Large model interactive question and answer method, device and equipment and storage medium

Abstract

The invention discloses a question answering method, device and equipment for large model interaction and a storage medium, relates to the technical field of computers, is applied to a preset interaction interface of a client, and comprises the following steps: displaying large models in a model library in a classified manner based on a preset screening panel, and determining a to-be-scheduled large model list according to received questioner selection information; transmitting the to-be-scheduled large model list and question information sent by the questioner into an interaction system, so as to distribute the question information to each target large model in the list through the interaction system; after acquiring and displaying reply information output by each target large model, integrating the reply information based on the reply fusion instruction; and after answer integration is completed, if an answer adjustment instruction fed back by the questioner is received, adjusting an integration result based on the answer adjustment instruction to determine a target answer of the question information. According to the method, the limitation of question answering by adopting a single model is avoided, and the model scheduling capability and the accuracy, pertinence, reliability and comprehensiveness of question answering are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a large-scale model interactive question-answering method, device, equipment, and storage medium. Background Art

[0002] Due to differences in training data, training methods, and inference parameter configuration, different models may produce diverse outputs for the same question. To facilitate comparison of the outputs of various models and improve answer satisfaction, there are currently two approaches to integrating large models: The first approach provides a list of selectable large models, allowing the questioner to select a single model to answer. However, if the questioner wishes to compare answers from different models, they must manually switch models and re-enter the question. This not only adds steps but can also degrade the questioner's experience. Frequent switching and re-entering can disrupt the questioner's thinking process and reduce efficiency. The second approach uses a hybrid model, which automatically schedules the appropriate large model based on the questioner's current question. However, the questioner cannot directly control the model selection process. This is especially true in areas where evaluation criteria are more subjective, such as text generation. The choice of hybrid model may not fully meet the questioner's expectations, and the scheduling algorithm itself may contain biases or errors, further impacting the questioner's experience. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a large-model interactive question-answering method, apparatus, device, and storage medium. This method can achieve large-model interaction without manually switching between different models or repeatedly inputting questions, thereby avoiding the limitations of answering questions using a single model, improving model scheduling capabilities, and thereby improving the accuracy, pertinence, reliability, and comprehensiveness of question-answering, and enhancing service quality and the questioner's experience. The specific solution is as follows:

[0004] In a first aspect, the present application provides a large-scale model interactive question-answering method, which is applied to a preset interactive interface of a client, including:

[0005] Based on the preset filter panel, the major models in the model library are categorized and displayed, and the list of major models to be scheduled is determined based on the selection information received from the questioner;

[0006] The list of large models to be scheduled and the question information sent by the questioner are transmitted to the interactive system, so that the question information is distributed to each target large model in the list of large models to be scheduled through the interactive system and a preset distribution mechanism;

[0007] After obtaining and displaying the answer information corresponding to the question information output by each of the target large models, integrating the answer information based on the answer fusion instruction received from the questioner to determine an integration result;

[0008] After completing the reply integration, if a reply adjustment instruction is received from the questioner, an adjustment operation is triggered on the integration result based on the reply adjustment instruction to complete the corresponding large model interaction operation and determine the target reply to the question information.

[0009] Optionally, the method of classifying and displaying the major models in the model library based on a preset filter panel and determining a list of major models to be scheduled according to the questioner's selection information received includes:

[0010] Based on the preset screening panel and using the field classification method, the expertise information of each large model in the model library is visualized, and the large models are sorted based on the preset performance indicators and preset sorting rules to complete the corresponding large model display operation;

[0011] A list of large models to be scheduled for this interaction is determined from the model library according to the received questioner selection information.

[0012] Optionally, distributing the problem information to each target large model in the list of large models to be scheduled through the interactive system and a preset distribution mechanism includes:

[0013] The problem information is distributed in parallel to each target large model in the list of large models to be scheduled through the interactive system and based on the asynchronous message queue, so that each target large model is controlled to run in a containerized environment respectively, and the reply information corresponding to the problem information output by each target large model is obtained in parallel.

[0014] Optionally, the obtaining and displaying the answer information corresponding to the question information output by each of the target large models includes:

[0015] Obtaining answer information corresponding to the question information output by each of the target large models;

[0016] Performing multiple verifications on the reply information based on a preset multiple verification mechanism to determine a verification result;

[0017] Based on the verification results and the response area corresponding to each target large model in the preset interface column layout, each reply information is displayed in parallel; wherein, the response area includes an answer display window for displaying the reply information and an interactive operation bar for dragging reply paragraphs or comparing reply contents.

[0018] Optionally, integrating the reply information based on the received reply fusion instruction sent by the questioner to determine an integration result includes:

[0019] receiving a reply fusion instruction sent by the questioner after observing and comparing the reply information;

[0020] Based on the reply fusion instruction, drag the target paragraph in the first reply information of the first large model or the first reply information to the corresponding response area of ​​the second large model to trigger the pre-configured target instruction and complete the corresponding first reply integration operation; the target instruction includes a rewrite instruction, a comparison instruction and an integration instruction;

[0021] Alternatively, based on the reply fusion instruction, the reply information is checked, compared, and reconstructed by using prompt word intervention to complete the corresponding second reply integration operation;

[0022] Determine the current integration results.

[0023] Optionally, if a reply adjustment instruction fed back by the questioner is received, triggering an adjustment operation on the integration result based on the reply adjustment instruction includes:

[0024] After visually displaying the current integration result, if feedback is received from the questioner regarding a reply adjustment instruction for the integration result, based on the reply adjustment instruction, an adjustment operation is triggered on the integration result by dragging or prompt word intervention to complete the corresponding large model interaction operation and determine the target reply to the question information.

[0025] Optionally, performing multiple verifications on the reply information based on a preset multiple verification mechanism to determine a verification result includes:

[0026] Performing text alignment and denoising processing on each of the reply messages to obtain a processing result;

[0027] Performing multi-granularity similarity calculation based on the hybrid similarity algorithm and each processed reply in the processing result, and adjusting the weight value according to the determined similarity result to determine the weight value of each processed reply;

[0028] Performing a cross-model consensus analysis based on each of the processed responses and the similarity results, and determining whether to trigger an external verification operation based on the consensus analysis result to obtain a determination result;

[0029] If the judgment result is yes, then numerical verification is performed based on the preset heterogeneous data source, the preset bulldozer distance algorithm and each of the processed responses, and each of the processed responses is adjusted using the numerical verification result to determine an adjusted response;

[0030] The adjusted response is checked based on a preset logical consistency check strategy to complete multiple verification operations and determine a verification result.

[0031] In a second aspect, the present application provides a large-scale model interactive question-answering device, which is applied to a preset interactive interface of a client, including:

[0032] The list determination module is used to classify and display the major models in the model library based on the preset filtering panel, and determine the list of major models to be scheduled based on the selection information received from the questioner;

[0033] a question distribution module, configured to transmit the list of large models to be scheduled and the question information sent by the questioner to the interactive system, so as to distribute the question information to each target large model in the list of large models to be scheduled through the interactive system and a preset distribution mechanism;

[0034] a response integration module, configured to, after acquiring and displaying the response information corresponding to the question information output by each of the target large models, integrate the response information based on the response fusion instruction received from the questioner to determine an integration result;

[0035] The reply adjustment module is used to trigger an adjustment operation on the integration result based on the reply adjustment instruction after completing the reply integration, if the reply adjustment instruction is received from the questioner, so as to complete the corresponding large model interaction operation and determine the target reply to the question information.

[0036] In a third aspect, the present application provides an electronic device, comprising:

[0037] Memory, used to store computer programs;

[0038] A processor is used to execute the computer program to implement the steps of the aforementioned large model interactive question-answering method.

[0039] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the steps of the aforementioned large-model interactive question-and-answer method.

[0040] It can be seen that in this application, the preset interactive interface applied to the client includes: classifying and displaying the major models in the model library based on the preset filtering panel, and determining the list of major models to be scheduled based on the selection information received from the questioner; transmitting the list of major models to be scheduled and the question information sent by the questioner into the interactive system, so that the question information can be distributed to the target major models in the list of major models to be scheduled through the interactive system and the preset distribution mechanism; after obtaining and displaying the reply information corresponding to the question information output by each of the target major models, integrating the reply information based on the reply fusion instruction sent by the questioner to determine the integration result; after completing the reply integration, if a reply adjustment instruction is received from the questioner, triggering an adjustment operation on the integration result based on the reply adjustment instruction to complete the corresponding major model interaction operation and determine the target reply to the question information. That is to say, in this application, large model interaction is achieved through a unified preset interactive interface. Specifically, the large models are first displayed based on the preset filtering panel classification to determine the list of large models to be scheduled, and then the question and the list are passed to the interactive system so that the interactive system can distribute the question to each target large model in the list. After displaying the reply information output by each target large model, the reply information is integrated using the reply fusion instruction sent by the questioner, and after the integration is completed, if the reply adjustment instruction is received from the questioner, the integration result is further adjusted to determine the target reply. In this way, large model interaction can be achieved without manually switching between different models or repeatedly inputting questions, thereby avoiding the limitations of answering questions using a single model, improving the model scheduling capability, and thereby improving the accuracy, pertinence, reliability and comprehensiveness of questions and answers, and improving the service quality and the questioner experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0042] Figure 1 A flow chart of a large-scale model interactive question-answering method provided in this application;

[0043] Figure 2 A schematic diagram of the answers to the General Thousand Questions model provided for this application;

[0044] Figure 3 A schematic diagram of the answer of a DeepSeek large model provided for this application;

[0045] Figure 4A schematic diagram of an integrated answer provided for this application;

[0046] Figure 5 A schematic diagram of the structure of a large-scale interactive question-answering device provided in this application;

[0047] Figure 6 This is a structural diagram of an electronic device provided in this application. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] There are currently two ways to integrate large models: the first way is to provide a list of optional large models, and the questioner selects a single model to answer. However, if the questioner wants to compare the answers of different models, he needs to manually switch the model and re-enter the question, which not only increases the operation steps, but may also lead to a decline in the questioner's experience. Frequent switching and re-entering of questions may interrupt the questioner's thinking process and reduce efficiency. The second way is to use a hybrid model, and the hybrid model automatically schedules the appropriate large model according to the questioner's current question. However, the questioner cannot directly control the model selection process, especially in areas such as text generation where the evaluation criteria are more subjective. The choice of hybrid model may not fully meet the questioner's expectations, and the scheduling algorithm of the hybrid model itself may also have deviations or errors, further affecting the questioner's experience. To this end, the present application provides a large-model interactive question-and-answer solution that can effectively avoid the limitations of using a single model to answer questions, improve the model scheduling capability, and thereby improve the accuracy, pertinence, reliability and comprehensiveness of questions and answers.

[0050] See also Figure 1 As shown, the embodiment of the present invention discloses a large model interactive question-answering method, which is applied to a preset interactive interface of a client, including:

[0051] Step S11: Classify and display the major models in the model library based on the preset screening panel, and determine the list of major models to be scheduled according to the received questioner's selection information.

[0052] Specifically, in this embodiment, the questioner is first required to select the model category that he wants to schedule. That is, based on the preset filtering panel and using the field classification method, the expertise information of each major model in the model library is visualized, and the major models are sorted based on the preset performance indicators and preset sorting rules to complete the corresponding large model display operation; according to the received questioner selection information, the list of large models to be scheduled for this interaction is determined from the model library. It should be understood that the preset interactive interface has a built-in model library and provides a filtering panel for displaying model expertise by field classification, and presenting key indicators such as response speed, knowledge update timeliness, and context window length by performance index sorting. The questioner can select multiple large models that he wants to interact with later according to his own preferences. After selecting the model he wants to interact with, the interface obtains the selected model list S, , among which Refers to the 1st, 2nd, ..., nth target large model in the list.

[0053] Step S12: The list of large models to be scheduled and the question information sent by the questioner are transmitted to the interactive system, so that the question information is distributed to each target large model in the list of large models to be scheduled through the interactive system and a preset distribution mechanism.

[0054] In this embodiment, after the list S of large models to be scheduled is determined and the questioner enters the desired question x, the interface transmits the question and the determined list S to the system. The system then distributes x in parallel to all target large models in list S via an asynchronous message queue. That is, through the interactive system and based on the asynchronous message queue, the question information is distributed in parallel to each target large model in the list of large models to be scheduled, so that by controlling each target large model to run in a containerized environment, the response information corresponding to the question information output by each target large model can be obtained in parallel.

[0055] It should be understood that each target large model runs independently in a containerized environment to ensure the isolation of computing resources and a secure sandbox mechanism, and obtains the response information A(x) of multiple target large models selected by the questioner for question x in parallel, and .in, Represents the answer information of the nth target large model to question x.

[0056] Step S13: After obtaining and displaying the reply information corresponding to the question information output by each of the target large models, the reply information is integrated based on the reply fusion instruction received from the questioner to determine the integration result.

[0057] Specifically, in this embodiment, the main interface displays the answers of each target large model in columns in parallel. That is, the reply information corresponding to the question information output by each target large model is obtained; the reply information is multi-verified based on the preset multiple verification mechanism to determine the verification result; based on the verification result and the response area corresponding to each target large model in the preset interface column layout, each reply information is displayed in parallel; wherein, the response area includes an answer display window for displaying the reply information and an interactive operation bar for dragging the reply paragraph or comparing the reply content. It can be understood that the interface adopts an adaptive column layout (for example, a 2-4 column window), and the response area of ​​each target large model includes an answer display window and an interactive operation bar for the questioner to perform operations such as dragging paragraphs or comparing versions, and then feedback on how to integrate, summarize or readjust the answers of multiple target large models.

[0058] Furthermore, regarding the integration and summarization of answers, the questioner can, after observing the answers of multiple models, feed back the answer of the integrated model, that is, receive the answer fusion instruction sent by the questioner after observing and comparing the answer information; based on the answer fusion instruction, drag the target paragraph or the first answer information in the first answer information of the first model to the corresponding answer area of ​​the second model to trigger the pre-configured target instruction and complete the corresponding first answer integration operation; the target instruction includes rewriting instruction, comparison instruction and integration instruction; or, based on the answer fusion instruction, use prompt word intervention to check, compare and reconstruct the answer information to complete the corresponding second answer integration operation; determine the current integration result. It should be understood that the answer fusion instruction based on the feedback of the questioner intervenes in the integration of multiple model answers through prompt words, and can further guide the knowledge fusion between models through natural language instructions or visualization operations. The specific steps for the integration of natural language instructions and visualization operations are as follows.

[0059] (1) Natural language instructions:

[0060] Use keywords, such as "@", to specify how a large model should adjust or integrate the answers of a specific large model with the help of prompt words. The functions that can be completed include inspection, comparison, reconstruction, etc. For example, "Please @model Check @Model Is there any bug in the code I just wrote? "Please @model Compare @model and @model Which answer is better? "Please use @model The format and @model Rewrite this work summary based on the outline of the work";

[0061] 2) Visual operation:

[0062] The system can have built-in template prompts for rewriting or comparison. When the questioner drags a model answer paragraph or a complete answer to the answer area of ​​the corresponding large model, rewriting or other preset instructions are triggered. Dragging the answers of multiple models to the answer area of ​​the corresponding model triggers comparison, integration or other preset instructions.

[0063] It should be further understood that after receiving the response output by the model and before integrating the response, this embodiment also requires multiple verifications of the credibility of the response. That is, in a specific implementation, this embodiment adopts a triple verification mechanism, namely: cross-model consensus analysis, external knowledge verification, and logical consistency check. Specifically: text alignment and denoising are performed on each response information to obtain a processing result; multi-granularity similarity calculation is performed based on the hybrid similarity algorithm and each processed response in the processing result, and the weight value is adjusted according to the determined similarity result to determine the weight value of each processed response; cross-model consensus analysis is performed based on each processed response and the similarity result, and it is determined whether to trigger an external verification operation based on the consensus analysis result to obtain a judgment result; if the judgment result is yes, numerical verification is performed based on the preset heterogeneous data source, the preset bulldozer distance algorithm, and each processed response, and the numerical verification result is used to adjust each processed response to determine the adjusted response; the adjusted response is checked based on the preset logical consistency check strategy to complete multiple verification operations and determine the verification result.

[0064] It can be understood that in the first-level verification mechanism, text alignment and denoising are first performed to remove model-specific tag information in the response and unify the digital format. Then, a mixed similarity algorithm (such as Jaccard similarity + semantic similarity) and dynamically adjustable weight values ​​are used to analyze the consensus of the response. The weights can be adjusted in combination with text length adjustment and similarity results to finally determine the consensus analysis results. Afterwards, a judgment is made. If the consensus of a certain reply information is greater than the configured threshold, it is directly adopted as a credible reply. Otherwise, a second verification mechanism is triggered, namely external knowledge verification. External verification requires calling an external API (Application Programming Interface) (such as the interface of Wolfram Alpha and Google SGE) to access the preset heterogeneous data source. After the access is completed, the processed reply is numerically verified using a preset bulldozer distance algorithm (for example, the improved Wasserstein distance algorithm (also known as Earth Mover's Distance, a metric used to measure the difference between two probability distributions)) to complete the knowledge alignment. If the numerical verification result indicates that the external verification matches, it is marked as verified. Otherwise, a manual review mark is triggered to confirm the adjusted reply. In addition, in this second verification, the adjusted reply can also be time-sensitive based on the accessed data source to ensure the accuracy of the reply. In the third level verification mechanism, the response is first converted from natural language to formal logic. For example, a large language model + rule template can be used to complete the conversion. Then, the Z3 solver is integrated to build logical constraints and add non-conflicting constraints, and / or a domain-specific rule engine is used to complete logical consistency checks to complete multiple verification operations and determine the verification results.

[0065] In addition, the aforementioned multiple verifications can be further optimized. For example, 1) by designing a cache layer to build a verification result cache tree, in which different cache validity periods can be set according to different fields. For questions and answers related to specific fields, a longer cache validity period can be set, and user-defined settings are supported; 2) when there are many questions, parallel verification can be implemented to improve efficiency; 3) incremental verification can be implemented, such as using the edit distance algorithm to detect text changes. In addition, the verification results and the verification process can be visualized to show the verification evidence chain. In this way, the accuracy of questions and answers can be further improved and response delays can be reduced.

[0066] Step S14: After completing the reply integration, if a reply adjustment instruction is received from the questioner, an adjustment operation is triggered on the integration result based on the reply adjustment instruction to complete the corresponding large model interaction operation and determine the target reply to the question information.

[0067] In this embodiment, after the integrated answer is determined, the integrated answer result can be further adjusted through natural language instructions or visual operations. A Rewrite the answer and further expand and improve it. That is, after visually displaying the current integration results, if the questioner receives feedback from the questioner regarding the integrated results and sends a response adjustment instruction, based on the response adjustment instruction, the integrated results are adjusted by dragging or prompt word intervention to complete the corresponding large model interaction operation and determine the target response to the question information. The target response is the response that satisfies the questioner.

[0068] In summary, the multi-model intelligent interaction method in this embodiment establishes a dynamic collaborative workflow, enabling deep collaboration between questioners and multiple large models. It efficiently integrates and schedules the outputs of these models, allowing questioners to quickly obtain satisfactory answers within a unified interactive interface without having to manually switch between models or repeatedly input questions. Furthermore, it transcends the knowledge boundaries of a single model, aggregating information from multiple sources for cross-validation. The questioner intervention mechanism (answer fusion / adjustment instructions fed back by the questioner) effectively corrects deviations between the model and the questioner's expectations. Through a structured collaborative process, multiple large models are transformed into a programmable cluster of intelligent agents, demonstrating the dual benefits of significantly improved quality and lowered user barriers in complex problem-solving scenarios.

[0069] It can be seen that in this application, large model interaction is achieved through a unified preset interactive interface. Specifically, the large models are first displayed based on the preset filter panel classification to determine the list of large models to be scheduled, and then the question and the list are passed to the interactive system so that the interactive system can distribute the question to each target large model in the list. After displaying the reply information output by each target large model, the reply information is integrated using the reply fusion instruction sent by the questioner, and after the integration is completed, if the reply adjustment instruction is received from the questioner, the integration result is further adjusted to determine the target reply. In this way, large model interaction can be achieved without manually switching different models or repeatedly inputting questions, thereby avoiding the limitations of answering questions using a single model, improving the model scheduling capability, and thereby improving the accuracy, pertinence, reliability and comprehensiveness of questions and answers, and improving the service quality and the questioner experience.

[0070] The following combination Figure 2-Figure 4 The schematic diagram disclosed in the figure specifically illustrates the technical solution of the embodiment of the present application.

[0071] Assume that the question input by the questioner is x = "Write a travelogue of Chengdu", and the list of large models to be scheduled selected by the questioner is S = [ , ],in, It is the Tongyi Qianwen model (i.e. Qwen), is the DeepSeek big model, and the answer of the big model selected by the questioner to question x is ,in, like Figure 2 As shown, like Figure 3 The complete process of this method is as follows:

[0072] 1) Select the model:

[0073] The questioner selects the model he wants to schedule. Assume that the questioner selects Tongyi Qianwen and DeepSeek. After the questioner completes the selection, the model list S selected by the questioner is obtained. , ],in, For the Tongyi Qianwen large model, It is the DeepSeek large model;

[0074] 2) Input problem:

[0075] The questioner enters the question x they want to ask, assuming x is "Write a travelogue about Chengdu." The question x and the model list S selected by the questioner are passed to the system.

[0076] 3) Parallel display model answer:

[0077] The answers generated by each model selected by the questioner for the question x are displayed in parallel on the main interface, such as Figure 2 、 3 As shown in the figure, the questioner can clearly observe and compare the answers of each model to question x;

[0078] 4) Instruction fusion:

[0079] If the questioner observes the model through comparison and model After answering, we expect to get the model Text frame and model The model can be instructed to write new answers in a reorganized text style by simple instructions. Further integration of answers, the new instruction can be "Please @Qwen write me a travelogue of Chengdu in the style of the text generated by @DeepSeek and the framework of the text generated by @Qwen." The questioner can click on the model Or drag the model directly In the dialog box, the text "@DeepSeek" is automatically generated in the interactive input box, or the keyword "@" is directly used. The keyword is not limited to "@". Figure 4To indicate the answer generated by Qwen after fusion;

[0080] 5) Further interaction:

[0081] If the questioner still needs to further modify the answer indicated by the fusion, for example, if they want the text to be richer or longer, they can further instruct the model to expand and improve the answer. The further instruction could be "Please expand and improve the above answer, with a word count of at least 3,000 words."

[0082] Through the above steps, the questioner can more conveniently interact with multiple models and more intuitively instruct the model how to selectively fuse the responses from multiple models. This allows the questioner to more precisely instruct the model how to combine the strengths of different models to generate more comprehensive, accurate, or targeted responses, thereby obtaining the desired answer, rather than relying solely on the decision accuracy of the hybrid model. This method lowers the threshold for multi-model collaboration through visual interaction. While maintaining control over the questioner, it leverages the generative power of large models to achieve a "1+1>2" effect. By effectively integrating the capabilities of multiple models, the value generated exceeds the effect of simply adding the functions of individual models.

[0083] See also Figure 5 As shown, the embodiment of the present application also discloses a large-scale model interactive question-answering device, which is applied to a preset interactive interface of a client, including:

[0084] The list determination module 11 is used to classify and display the major models in the model library based on the preset filtering panel, and determine the list of major models to be scheduled according to the questioner's selection information received;

[0085] The question distribution module 12 is used to transmit the list of large models to be scheduled and the question information sent by the questioner to the interactive system, so as to distribute the question information to each target large model in the list of large models to be scheduled through the interactive system and a preset distribution mechanism;

[0086] The answer integration module 13 is configured to, after acquiring and displaying the answer information corresponding to the question information output by each target large model, integrate the answer information based on the answer fusion instruction received from the questioner to determine an integration result;

[0087] The answer adjustment module 14 is used to trigger an adjustment operation on the integration result based on the answer adjustment instruction after completing the answer integration, if an answer adjustment instruction is received from the questioner, so as to complete the corresponding large model interaction operation and determine the target answer to the question information.

[0088] Among them, for more specific working processes of the above modules, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0089] It can be seen that in this application, large model interaction is achieved through a unified preset interactive interface. Specifically, the large models are first displayed based on the preset filter panel classification to determine the list of large models to be scheduled, and then the question and the list are passed to the interactive system so that the interactive system can distribute the question to each target large model in the list. After displaying the reply information output by each target large model, the reply information is integrated using the reply fusion instruction sent by the questioner, and after the integration is completed, if the reply adjustment instruction is received from the questioner, the integration result is further adjusted to determine the target reply. In this way, large model interaction can be achieved without manually switching different models or repeatedly inputting questions, thereby avoiding the limitations of answering questions using a single model, improving the model scheduling capability, and thereby improving the accuracy, pertinence, reliability and comprehensiveness of questions and answers, and improving the service quality and the questioner experience.

[0090] In some specific embodiments, the list determination module 11 may specifically include:

[0091] A large model sorting unit is used to visualize the expertise information of each large model in the model library based on a preset screening panel and in a field classification manner, and to sort each large model based on preset performance indicators and preset sorting rules to complete the corresponding large model display operation;

[0092] The list determining unit is used to determine the list of large models to be scheduled for this interaction from the model library according to the received questioner selection information.

[0093] In some specific embodiments, the question distribution module 12 may specifically include:

[0094] A question distribution unit is used to distribute the question information in parallel to each target large model in the list of large models to be scheduled through the interactive system and based on an asynchronous message queue, so as to obtain the reply information corresponding to the question information output by each target large model in parallel by controlling each target large model to run in a containerized environment respectively.

[0095] In some specific embodiments, the response integration module 13 may specifically include:

[0096] a response obtaining unit, configured to obtain response information corresponding to the question information output by each of the target large models;

[0097] A multiple verification unit, configured to perform multiple verifications on the reply information based on a preset multiple verification mechanism to determine a verification result;

[0098] A reply display unit is used to display each reply information in parallel based on the verification results and the response area corresponding to each target large model in the preset interface column layout; wherein, the response area includes an answer display window for displaying the reply information and an interactive operation bar for dragging reply paragraphs or comparing reply contents.

[0099] In some specific embodiments, the response integration module 13 may specifically include:

[0100] An instruction receiving unit, configured to receive an answer fusion instruction sent by the questioner after observing and comparing the answer information;

[0101] A first integration unit is configured to drag a target paragraph in the first reply information of the first large model or the first reply information to the corresponding response area of ​​the second large model based on the reply fusion instruction, so as to trigger a pre-configured target instruction and complete the corresponding first reply integration operation; the target instruction includes a rewrite instruction, a comparison instruction, and an integration instruction;

[0102] A second integration unit is configured to, based on the reply fusion instruction, examine, compare, and reconstruct the reply information by means of prompt word intervention to complete a corresponding second reply integration operation;

[0103] The result determination unit is used to determine the current integration result.

[0104] In some specific embodiments, the response adjustment module 14 may specifically include:

[0105] The answer adjustment unit is used to trigger an adjustment operation on the integration result by dragging or prompt word intervention based on the answer adjustment instruction after visually displaying the current integration result, if the answer adjustment instruction is received from the questioner regarding the integration result, so as to complete the corresponding large model interaction operation and determine the target answer to the question information.

[0106] In some specific embodiments, the multiple verification unit may specifically include:

[0107] A reply processing subunit, configured to perform text alignment and denoising processing on each reply message to obtain a processing result;

[0108] a weight value determination subunit, configured to perform multi-granularity similarity calculation based on the hybrid similarity algorithm and each processed reply in the processing result, and adjust the weight value according to the determined similarity result to determine the weight value of each processed reply;

[0109] an external verification judgment subunit, configured to perform a cross-model consensus analysis based on each of the processed responses and the similarity results, and determine whether to trigger an external verification operation based on the consensus analysis result to obtain a judgment result;

[0110] an external verification operation execution subunit, configured to, if the judgment result is yes, perform numerical verification based on a preset heterogeneous data source, a preset bulldozer distance algorithm, and each of the processed responses, and adjust each of the processed responses using the numerical verification result to determine an adjusted response;

[0111] The verification result determination subunit is used to check the adjusted response based on a preset logical consistency check strategy to complete multiple verification operations and determine the verification result.

[0112] Furthermore, the embodiment of the present application also discloses an electronic device, Figure 6 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram should not be considered as any limitation to the scope of application of the present application.

[0113] Figure 6 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps of the large-scale model interactive question-and-answer method disclosed in any of the aforementioned embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0114] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0115] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0116] The operating system 221 is used to manage and control the hardware devices on the electronic device 20 and the computer program 222, and can be Windows Server, NetWare, Unix, Linux, etc. In addition to including a computer program capable of implementing the large-model interactive question-and-answer method performed by the electronic device 20 as disclosed in any of the aforementioned embodiments, the computer program 222 can further include computer programs capable of performing other specific tasks.

[0117] Furthermore, this application discloses a computer-readable storage medium for storing a computer program; wherein, when executed by a processor, the computer program implements the aforementioned large-scale model interactive question-and-answer method. The specific steps of this method can be found in the corresponding content disclosed in the aforementioned embodiments and will not be repeated here.

[0118] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0119] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0120] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0121] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0122] The above is a detailed introduction to the technical solution provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A large-scale interactive question-answering method, characterized in that: The preset interactive interface applied to the client includes: Based on the preset filter panel, the major models in the model library are categorized and displayed, and the list of major models to be scheduled is determined based on the selection information received from the questioner; The list of large models to be scheduled and the question information sent by the questioner are transmitted to the interactive system, so that the question information is distributed to each target large model in the list of large models to be scheduled through the interactive system and a preset distribution mechanism; After obtaining and displaying the answer information corresponding to the question information output by each of the target large models, integrating the answer information based on the answer fusion instruction received from the questioner to determine an integration result; After completing the reply integration, if a reply adjustment instruction is received from the questioner, an adjustment operation is triggered on the integration result based on the reply adjustment instruction to complete the corresponding large model interaction operation and determine the target reply to the question information.

2. The large-scale interactive question-answering method according to claim 1, characterized in that: The method of classifying and displaying the major models in the model library based on the preset screening panel and determining the list of major models to be scheduled according to the questioner's selection information received includes: Based on the preset screening panel and using the field classification method, the expertise information of each large model in the model library is visualized, and the large models are sorted based on the preset performance indicators and preset sorting rules to complete the corresponding large model display operation; A list of large models to be scheduled for this interaction is determined from the model library according to the received questioner selection information.

3. The large-scale interactive question-answering method according to claim 1, characterized in that: The step of distributing the problem information to each target large model in the list of large models to be scheduled through the interactive system and a preset distribution mechanism includes: The problem information is distributed in parallel to each target large model in the list of large models to be scheduled through the interactive system and based on the asynchronous message queue, so that each target large model is controlled to run in a containerized environment respectively, and the reply information corresponding to the problem information output by each target large model is obtained in parallel.

4. The large-scale model interactive question-answering method according to any one of claims 1 to 3, characterized in that: The obtaining and displaying the answer information corresponding to the question information output by each target large model includes: Obtaining answer information corresponding to the question information output by each of the target large models; Performing multiple verifications on the reply information based on a preset multiple verification mechanism to determine a verification result; Based on the verification results and the response area corresponding to each target large model in the preset interface column layout, each reply information is displayed in parallel; wherein, the response area includes an answer display window for displaying the reply information and an interactive operation bar for dragging reply paragraphs or comparing reply contents.

5. The large-scale model interactive question-answering method according to claim 4, characterized in that: The step of integrating the reply information based on the reply fusion instruction received from the questioner to determine an integration result includes: receiving a reply fusion instruction sent by the questioner after observing and comparing the reply information; Based on the reply fusion instruction, drag the target paragraph in the first reply information of the first large model or the first reply information to the corresponding response area of ​​the second large model to trigger the pre-configured target instruction and complete the corresponding first reply integration operation; the target instruction includes a rewrite instruction, a comparison instruction and an integration instruction; Alternatively, based on the reply fusion instruction, the reply information is checked, compared, and reconstructed by using prompt word intervention to complete the corresponding second reply integration operation; Determine the current integration results.

6. The large-scale model interactive question-answering method according to claim 1, characterized in that: If a reply adjustment instruction fed back by the questioner is received, triggering an adjustment operation on the integration result based on the reply adjustment instruction includes: After visually displaying the current integration result, if feedback is received from the questioner regarding a reply adjustment instruction for the integration result, based on the reply adjustment instruction, an adjustment operation is triggered on the integration result by dragging or prompt word intervention to complete the corresponding large model interaction operation and determine the target reply to the question information.

7. The large-scale interactive question-answering method according to claim 4, characterized in that: The performing multiple verifications on the reply information based on a preset multiple verification mechanism to determine a verification result includes: Performing text alignment and denoising processing on each of the reply messages to obtain a processing result; Performing multi-granularity similarity calculation based on the hybrid similarity algorithm and each processed reply in the processing result, and adjusting the weight value according to the determined similarity result to determine the weight value of each processed reply; Performing a cross-model consensus analysis based on each of the processed responses and the similarity results, and determining whether to trigger an external verification operation based on the consensus analysis result to obtain a determination result; If the judgment result is yes, then numerical verification is performed based on the preset heterogeneous data source, the preset bulldozer distance algorithm and each of the processed responses, and each of the processed responses is adjusted using the numerical verification result to determine an adjusted response; The adjusted response is checked based on a preset logical consistency check strategy to complete multiple verification operations and determine a verification result.

8. A large-scale interactive question-answering device, characterized in that: The preset interactive interface applied to the client includes: The list determination module is used to classify and display the major models in the model library based on the preset filtering panel, and determine the list of major models to be scheduled based on the selection information received from the questioner; a question distribution module, configured to transmit the list of large models to be scheduled and the question information sent by the questioner to the interactive system, so as to distribute the question information to each target large model in the list of large models to be scheduled through the interactive system and a preset distribution mechanism; a response integration module, configured to, after acquiring and displaying the response information corresponding to the question information output by each of the target large models, integrate the response information based on the response fusion instruction received from the questioner to determine an integration result; The reply adjustment module is used to trigger an adjustment operation on the integration result based on the reply adjustment instruction after completing the reply integration, if the reply adjustment instruction is received from the questioner, so as to complete the corresponding large model interaction operation and determine the target reply to the question information.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the large model interactive question-answering method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Used to store a computer program, which, when executed by a processor, implements the large model interactive question-answering method as described in any one of claims 1 to 7.

Images