.

[0097] For example, for a technical question, the system may intelligently adjust the text tags based on the content of the previous conversation and the user's specific needs to generate a more accurate and personalized question.

[0098] Through this multi-dimensional and intelligent question generation mechanism, the system can continuously optimize the quality of dialogue and provide users with a more intelligent and smooth interactive experience.

[0099] In step S104, the intelligent input optimization module outputs dual-path problems, such as the user's original problem and the system optimization problem, and then inputs them to the user to retain or reselect multiple large models.

[0100] In a specific implementation, the intelligent input optimization module achieves intelligent optimization and accurate conversion of user input through a dual-path problem generation mechanism.

[0101] In some embodiments, the module can simultaneously output user-native questions and system optimization questions, and input these questions into multiple large models that the user retains or reselects.

[0102] Specifically, the generation of user-generated questions comprehensively considers multiple dimensions, including: the output content of the large model and its percentage, the total performance score of the user-evaluated model, the scores of various user evaluation indicators, the content of answer improvement direction tags, user-provided examples, and the user's original input content. These elements are directly combined to form a complete expression of the user-generated question.

[0103] In some embodiments, the system optimization problem is based on a deep optimization of the user's original problem. The optimization process follows these core principles: preserving the original semantics and logical structure, refining the text through multi-weighted fusion, user context reasoning, eliminating ambiguity and vagueness in the input, and improving the relevance and operability of the problem.

[0104] Specifically, the system optimization module will provide a more suitable expression method for users' understanding based on their knowledge level and background. The key elements considered in the optimization process include: (1) user evaluation scores of various indicators, mainly including: accuracy, the precision of the problem statement; relevance, the degree of fit between the problem and the expected goal; readability, the clarity of the problem statement and the ease of understanding.

[0105] For example, when faced with a technical problem, the system may, based on the user's technical background, transform complex technical terms into more easily understood expressions while retaining the original technical accuracy.

[0106] The selected model module will further process the dual-path problem generated by the intelligent output optimization module, ensuring that the problem can accurately match the user's actual needs.

[0107] Through this multi-dimensional and intelligent input optimization mechanism, the system can significantly improve the efficiency and experience of users interacting with large models, and achieve a more intelligent and personalized dialogue process.

[0108] In step S105, the output results are displayed through the multi-turn dialogue result output unit and the result sorting unit.

[0109] In a specific implementation, the multi-turn dialogue result output unit provides users with a more intelligent and personalized presentation of dialogue results through an innovative display and sorting mechanism.

[0110] In some embodiments, the system intelligently displays and organizes the output results of multiple models through a multi-turn dialogue result output unit and a result sorting unit. This unit consists of two core modules: a raw multi-model result display unit and an integrated large model result display unit.

[0111] Specifically, the multi-turn dialogue result output unit uses a dual-path display module to present the results. This module assigns the user's original question to path 1 and the system optimization question to path 2, providing users with a richer and more flexible way to view the results.

[0112] In some embodiments, the result sorting unit adjusts the display order of answers for different large models in the result display unit of multiple large models. That is, it adopts a dynamic sorting mechanism based on user evaluation weights, dynamically adjusting and optimizing the path according to the speed, accuracy, and emotional fluctuations of user responses, and intelligently reorganizing the output results of multiple models. The sorting process follows the core principle of "highest evaluation first" to ensure that users can quickly obtain the most relevant and highest quality answers.

[0113] Specifically, the key factors considered by the ranking mechanism include: users' historical evaluation scores of different models, the model's performance in a specific domain, the accuracy, relevance and originality of the answers, users' response speed, accuracy and emotional fluctuations, and users' personal preferences and interaction history.

[0114] For example, if a user gives a high score to a particular model's answers in multiple technical conversations, that model will receive a higher weight in subsequent result rankings.

[0115] The integrated large model results display unit follows the original multiple large model results display unit. This unit weights and fuses the outputs of each model based on the weights of user ratings in the previous rounds of dialogue, ultimately generating a comprehensive integrated answer. It provides a user-controllable feedback learning mode, allowing users to actively fine-tune the model's responses during the dialogue, or automatically adjust subsequent dialogue strategies through reinforcement learning if user satisfaction is low, ensuring that the output of each question matches user expectations. The trust score of each model is evaluated based on its historical performance, domain knowledge, and output consistency, calculated according to the user rating weights for each model.

[0116] Key strategies considered in the fusion process include: weight allocation of different model outputs, complementarity of answers between models, consistency and difference of answers, and dynamic adjustment of user preferences. Through this multi-dimensional and intelligent result output and ranking mechanism, the system can provide users with a more personalized and efficient multi-model dialogue experience, significantly improving user satisfaction with interactions with the artificial intelligence system.

[0117] In step S106, the multi-model collaboration mechanism allows for continuous and progressive interaction between models, enabling users to select a large model that is completely different from the previous model to answer subsequent questions.

[0118] In specific implementations, the multi-model collaboration mechanism enables continuous and progressive collaboration between models through innovative interaction modes, providing users with a more comprehensive and in-depth dialogue experience.

[0119] In some embodiments, this mechanism allows users to select a large model that is completely different from the front wheel for subsequent question answers, breaking the limitations of traditional dialogue modes.

[0120] Specifically, the typical process of a multi-model collaboration mechanism is as follows: First, model A generates an initial answer; second, model B provides supplementary or improvement suggestions based on the output of model A; third, model C finally integrates the outputs of the first two models to generate a highly optimized comprehensive answer.

[0121] In some embodiments, the model selection and switching unit provides users with a highly flexible interaction method. Users can: continue to interact with all or part of the previous large model, or add new models to interact with.

[0122] The system will provide an intuitive user interface, enabling users to: quickly view the list of selected models, easily select the model they want to continue the conversation with, and conveniently add new models.

[0123] Specifically, when a user selects multiple models, the system adopts a differentiated color display strategy: different models' answers will use unique colors, the overall page will present different color themes according to different models, and the color selection is based on the characteristics of the model, the style of the answer, etc.

[0124] For example, when faced with a complex technical problem, a user might choose: (1) an academic model that excels in theoretical analysis; (2) an engineering model that excels in practical application; or (3) a popular model that excels in easy-to-understand explanation. These models will collaborate to generate a comprehensive answer from multiple perspectives and levels.

[0125] Through this intelligent multi-model collaboration mechanism, the system can overcome the limitations of a single model, provide more comprehensive and in-depth dialogue content, enhance users' sense of participation and control over the dialogue process, and significantly improve the intelligence level of the artificial intelligence dialogue system.

[0126] In step S107, the default mechanism is supported, allowing users to choose not to select the user rating area, dynamic sorting mechanism, or text tag content.

[0127] In specific implementations, a flexible default mechanism is provided to ensure that the basic functions and interactive experience of multi-model collaboration are maintained even when the user does not actively select a specific configuration.

[0128] In some embodiments, the system presets multiple default configurations to meet the needs of different users.

[0129] Specifically, the default mechanism for large model selection follows these core principles: the system defaults to requiring at least 2 large models (n) to ensure basic interaction requirements for multi-model collaboration; and it provides users with the minimum available model interaction configuration.

[0130] In the default handling of the user rating area: users can choose not to rate, in which case the weight values ​​of all large models are equal by default; the subsequent dynamic sorting mechanism will remain in the default state; to avoid affecting the basic functions of the system due to users not rating.

[0131] The default mechanism for text tag content also offers a high degree of flexibility: users can choose not to specify specific text tags; the system will use tags consistent with the output content by default; and the original characteristics and semantic integrity of the content are maintained.

[0132] The handling of user conversation history follows the principle of user self-control: conversation history will be retained until the user chooses to manually delete it, providing users with the ability to track and review conversation content over the long term.

[0133] For example, for a first-time user of the system, without any additional configuration, the system will automatically select two or more models, provide a basic multi-model dialogue experience, and maintain the original semantics and style of the output.

[0134] Through this intelligent default mechanism, the system can: lower the barrier to entry for users, provide a ready-to-use interactive experience, maintain the integrity and flexibility of system functions, and adapt to the usage habits and preferences of different users. This user-friendly design philosophy reflects the system's balance between user experience and functionality, providing users with a more intelligent and personalized interactive environment.

[0135] Corresponding to the above method, the present invention also provides an apparatus / system including a computer device, the computer device including a processor and a memory, the memory storing computer instructions, the processor executing the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the apparatus / system performs the steps of the method as described above.

[0136] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned edge computing server deployment method. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0137] In summary, this invention provides a user-initiated question-answering method and system that aggregates multiple large models, offering an innovative approach to multi-model selection, interaction, and result display for user text-based question-answering interactions. The system supports manual user selection and intelligent model recommendation through multiple large model selection units, employing a parallel request processing mechanism to simultaneously call multiple large models to generate results. The core innovations of this invention include: an intelligent multi-model selection mechanism, flexible question generation and optimization units, and dynamic result sorting and display methods. The system allows users to freely switch models in multi-turn dialogues, achieving dynamic optimization of dialogue content through a previous-turn answer module and an answer improvement direction module. The multi-model collaboration mechanism allows different models to continuously and progressively interact during the dialogue, generating more comprehensive and in-depth answers. This invention covers various needs of multi-model dialogue interaction, significantly improving the user's interaction experience with the artificial intelligence system through innovative mechanisms such as intelligent recommendation, question optimization, and result sorting. It provides flexible default configurations and user-defined options, ensuring high-quality dialogue services in different usage scenarios. The innovative multi-model collaboration and intelligent optimization mechanism effectively reduces the limitations of a single model, shortens the path to obtaining accurate and comprehensive information, and provides a more intelligent and efficient solution for human-computer interaction.

[0138] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0139] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0140] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0141] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A user-initiated question-answering method and system that aggregates multiple large models, characterized in that, include: When users engage in text-based question-and-answer interactions, the interaction begins through multiple large model selection units, including a manually selected model module and an intelligent recommendation model module. Based on user input, parallel request processing is performed, multiple large models are invoked simultaneously and corresponding results are generated, and the results are displayed using multiple large model result display units. In the second and subsequent rounds of dialogue, users can use the question generation unit to flexibly select parts of the output answers from the previous round of the large model and combine them with their own new input needs. Through the intelligent input optimization module, dual-path questions are generated to complete multiple rounds of dialogue. The intelligent input optimization module outputs dual-path questions, such as user-native questions and system optimization questions, and then inputs them to the user to retain or reselect multiple large models. The output results are displayed through a multi-turn dialogue result output unit and a result sorting unit. The multi-model collaboration mechanism allows for continuous and progressive interaction between models, enabling users to select a large model that is completely different from the previous one to answer subsequent questions. It supports the default mechanism for users not to select user rating areas, dynamic sorting mechanisms, and text tag content.

2. The user-initiated question-answering method for aggregating multiple large models according to claim 1 is characterized in that, During user-text-based question-and-answer interactions, the interaction begins through multiple large model selection units, including a manually selected model module and an intelligent recommendation model module. This supports both manual selection of n dialogue models by the user and automatic recommendation of domain-relevant models based on user input by the intelligent recommendation module. The manual model selection module provides a large list of models for users to manually select one or more. Users can browse and choose models independently. Based on their expertise and experience, users can filter and compare models from the list to ultimately select the one that meets their needs. Model information includes model name, brief description, model version information, model release and update time, model provider, model's strengths, applicable scenarios, known limitations, sample output, maximum token count, etc., and filtering functionality is provided to help users accurately find their target model. The intelligent recommendation model module analyzes the semantics and domain of the user's input question, recommending relevant large-scale models for the user to choose from. Based on the user's input data and preset conditions, it analyzes the user's input question, identifies keywords, determines the domain of the question, and intelligently recommends suitable models to the user by combining user question sentiment, historical dialogue interaction records, etc. Based on the characteristics of user data, such as data type, scale, and distribution, combined with the performance metrics and applicable scenarios of pre-trained models, a comprehensive evaluation is conducted to provide recommendation results. Detailed reasons for the recommendations are also provided, explaining why these models are recommended to help users understand the logic behind the recommendations. This includes the model name and the rationale for the recommendation.

3. The user-initiated question-answering method for aggregating multiple large models according to claim 1, characterized in that, Based on user input, parallel request processing is performed, multiple large models are invoked simultaneously and corresponding results are generated, and the results are displayed using multiple large model result display units. The results display units for multiple large models employ differentiated display strategies for different terminals: the desktop version presents a 3x3 format, while the mobile version presents an nx3 format. Parallel request processing is used; when a user selects multiple large models, the user's question is sent to all selected models in parallel. The results display unit for multiple large models includes a basic display module and a dual-path display module. The basic display module includes the large model name, the model's answer, and a user rating area. The dual-path display module is a further subdivision of the basic display module, including the large model name, the model's answer (subdivided into Path 1 and Path 2), and a user rating area (subdivided into Path 1 and Path 2). The first round of dialogue uses the basic display module to show the results. The user rating area also includes a user preference recording unit, which is used to record users' rating preferences.

4. The user-initiated question-answering method for aggregating multiple large models according to claim 1, characterized in that, In the second and subsequent rounds of dialogue, users can use the question generation unit to flexibly select parts of the output answers from the previous round of the large model and combine them with their own new input needs. Through the intelligent input optimization module, dual-path questions are generated to complete multiple rounds of dialogue. The question generation unit intelligently transforms and integrates user interaction information to form the foundation for a new round of dialogue. The question generation unit consists of six core components: a previous round answer module, an answer improvement direction module, a user-inputted content module, user-provided examples, a model selection module, and an intelligent input optimization module. Its aim is to achieve dynamic optimization and deep customization of dialogue content. The first-round answer module accurately captures key interaction information from the user's previous conversation: recording the main model name, the content selected by the user, and its proportion in the original answer. The first-round answer module provides both an automatic semantic segmentation module and user-selected text. The user-provided sample is a text that the user finds particularly appealing during a non-question-and-answer process. The answer improvement direction module is completed using text tags. The text tags include: writing style, stylistic features, text content, word limit (numerical), writing tone, structural logic, emotional attitude, artistic conception, detailed description, data argumentation, explanatory function, and content conciseness and detail.

5. The user-initiated question-answering method for aggregating multiple large models according to claim 1, characterized in that, The intelligent input optimization module outputs dual-path questions, such as user-native questions and system optimization questions, and then inputs them to the user to retain or reselect multiple large models. The user's original question is obtained by combining the output content of the large model and its percentage; the user evaluation model's total performance score is obtained by directly concatenating the scores of various user evaluation indicators, the answer improvement direction tags, user-provided examples, and the user input content module; The system optimization problem involves refining the text based on the user's original question, while preserving the original semantics and logic, through multi-weighted fusion and contextual reasoning. Selecting the model module will handle the dual-path problem of the intelligent output optimization module. The user review scores are tagged with accuracy, relevance, and readability. The system optimization problem reconstructs the user's input, eliminates ambiguity and vagueness, and ensures that the generated problem is more targeted and actionable. During the optimization process, the system provides expressions that are suitable for the user's understanding, taking into account the user's knowledge level and background.

6. The user-initiated question-answering method for aggregating multiple large models according to claim 1, characterized in that, The system displays the output results through a multi-turn dialogue result output unit and a result sorting unit. The multi-turn dialogue result output unit also includes a raw multi-model result display unit and an integrated multi-model result display unit. The raw multi-model result display unit uses a dual-path display module for displaying the results. User's original question corresponds to path 1, and system optimization question corresponds to path 2. The results sorting unit adjusts the display order of answers for different large models in the results display unit of multiple large models. That is, it adopts a dynamic sorting mechanism based on user evaluation weights, dynamically adjusts and optimizes the path according to the speed, accuracy and emotional fluctuations of user responses, and intelligently reorganizes the output results of multiple models. The results presentation follows the principle of prioritizing those with higher evaluations and outputs results according to the multiple large model results display units. The integrated large model results display unit follows the outputs of multiple large model results display units. This unit weights user ratings from previous rounds of dialogue, trust scores for each model, and the outputs of each model, then weights and fuses them to output an integrated answer. It also provides a user-controllable feedback learning mode, allowing users to actively fine-tune the model's responses during the dialogue, or automatically adjust subsequent dialogue strategies through reinforcement learning if user satisfaction is low, ensuring that the output of each question matches user expectations. The trust score for each model is calculated based on the user rating weights for each model, and historical performance, domain knowledge, and output consistency to assess the trustworthiness of the model.

7. The user-initiated question-answering method for aggregating multiple large models according to claim 1, characterized in that, The multi-model collaboration mechanism allows for continuous and progressive interaction between models, enabling users to select a large model that is completely different from the previous one to answer subsequent questions. The multi-model collaboration mechanism involves Model A generating an initial answer, Model B providing supplementary or improvement suggestions based on this, and finally Model C performing comprehensive integration to output a highly optimized integrated answer. The model selection and switching unit also includes the ability for users to easily choose to continue interacting with all or some of the previous large models, or to add new models for interaction. The system provides an intuitive interface that allows users to quickly view the list of selected models and choose which model to continue the interaction with or add a new model. When multiple models are selected, the answers from all models are displayed on the entire page in different colors.

8. The user-initiated question-answering method for aggregating multiple large models according to claim 1, characterized in that, It supports the default mechanism for users not to select user rating areas, dynamic sorting mechanisms, and text tag content. The default mechanism also includes a system default configuration requiring a maximum of 2 large models (n ≥ 2) to ensure the basic need for multi-model collaboration. The user rating area can be left unselected by the user; this weight value is equal for all models, and the subsequent dynamic ranking mechanism remains unchanged by default. Text tag content can be left unselected by the user and defaults to be consistent with the output. User conversation history is retained until manually cleared by the user.

9. 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 as described in any one of claims 1 to 8.

10. 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 8.

Images