Question answering method, device and equipment based on multi-model cooperation and storage medium

Abstract

The invention relates to the technical field of large language models, and discloses a question answering method, device and equipment based on multi-model cooperation and a storage medium, the method comprises the following steps: in response to input question information, generating first conclusion information corresponding to the question information through a first large language model; based on the first conclusion information and the question information, generating correction information corresponding to the first conclusion information through a second big language model; and generating and displaying second conclusion information of the question information through a third language model according to the first conclusion information, the question information and the correction information. The question answering accuracy can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of large language model, and particularly relates to a question answering method and device based on multi-model cooperation, equipment and a storage medium. BACKGROUND

[0002] With the rapid development of artificial intelligence and natural language processing technology, large language models (such as GPT series models) have gradually become an important tool in the field of information retrieval. These large models can analyze complex query words input by users, split and recall relevant search results from massive databases or literature. However, although large models have great advantages in understanding and generating language, when dealing with complex search words, only one large language model based on search results to give the answer corresponding to the search word often leads to low accuracy of the answer.

[0003] Therefore, how to effectively improve the accuracy of question answering is a problem to be solved at present.

[0004] The above content is only used to assist in understanding the technical solutions of the present application, and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0005] The main purpose of the present application is to provide a question answering method, device, equipment and storage medium based on multi-model cooperation, aiming at solving the technical problem of how to effectively improve the accuracy of question answering.

[0006] To achieve the above purpose, the present application provides a question answering method based on multi-model cooperation, which comprises:

[0007] In response to the input question information, a first conclusion information corresponding to the question information is generated by a first large language model;

[0008] Based on the first conclusion information and the question information, a correction information corresponding to the first conclusion information is generated by a second large language model;

[0009] According to the first conclusion information, the question information and the correction information, a second conclusion information of the question information is generated and displayed by a third large language model.

[0010] In an embodiment, the correction information corresponding to the first conclusion information is generated by the second large language model based on the first conclusion information and the question information, comprising:

[0011] The semantic fusion analysis is performed on the first conclusion information and the question information to obtain the ambiguous content in the first conclusion information;

[0012] generate, based on the ambiguous content, the first conclusion information corresponding to the correction information by using a second large language model.

[0013] In an embodiment, the semantic fusion analysis of the first conclusion information and the question information obtains the ambiguous content in the first conclusion information, including:

[0014] extracting the search keywords in the question information;

[0015] performing semantic relevance analysis based on the search keywords to determine the information correlation degree of each search keyword and the first conclusion information;

[0016] determining the target search keyword according to the information correlation degree;

[0017] performing deep semantic analysis on the first conclusion information according to the target search keyword to obtain lexical ambiguity information, syntactic ambiguity information, and pragmatic ambiguity information;

[0018] determining the ambiguous content in the first conclusion information according to the lexical ambiguity information, the syntactic ambiguity information, and the pragmatic ambiguity information.

[0019] In an embodiment, the generation of the first conclusion information corresponding to the correction information by using the second large language model based on the ambiguous content includes:

[0020] constructing an ambiguity analysis model, and analyzing the ambiguous content by using the ambiguity analysis model to generate the correction strategy corresponding to the ambiguous content;

[0021] generating the first conclusion information corresponding to the correction information by using the second large language model according to the ambiguous content and the corresponding correction strategy.

[0022] In an embodiment, the generation of the first conclusion information corresponding to the correction information by using the second large language model based on the first conclusion information and the question information further includes:

[0023] performing content analysis on the first conclusion information and the question information to obtain the error content in the first conclusion information;

[0024] generating the first conclusion information corresponding to the correction information by using the second large language model based on the error content.

[0025] In an embodiment, the generation and display of the second conclusion information of the question information by using the third large language model according to the first conclusion information, the question information, and the correction information include:

[0026] correct the first conclusion information based on the problem information and the correction information by a third large language model to generate second conclusion information of the problem information;

[0027] display the second conclusion information of the problem information as an answer to the problem information, and display the answer.

[0028] In an embodiment, before the step of generating the correction information corresponding to the first conclusion information based on the first conclusion information and the problem information by a second large language model, the method further comprises:

[0029] performing quality evaluation based on the first conclusion information to obtain a quality evaluation index;

[0030] determining a quality evaluation score of the first conclusion information based on the quality evaluation index;

[0031] if the quality evaluation score does not reach a preset score threshold, performing the step of generating the correction information corresponding to the first conclusion information based on the first conclusion information and the problem information by a second large language model.

[0032] In an embodiment, the step of performing quality evaluation based on the first conclusion information to obtain a quality evaluation index comprises:

[0033] performing content quality evaluation based on the first conclusion information to obtain content quality information;

[0034] performing text quality evaluation based on the first conclusion information to obtain text quality information;

[0035] determining a quality evaluation index according to the content quality information and the text quality information.

[0036] In an embodiment, the step of performing content quality evaluation based on the first conclusion information to obtain content quality information comprises:

[0037] determining a question-answer matching degree and a number of co-linear words of questions and answers based on the problem information and the first conclusion information;

[0038] determining answer accuracy information according to the question-answer matching degree and the number of co-linear words of questions and answers;

[0039] determining an answer length, information entropy, and a number of complete sentences based on the first conclusion information;

[0040] determining an amount of answer information according to the answer length, the information entropy, and the number of complete sentences;

[0041] performing content quality evaluation according to the answer accuracy information and the amount of answer information to obtain content quality information.

[0042] In an embodiment, the text quality evaluation based on the first conclusion information to obtain text quality information comprises:

[0043] performing semantic analysis on the first conclusion information to obtain semantic clarity;

[0044] performing syntax structure analysis on the first conclusion information to obtain text complexity;

[0045] performing text quality evaluation according to the semantic clarity and the text complexity to obtain text quality information.

[0046] In addition, to achieve the above-mentioned purpose, the application further provides a question answering device based on multi-model cooperation, which comprises:

[0047] a generation module configured to generate first conclusion information corresponding to the question information through a first large language model in response to input question information;

[0048] The generation module is further configured to generate correction information corresponding to the first conclusion information through a second large language model based on the first conclusion information and the question information;

[0049] a display module configured to generate and display second conclusion information of the question information through a third large language model according to the first conclusion information, the question information and the correction information.

[0050] In an embodiment, the generation module is further configured to perform semantic fusion analysis on the first conclusion information and the question information to obtain ambiguous content in the first conclusion information;

[0051] generate correction information corresponding to the first conclusion information through a second large language model based on the ambiguous content.

[0052] In an embodiment, the generation module is further configured to extract search keywords in the question information;

[0053] perform semantic relevance analysis based on the search keywords to determine information correlation degrees of each search keyword and the first conclusion information;

[0054] determine target search keywords according to the information correlation degrees;

[0055] perform deep semantic analysis on the first conclusion information according to the target search keywords to obtain lexical ambiguity information, syntactic ambiguity information and pragmatic ambiguity information;

[0056] determine ambiguous content in the first conclusion information according to the lexical ambiguity information, the syntactic ambiguity information and the pragmatic ambiguity information.

[0057] In an embodiment, the generating module is further configured to construct an ambiguity resolution model, and resolve the ambiguous content through the ambiguity resolution model to generate a correction strategy corresponding to the ambiguous content.

[0058] According to the ambiguous content and the corresponding correction strategy, the first conclusion information corresponding to the correction information is generated through a second large language model.

[0059] In an embodiment, the generating module is further configured to perform content analysis on the first conclusion information and the question information to obtain error content in the first conclusion information.

[0060] Based on the error content, the first conclusion information corresponding to the correction information is generated through a second large language model.

[0061] In an embodiment, the display module is further configured to correct the first conclusion information based on the question information and the correction information through a third large language model to generate second conclusion information of the question information.

[0062] The second conclusion information of the question information is taken as an answer to the question information, and the answer is displayed.

[0063] In an embodiment, the question answering device based on multi-model cooperation further includes a determining module, configured to perform quality evaluation based on the first conclusion information to obtain a quality evaluation index.

[0064] Based on the quality evaluation index, a quality evaluation score of the first conclusion information is determined.

[0065] In a case where the quality evaluation score does not reach a preset score threshold, the step of generating the first conclusion information corresponding to the correction information through a second large language model based on the first conclusion information and the question information is performed.

[0066] In an embodiment, the determining module is further configured to perform content quality evaluation based on the first conclusion information to obtain content quality information.

[0067] Text quality evaluation is performed based on the first conclusion information to obtain text quality information.

[0068] According to the content quality information and the text quality information, a quality evaluation index is determined.

[0069] In addition, to achieve the above object, the present application also provides a question answering device based on multi-model cooperation, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the question answering method based on multi-model cooperation as described above.

[0070] In addition, to achieve the above object, the present application also provides a storage medium, which is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the question answering method based on multi-model cooperation as described above.

[0071] In addition, to achieve the above object, the present application also provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the question answering method based on multi-model cooperation as described above.

[0072] The present application provides a question answering method based on multi-model cooperation, which first generates first conclusion information corresponding to the problem information through a first large language model in response to the input problem information; generates correction information corresponding to the first conclusion information through a second large language model based on the first conclusion information and the problem information; and generates and displays second conclusion information of the problem information through a third large language model according to the first conclusion information, the problem information and the correction information, which can effectively improve the accuracy of question answering.

[0073] In summary, the present application generates first conclusion information corresponding to the problem information through a first large language model, which can quickly respond to the problem and provide a preliminary answer, generates correction information corresponding to the first conclusion information through a second large language model, which can further improve the accuracy and relevance of the answer, and then accurately corrects the first conclusion information according to the correction information through a third large language model to generate more perfect and accurate second conclusion information, thereby providing more satisfactory answers for users. Through the cooperation of multiple large language models, the answer to the question is quickly and accurately generated, which overcomes the technical defect of low accuracy of answering complex search requests, and can effectively improve the accuracy of question answering. BRIEF DESCRIPTION OF DRAWINGS

[0074] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, for those of ordinary skill in the art, the other drawings can also be obtained based on these drawings without any creative work.

[0076] Figure 1 The flowchart provided by the first embodiment of the problem answering method based on multi-model cooperation of the present application;

[0077] Figure 2 The flowchart provided by the second embodiment of the problem answering method based on multi-model cooperation of the present application;

[0078] Figure 3 The flowchart provided by the third embodiment of the problem answering method based on multi-model cooperation of the present application;

[0079] Figure 4 The module structure diagram of the problem answering device based on multi-model cooperation of the embodiments of the present application;

[0080] Figure 5 The device structure diagram of the hardware running environment involved in the problem answering method based on multi-model cooperation of the embodiments of the present application.

[0081] The object implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0082] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application, and are not used to limit the present application.

[0083] In order to better understand the technical solutions of the present application, the following will be described in detail in combination with the drawings in the specification and specific embodiments.

[0084] The main solution of the embodiments of the present application is: in response to the input question information, the first conclusion information corresponding to the question information is generated through the first large language model; the correction information corresponding to the first conclusion information is generated through the second large language model based on the first conclusion information and the question information; the second conclusion information of the question information is generated and displayed through the third large language model according to the first conclusion information, the question information and the correction information.

[0085] With the rapid development of artificial intelligence and natural language processing technology, large language models such as the GPT series of models have gradually become important tools in the field of information retrieval. These large models can analyze complex query words input by users, split them, and recall relevant search results from massive databases or literature. However, although large models have great advantages in understanding and generating language, when dealing with complex search words, simply giving an answer corresponding to the search words based on search results through a large language model often leads to low accuracy of the answer. Therefore, how to effectively improve the accuracy of question answering is a problem that needs to be solved at present.

[0086] The present application generates first conclusion information corresponding to question information through a first large language model, can quickly respond to questions and provide preliminary answers, generates correction information corresponding to the first conclusion information through a second large language model, can further improve the accuracy and relevance of the answers, and then generates more perfect and accurate second conclusion information through a third large language model by accurately modifying the first conclusion information according to the correction information, thereby providing more satisfactory answers for users. Through the cooperation of multiple large language models, the answer to the question is quickly and accurately generated, overcoming the technical defect of low accuracy of answers to complex search requests, and effectively improving the accuracy of question answering.

[0087] It should be noted that the execution subject of the present embodiment can be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a question answering device based on multi-model cooperation, etc. The present embodiment and each of the following embodiments will be described below with the question answering device based on multi-model cooperation as an example.

[0088] Based on this, the present embodiment provides a question answering method based on multi-model cooperation, which refers to Figure 1 , Figure 1 is a flowchart of the first embodiment of the question answering method based on multi-model cooperation of the present application.

[0089] In the present embodiment, the question answering method based on multi-model cooperation includes steps S10-S50:

[0090] Step S10, in response to input question information, generates first conclusion information corresponding to the question information through a first large language model.

[0091] It should be noted that the question information is the question input by the user, which can be a query in the form of text or text information converted from voice input. The present embodiment does not make specific limitations on this, and the present embodiment will be described below with the question information as an example of a text query request input by the user through an interface.

[0092] It can be understood that the first large language model can be a pre-trained model, such as GPT, BERT, etc., which can understand the semantics of the question and generate a preliminary answer. In the present embodiment, the training data set of the first large language model can include a large number of question-answer pairs, so that the model can learn how to extract key information from the question and generate relevant conclusion information.

[0093] In a specific implementation, the text query request input by the user through the interface is input as question information into the first large language model, and the first large language model parses the question according to the pattern in its training data set and generates a preliminary answer, so as to quickly respond to the user's demand and provide a preliminary conclusion information, for example, for the question "What is the fastest subway line in Wuhan in the past 10 years?", the corresponding first conclusion information is "The fastest subway line in Wuhan in the past 10 years is subway line 7".

[0094] Step S20, generating the correction information corresponding to the first conclusion information through the second large language model based on the first conclusion information and the question information.

[0095] It should be noted that the second large language model can also be a pre-trained model, which functions to further improve the accuracy and relevance of the answer.

[0096] It can be understood that after receiving the first conclusion information and the question information, the second large language model will analyze and correct the preliminary conclusion using the pattern in its training data set, to ensure that the answer is more in line with the actual needs of the user, for example, if the preliminary conclusion information has ambiguity or errors, the second large language model can identify these problems and provide more accurate and detailed correction information.

[0097] In a feasible implementation, the step S20 can include: performing content analysis on the first conclusion information and the question information to obtain error content in the first conclusion information; generating the correction information corresponding to the first conclusion information through the second large language model based on the error content.

[0098] It should be noted that in the first conclusion information, the error content can include but is not limited to problems such as ambiguous semantics, factual errors, and incoherent logic. The content analysis on the first conclusion information and the question information to obtain the error content in the first conclusion information specifically includes

[0099] It can be understood that the second large language model can identify the part that needs to be corrected by analyzing these error contents, and generate correction information using the patterns in its training data set. If there is a factual error in the preliminary conclusion information, the second large language model will correct the error fact according to the correct information in its training data set, for example, for the above example, the preliminary conclusion that "the line with the fastest average running speed in Wuhan subway in the past 10 years is subway line 7", the second large language model may analyze that there is a factual error in the conclusion information. Because according to the latest data, actually the average running speed of subway line 2 may be faster. Therefore, the second large language model will generate correction information according to the correct information in its training data set, for example, "according to the latest data, the line with the fastest average running speed in Wuhan subway in the past 10 years is subway line 2". If there is a problem of logical inconsistency, the model will provide more reasonable conclusion information according to the logical relationship between the problem information and the first conclusion information, for example, if there is a problem of logical inconsistency in the preliminary conclusion information, the second large language model will analyze the logical relationship between the problem information and the first conclusion information, and then provide more reasonable conclusion information. For example, if the preliminary conclusion information is "the growth cycle of the plant is 6 months", and the problem information asks "whether the plant is suitable for planting in cold areas", the second large language model will analyze the relationship between the growth cycle of the plant and the planting conditions in cold areas according to the relevant knowledge in its training data set, and then generate correction information such as "considering the growth cycle of the plant is 6 months, it may not be suitable for planting in cold areas, because the area may not be able to provide enough warm season for it to complete its life cycle." In this way, the second large language model can provide more accurate and relevant correction information to meet the user's expectations for the quality of the answer.

[0100] In step S30, a second conclusion information of the problem information is generated and displayed by a third large language model according to the first conclusion information, the problem information and the correction information.

[0101] It should be noted that after receiving the first conclusion information, the problem information and the correction information, the third large language model will integrate these information and correct the first conclusion information using the patterns in its training data set. The purpose of the third large language model is to ensure that the generated answer is not only accurate, but also more fluent and natural in expression, to meet the high requirements of users for the quality of the answer.

[0102] Specifically, the third large language model can identify any inconsistencies or areas that need further clarification in the first conclusion information and the correction information by analyzing them, and generate a more refined and accurate second conclusion information in combination with the question information. For example, if the correction information indicates that certain details in the first conclusion information need to be adjusted, the third large language model will adjust the answer according to these details to ensure that the final answer provided is both accurate and comprehensive.

[0103] It is worth noting that the present embodiment effectively improves the accuracy of question answering through multi-model collaboration, which includes at least the first large language model, the second large language model, and the third large language model, and can also include more language models to achieve deeper collaboration and optimization. For example, a fourth large language model can be introduced to handle specific types of questions, and the present embodiment does not make specific limitations. The first large language model, the second large language model, and the third large language model in the present embodiment can be integrated as sub-models in the same model or run independently, and they exchange data and collaborate through different interfaces.

[0104] In a feasible embodiment, the step S30 can include: correcting the first conclusion information based on the question information and the correction information through a third large language model to generate second conclusion information of the question information; taking the second conclusion information of the question information as an answer to the question information, and displaying the answer.

[0105] It should be noted that the correction process of the third large language model includes optimization of the syntax, word choice, semantic coherence, etc. of the first conclusion information. For example, if the first conclusion information has grammatical flaws, the third large language model will adjust the sentence structure according to the patterns in the training data set of the language model, so that it is more consistent with the grammatical rules and natural language expression habits. If the word choice is not accurate or professional enough, the model will replace it according to the context and professional terminology library to improve the professionalism and accuracy of the answer. In addition, if the first conclusion information is not coherent in semantics, the third large language model will reorganize the information to ensure the logicality and orderliness of the answer, so as to provide a more perfect and easy-to-understand answer to the user. Ultimately, through the collaboration and optimization of multiple models, the quality of question answering and user experience can be significantly improved.

[0106] The embodiment provides a question answering method based on multi-model cooperation. The embodiment can effectively improve the accuracy of question answering by first responding to input question information, generating first conclusion information corresponding to the question information through a first large language model, generating correction information corresponding to the first conclusion information through a second large language model based on the first conclusion information and the question information, and generating and displaying second conclusion information of the question information through a third large language model according to the first conclusion information, the question information and the correction information.

[0107] In conclusion, the embodiment can quickly respond to a question and provide a preliminary answer by generating first conclusion information corresponding to question information through a first large language model, further improve the accuracy and relevance of an answer by generating correction information corresponding to the first conclusion information through a second large language model, and then accurately correct the first conclusion information according to the correction information through a third large language model to generate more perfect and accurate second conclusion information, thereby providing a more satisfactory answer for a user. The multi-model cooperation can quickly and accurately generate an answer to a question, overcome the technical defect of low accuracy of answers to complex search requests, and effectively improve the accuracy of question answering.

[0108] Based on the first embodiment of the application, the same or similar content as or to the above-mentioned first embodiment can be referred to in the above description, and will not be described again. On this basis, please refer to Figure 2 , the step S20 further includes steps S201-S202:

[0109] Step S201, performing semantic fusion analysis on the first conclusion information and the question information to obtain ambiguous content in the first conclusion information.

[0110] It should be noted that ambiguous content refers to information segments that have multiple possible interpretations in semantics, including lexical ambiguity, syntactic ambiguity and pragmatic ambiguity, which are not specifically limited in the embodiment.

[0111] It can be understood that semantic fusion analysis includes using natural language processing technology to perform in-depth analysis on the text to identify and analyze ambiguous content, and then using context information to disambiguate the ambiguous content to ensure the accuracy of understanding. Combined with the semantic analysis result, the first conclusion information is adjusted and optimized as necessary to eliminate the influence of ambiguity, thereby improving the accuracy of question answering.

[0112] In an implementation, the step S201 specifically comprises: extracting search keywords in the question information; performing semantic correlation analysis based on the search keywords to determine information correlation degrees of the search keywords and the first conclusion information; determining target search keywords according to the information correlation degrees; performing deep semantic analysis on the first conclusion information according to the target search keywords to obtain lexical ambiguity information, syntactic ambiguity information and pragmatic ambiguity information; and determining ambiguous content in the first conclusion information according to the lexical ambiguity information, the syntactic ambiguity information and the pragmatic ambiguity information.

[0113] It should be noted that extracting search keywords in the question information refers to identifying key words or phrases from the question information, which can represent the core meaning or requirements of the question. For example, if the question is “how to treat a cold”, “treat” and “cold” are search keywords. Next, semantic correlation analysis is performed based on these keywords, that is, the strength of the connection between these keywords and the first conclusion information is evaluated. The determination of information correlation degrees can be based on factors such as the frequency, position and context environment of keywords in the first conclusion information. Through such analysis, the most relevant search keywords to the question, i.e. target search keywords, can be screened out.

[0114] It can be understood that deep semantic analysis is performed on the first conclusion information using the target search keywords. Deep semantic analysis may involve natural language processing techniques such as word sense disambiguation, syntactic structure analysis and context understanding, etc. to identify content with lexical ambiguity, syntactic ambiguity and pragmatic ambiguity. For example, lexical ambiguity may occur in the word “apple”, which can refer to a fruit or a technology company; syntactic ambiguity may occur in sentence structure, such as “he saw the person and the dog”, which can be interpreted as “he saw the person, and the person has a dog”, or as “he saw the person and the dog”; pragmatic ambiguity may be related to a specific context, such as “can you open the window? ” may represent a request or a command in different contexts.

[0115] The step S202 generates the corrected information corresponding to the first conclusion information based on the ambiguous content through a second large language model.

[0116] It should be noted that the second large language model can generate correction information related to the ambiguous content according to the context information and the training data of the language model. For example, if there is lexical ambiguity in the first conclusion information, the language model will select the most appropriate word meaning according to the context clues; for syntactic ambiguity, the model will infer the most reasonable sentence structure according to the structure and grammar rules of the sentence; for pragmatic ambiguity, the model will consider specific contextual factors to determine the most appropriate expression. In this way, the language model can provide more accurate and contextually appropriate correction information, thereby improving the accuracy and reliability of the problem answer.

[0117] In a feasible implementation, step S202 specifically includes: constructing an ambiguity resolution model, and resolving the ambiguous content through the ambiguity resolution model to generate a correction strategy corresponding to the ambiguous content; and generating correction information corresponding to the first conclusion information through the second large language model according to the ambiguous content and the corresponding correction strategy.

[0118] It should be noted that the process of constructing the ambiguity resolution model involves classifying and identifying different types of ambiguity, and then designing corresponding resolution algorithms. For example, for lexical ambiguity, the model needs a word sense disambiguation module that can determine the correct meaning of a word based on context; for syntactic ambiguity, the model needs a syntax analyzer that can analyze sentence structure and determine the most likely grammatical interpretation; for pragmatic ambiguity, the model needs a context analyzer that can infer the most appropriate expression intent based on the history and context of the conversation. These resolution modules will work together to ensure accurate resolution and understanding of ambiguous content.

[0119] It can be understood that the correction strategy refers to the processing scheme or method adopted for a specific ambiguity type. For example, for lexical ambiguity, the correction strategy may include selecting the most common or most relevant word sense; for syntactic ambiguity, the strategy may involve restructuring the sentence to eliminate ambiguity; and for pragmatic ambiguity, the strategy may include adjusting the tone of the sentence or adding additional contextual information to clarify the intent. These strategies will be used to guide the second large language model to generate correction information, ensuring that the corrected information can more accurately reflect the true intent and context of the problem. In this way, understanding problems caused by ambiguity can be effectively solved, thereby providing higher quality problem solving.

[0120] Specifically, when generating the correction information, the second large language model can use the correction strategy provided by the ambiguity resolution model, combined with its deep understanding of the language and rich training data, to generate more accurate and targeted correction information. For example, if there is ambiguity in the word "apple" in the first conclusion information, the language model will select "fruit" or "technology company" as the correct word meaning according to the context clues, and generate correction information accordingly. Similarly, for syntactic ambiguity and pragmatic ambiguity, the model can generate correction information that reflects the actual context and intent, providing more accurate and useful answers. In this way, the quality and user experience of question answering can be significantly improved.

[0121] In this embodiment, by performing semantic fusion analysis on the first conclusion information and the question information, the ambiguous content in the first conclusion information can be quickly and accurately located, and then the corresponding correction information can be generated according to the corresponding correction strategy of the ambiguous content, further improving the accuracy of question answering.

[0122] It should be noted that the above examples are only for understanding the present application and do not constitute a limitation on the question answering method based on multi-model cooperation of the present application. Further simple transformations based on this technical concept are within the scope of protection of the present application.

[0123] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar contents as the above embodiment one can refer to the above introduction, and the following will not be repeated. On this basis, please refer to Figure 3 , before the step S20, further comprising steps S01-S03:

[0124] Step S01, based on the first conclusion information, quality evaluation is performed to obtain a quality evaluation index.

[0125] It should be noted that quality evaluation refers to quantitative analysis of the content and text of the first conclusion information to determine its accuracy and reliability. The quality evaluation index refers to a parameter for measuring the quality of the first conclusion information, such as content quality and text quality, which is not specifically limited in this embodiment.

[0126] In a feasible implementation manner, the step S01 can include steps S010-S012:

[0127] Step S010, based on the first conclusion information, content quality evaluation is performed to obtain content quality information.

[0128] It should be noted that content quality evaluation refers to evaluating the semantic depth and relevance of the first conclusion information to ensure that it is highly relevant to the question and rich in information.

[0129] It can be understood that the content quality information refers to a quantitative description of the depth and breadth of the first conclusion information at the semantic level, which reflects the degree of conclusion information for problem solving and the degree of information detail. The content quality information can be quantified by the answer accuracy and the answer information amount. For example, the answer accuracy can be evaluated by the matching degree with the standard answer, and the answer information amount can be measured by the number of keywords and the richness of information points contained in the answer.

[0130] In an implementable embodiment, step S010 specifically comprises: determining the question-answer matching degree and the number of co-linear words based on the question information and the first conclusion information; determining the answer accuracy information according to the question-answer matching degree and the number of co-linear words; determining the answer length, the information entropy, and the number of complete sentences based on the first conclusion information; determining the answer information amount according to the answer length, the information entropy, and the number of complete sentences; and performing content quality evaluation according to the answer accuracy information and the answer information amount to obtain the content quality information.

[0131] It should be noted that the question-answer matching degree is an index for measuring the correlation degree between the question and the answer, which reflects the direct response degree of the answer to the question. The question-answer matching degree can be calculated by a semantic similarity algorithm. The number of co-linear words refers to the number of keywords that appear in both the question and the answer, which helps to evaluate the relevance of the answer to the question. By comprehensively considering the question-answer matching degree and the number of co-linear words, the accuracy information of the answer can be more accurately evaluated. For example, a vector space model or a cosine similarity algorithm can be used for calculation.

[0132] It can be understood that the answer length, the information entropy, and the number of complete sentences reflect the information amount of the answer from different angles. The answer length refers to the number of words or characters in the answer text, which reflects the detail degree of the answer. The information entropy is an index for measuring the information richness, which evaluates the diversity of information by calculating the uncertainty of information in the answer. The number of complete sentences directly reflects the structural integrity of the answer. By comprehensively considering these parameters, the information amount of the answer can be more comprehensively evaluated. For example, a longer answer length and a higher information entropy indicate that the answer provides more detailed and diverse information. A larger number of complete sentences indicates that the answer is more complete in structure and can better express a complete idea or concept.

[0133] Step S011 performs text quality evaluation based on the first conclusion information to obtain text quality information.

[0134] It should be noted that the text quality evaluation focuses on the clarity of the expression of the conclusion information, the grammatical correctness and whether it conforms to the use of professional terms, to ensure that the expression of the conclusion information is both accurate and easy to understand. The text quality information is an evaluation of the language expression level of the first conclusion information, including semantic clarity and text complexity, which reflects the readability and professionalism of the conclusion information.

[0135] In an implementable embodiment, step S011 specifically comprises: performing semantic analysis on the first conclusion information to obtain semantic clarity; performing syntactic structure analysis on the first conclusion information to obtain text complexity; and performing text quality evaluation according to the semantic clarity and the text complexity to obtain text quality information.

[0136] It should be noted that the semantic clarity is evaluated by analyzing the semantic coherence and logicality of the conclusion information, which can reflect whether the information is easy to understand. The evaluation of semantic clarity can be determined by the grammatical structure and semantic role labeling of the sentence.

[0137] It can be understood that the text complexity involves the sentence structure and vocabulary difficulty of the conclusion information, which can be evaluated by calculating the frequency of using professional terms and the use of complex sentence patterns. The reduction of text complexity helps to improve the readability of the information, so that non-professional readers can also understand the conclusion information. By comprehensively considering the semantic clarity and text complexity, the text quality can be comprehensively evaluated to obtain the text quality information.

[0138] Step S012 determines a quality evaluation index according to the content quality information and the text quality information.

[0139] It should be noted that the quality evaluation index is composed of the content quality information and the text quality information, which can comprehensively reflect the overall quality of the first conclusion information.

[0140] Step S02 determines a quality evaluation score of the first conclusion information based on the quality evaluation index.

[0141] It should be noted that the quality evaluation score is a quantitative index, which combines the content quality information and the text quality information to provide a comprehensive evaluation of the first conclusion information. The quality evaluation score can be used to compare the quality of different conclusion information, and can also be used as a basis for improving and optimizing the conclusion information. The quality evaluation score can be expressed in percentage or grade system, which is convenient for users to quickly understand the pros and cons of the conclusion information. Through such an evaluation system, it can be ensured that the conclusion information not only has accurate content and sufficient information, but also has clear expression and easy understanding, so as to meet the needs of different users.

[0142] It can be understood that the quality evaluation score of the first conclusion information is determined according to the quality evaluation index, that is, a comprehensive score is calculated according to the content quality information and the text quality information, so as to obtain a quantitative quality evaluation result. The calculation of the quality evaluation score can adopt a weighted average method, wherein the content quality information and the text quality information are given different weights according to their importance in the overall evaluation. For example, the content quality information can be more important, and therefore a higher weight can be given in the calculation of the score. Finally, the quality evaluation score can provide a direct indicator for the user, which helps to quickly judge the reliability and professionalism of the conclusion information.

[0143] In step S03, if the quality evaluation score does not reach the preset score threshold, a step of generating modified information corresponding to the first conclusion information based on the first conclusion information and the question information by a second large language model is performed.

[0144] It should be noted that if the quality evaluation score does not reach the preset score threshold, it indicates that the first conclusion information may not be perfect or accurate in some aspects and needs to be further modified. In this case, the modification process is started and the second large language model is used to analyze and process the original conclusion information. In this way, more accurate and reliable modified information can be generated to improve the overall quality of the conclusion information. The generation of the modified information will be based on the question information and the first conclusion information to ensure that the modified information can better meet the needs and expectations of the user.

[0145] It can be understood that if the quality evaluation score reaches the preset score threshold, it indicates that the quality of the first conclusion information has met the standard and does not need to be further modified. In this case, the first conclusion information can be directly provided to the user as an answer. The quality evaluation score as an evaluation tool not only helps to ensure the accuracy of the information, but also guides the further optimization and improvement of the information. Through such a quality control process, the usability and value of the information can be effectively improved to meet the information needs of the user in different scenarios.

[0146] It should be noted that the setting of the preset score threshold needs to be carefully considered according to the specific field and application scenario. For example, in the medical field, the preset score threshold may need to be set higher to ensure that the provided conclusion information has very high accuracy and reliability. In the daily consultation or entertainment field, the preset score threshold can be relatively relaxed, because the accuracy requirement of information in these scenarios can not be so strict. By flexibly adjusting the preset score threshold, it can be ensured that the quality evaluation system is strict and adaptive, meeting the specific needs in different fields and scenarios.

[0147] In this embodiment, the quality evaluation score of the first conclusion information is quickly and accurately determined according to the quality evaluation index of the first conclusion information, and then it is determined whether the first conclusion information needs to be corrected, thereby ensuring the high-quality output of the conclusion information. The process based on the quality evaluation score not only improves the work efficiency, but also enhances the trust of the user on the conclusion information.

[0148] It should be noted that the above examples are only used for understanding the present application and do not constitute a limitation on the problem answering method based on multi-model cooperation of the present application. Further simple transformations in more forms based on this technical concept are within the protection scope of the present application.

[0149] The present application also provides a problem answering device based on multi-model cooperation, which is described in detail with reference to Figure 4 , and the problem answering device based on multi-model cooperation comprises:

[0150] The generating module 10 is configured to generate first conclusion information corresponding to the input question information through a first large language model.

[0151] The generating module 10 is also configured to generate correction information corresponding to the first conclusion information through a second large language model based on the first conclusion information and the question information.

[0152] The display module 20 is configured to generate and display second conclusion information of the question information through a third large language model based on the first conclusion information, the question information and the correction information.

[0153] The present embodiment provides a problem answering device based on multi-model cooperation. The present embodiment generates first conclusion information corresponding to the input question information through a first large language model, generates correction information corresponding to the first conclusion information through a second large language model based on the first conclusion information and the question information, and generates and displays second conclusion information of the question information through a third large language model based on the first conclusion information, the question information and the correction information, thereby effectively improving the accuracy of problem answering.

[0154] As can be seen from the above, the first large language model is used to generate first conclusion information corresponding to the question information, which can quickly respond to the question and provide a preliminary answer. The second large language model is used to generate correction information corresponding to the first conclusion information, which can further improve the accuracy and relevance of the answer. Then, the third large language model is used to accurately correct the first conclusion information based on the correction information, thereby generating more perfect and accurate second conclusion information, and providing more satisfactory answers for the user. Through the cooperation of multiple large language models, the answer to the question is quickly and accurately generated, thereby overcoming the technical defect of low accuracy of the answer to the complex search request, and effectively improving the accuracy of problem answering.

[0155] Optionally, the generating module 10 is further configured to perform semantic fusion analysis on the first conclusion information and the question information to obtain ambiguous content in the first conclusion information; and generate, based on the ambiguous content, correction information corresponding to the first conclusion information by using a second large language model.

[0156] Optionally, the generating module 10 is further configured to extract search keywords in the question information; perform semantic relevance analysis based on the search keywords to determine information correlation degrees of the search keywords and the first conclusion information; determine a target search keyword according to the information correlation degrees; perform deep semantic analysis on the first conclusion information according to the target search keyword to obtain lexical ambiguity information, syntactic ambiguity information and pragmatic ambiguity information; and determine ambiguous content in the first conclusion information according to the lexical ambiguity information, the syntactic ambiguity information and the pragmatic ambiguity information.

[0157] Optionally, the generating module 10 is further configured to construct an ambiguity resolution model, resolve the ambiguous content by using the ambiguity resolution model, and generate a correction strategy corresponding to the ambiguous content; and generate, based on the ambiguous content and the corresponding correction strategy, correction information corresponding to the first conclusion information by using a second large language model.

[0158] Optionally, the generating module 10 is further configured to perform content analysis on the first conclusion information and the question information to obtain error content in the first conclusion information; and generate, based on the error content, correction information corresponding to the first conclusion information by using a second large language model.

[0159] Optionally, the display module 30 is further configured to correct the first conclusion information based on the question information and the correction information by using a third large language model to generate second conclusion information of the question information; display the second conclusion information of the question information as an answer to the question information; and display the answer.

[0160] Optionally, the question answering device based on multi-model cooperation further comprises a determining module; the determining module is configured to perform quality evaluation based on the first conclusion information to obtain a quality evaluation index; determine a quality evaluation score of the first conclusion information based on the quality evaluation index; and in a case where the quality evaluation score does not reach a preset score threshold, perform the step of generating, based on the first conclusion information and the question information, correction information corresponding to the first conclusion information by using a second large language model.

[0161] Optionally, the determining module is further configured to perform content quality evaluation based on the first conclusion information to obtain content quality information, perform text quality evaluation based on the first conclusion information to obtain text quality information, and determine a quality evaluation index according to the content quality information and the text quality information.

[0162] Optionally, the determining module is further configured to determine a question-answer matching degree and a number of co-linear words of questions and answers based on the question information and the first conclusion information, determine answer accuracy information according to the question-answer matching degree and the number of co-linear words of questions and answers, determine an answer length, information entropy, and a number of complete sentences based on the first conclusion information, determine an answer information amount according to the answer length, the information entropy, and the number of complete sentences, and perform content quality evaluation according to the answer accuracy information and the answer information amount to obtain content quality information.

[0163] Optionally, the determining module is further configured to perform semantic analysis on the first conclusion information to obtain semantic clarity, perform syntax structure analysis on the first conclusion information to obtain text complexity, and perform text quality evaluation according to the semantic clarity and the text complexity to obtain text quality information.

[0164] The question answering device based on multi-model cooperation provided in the application can solve the technical problem of how to effectively improve the accuracy of question answering by using the question answering method based on multi-model cooperation in the above embodiments. Compared with the prior art, the question answering device based on multi-model cooperation provided in the application has the same beneficial effects as the question answering method based on multi-model cooperation provided in the above embodiments, and other technical features in the question answering device based on multi-model cooperation are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0165] The application provides a question answering device based on multi-model cooperation, which comprises at least one processor and a memory in communication connection with the at least one processor. The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the question answering method based on multi-model cooperation in the above embodiment one.

[0166] The following refers to Figure 5, which shows a schematic diagram of the structure of a multi-model collaboration-based question answering device suitable for implementing an embodiment of the present application. The multi-model collaboration-based question answering device in the embodiment of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The question answering device based on multi-model collaboration shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0167] like Figure 5 As shown, the question answering device based on multi-model collaboration may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. Various programs and data required for the operation of the question answering device based on multi-model collaboration are also stored in RAM1004. The processing device 1001, ROM1002 and RAM1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the question answering device based on multi-model collaboration to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a question answering device based on multi-model collaboration with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.

[0168] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program codes for executing the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network through a communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiments disclosed in the present application are executed.

[0169] The problem answering device based on multi-model collaboration provided in the present application adopts the problem answering method based on multi-model collaboration in the above-mentioned embodiments, and can solve the technical problem of how to effectively improve the accuracy of problem answering. Compared with the prior art, the problem answering device based on multi-model collaboration provided in the present application has the same beneficial effects as the problem answering method based on multi-model collaboration provided in the above-mentioned embodiments, and other technical features in the problem answering device based on multi-model collaboration are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0170] It should be understood that parts of the present application can be realized by hardware, software, firmware or a combination thereof. In the description of the above-mentioned embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0171] The above is merely specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0172] The present application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e. computer program) for executing the problem answering method based on multi-model collaboration in the above-mentioned embodiments.

[0173] The computer readable storage medium provided in the application may, for example, be a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electric connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency), etc., or any suitable combination of the above.

[0174] The above computer readable storage medium can be included in the problem answering device based on multi-model cooperation, or can exist independently without being assembled into the problem answering device based on multi-model cooperation.

[0175] The above computer readable storage medium carries one or more programs, which, when executed by the problem answering device based on multi-model cooperation, cause the problem answering device based on multi-model cooperation to: in response to input question information, generate first conclusion information corresponding to the question information by a first large language model; generate correction information corresponding to the first conclusion information by a second large language model based on the first conclusion information and the question information; and generate and display second conclusion information of the question information by a third large language model according to the first conclusion information, the question information, and the correction information.

[0176] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0177] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0178] The modules involved in the embodiments of the present application can be implemented in the manner of software or hardware. Among them, the name of the module does not constitute a limitation to the unit itself in some cases.

[0179] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e. computer program) for executing the above-mentioned problem answering method based on multi-model cooperation, and can solve the technical problem of how to effectively improve the accuracy of problem answering. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the problem answering method based on multi-model cooperation provided by the above-mentioned embodiments, which will not be described here.

[0180] The application further provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the problem answering method based on multi-model cooperation as described above.

[0181] The computer program product provided by the application can solve the technical problem of how to effectively improve the accuracy of problem answering. Compared with the prior art, the beneficial effects of the computer program product provided by the application are the same as those of the problem answering method based on multi-model cooperation provided by the above-mentioned embodiments, and are not repeated here.

[0182] The above only describes some embodiments of the application, and does not limit the patent scope of the application. Any equivalent structural transformation made by using the content of the application specification and drawings, or direct / indirect application in other related technical fields is included in the patent protection scope of the application.

[0183] The application discloses A1. A problem answering method based on multi-model cooperation, the method comprising:

[0184] In response to input question information, a first conclusion information corresponding to the question information is generated by a first large language model;

[0185] Based on the first conclusion information and the question information, a correction information corresponding to the first conclusion information is generated by a second large language model;

[0186] According to the first conclusion information, the question information and the correction information, a second conclusion information of the question information is generated and displayed by a third large language model.

[0187] A2. The method of A1, wherein the correction information corresponding to the first conclusion information is generated by a second large language model based on the first conclusion information and the question information, comprising:

[0188] Performing semantic fusion analysis on the first conclusion information and the question information to obtain ambiguous content in the first conclusion information;

[0189] Based on the ambiguous content, the correction information corresponding to the first conclusion information is generated by a second large language model.

[0190] A3. The method of A2, wherein the semantic fusion analysis on the first conclusion information and the question information to obtain ambiguous content in the first conclusion information comprises:

[0191] Extracting search keywords in the question information;

[0192] Based on the search keywords, performing semantic correlation analysis to determine the information correlation degree of each search keyword and the first conclusion information;

[0193] determine a target search keyword according to the information relevance degree;

[0194] perform deep semantic analysis on the first conclusion information according to the target search keyword to obtain lexical ambiguity information, syntactic ambiguity information, and pragmatic ambiguity information;

[0195] determine ambiguous content in the first conclusion information according to the lexical ambiguity information, the syntactic ambiguity information, and the pragmatic ambiguity information.

[0196] A4. The method of A2, wherein the generating, based on the ambiguous content, revised information corresponding to the first conclusion information by a second large language model comprises:

[0197] constructing an ambiguity resolution model and resolving the ambiguous content by the ambiguity resolution model to generate a revision strategy corresponding to the ambiguous content;

[0198] generating, according to the ambiguous content and the corresponding revision strategy, revised information corresponding to the first conclusion information by a second large language model.

[0199] A5. The method of A1, wherein the generating, based on the first conclusion information and the question information, revised information corresponding to the first conclusion information by a second large language model further comprises:

[0200] performing content analysis on the first conclusion information and the question information to obtain erroneous content in the first conclusion information;

[0201] generating, based on the erroneous content, revised information corresponding to the first conclusion information by a second large language model.

[0202] A6. The method of A1, wherein the generating and displaying, according to the first conclusion information, the question information, and the revised information, second conclusion information of the question information by a third large language model comprises:

[0203] revising, based on the question information and the revised information, the first conclusion information by a third large language model to generate second conclusion information of the question information;

[0204] displaying the second conclusion information of the question information as an answer to the question information.

[0205] A7. The method of A1, further comprising, before the generating, based on the first conclusion information and the question information, revised information corresponding to the first conclusion information by a second large language model:

[0206] perform quality evaluation based on the first conclusion information to obtain a quality evaluation index;

[0207] determine a quality evaluation score of the first conclusion information based on the quality evaluation index;

[0208] if the quality evaluation score does not reach a preset score threshold, perform a step of generating, by a second large language model, revised information corresponding to the first conclusion information based on the first conclusion information and the question information.

[0209] A8. The method of A7, wherein the performing quality evaluation based on the first conclusion information to obtain a quality evaluation index comprises:

[0210] performing content quality evaluation based on the first conclusion information to obtain content quality information;

[0211] performing text quality evaluation based on the first conclusion information to obtain text quality information;

[0212] determining a quality evaluation index according to the content quality information and the text quality information.

[0213] A9. The method of A8, wherein the performing content quality evaluation based on the first conclusion information to obtain content quality information comprises:

[0214] determining a question-answer matching degree and a number of co-linear words of questions and answers based on the question information and the first conclusion information;

[0215] determining answer accuracy information according to the question-answer matching degree and the number of co-linear words of questions and answers;

[0216] determining a length of an answer, information entropy, and a number of complete sentences based on the first conclusion information;

[0217] determining an amount of answer information according to the length of the answer, the information entropy, and the number of complete sentences;

[0218] performing content quality evaluation according to the answer accuracy information and the amount of answer information to obtain content quality information.

[0219] A10. The method of A8, wherein the performing text quality evaluation based on the first conclusion information to obtain text quality information comprises:

[0220] performing semantic analysis on the first conclusion information to obtain semantic clarity;

[0221] performing syntax structure analysis on the first conclusion information to obtain text complexity;

[0222] A text quality evaluation is performed based on the semantic clarity and the text complexity to obtain text quality information.

[0223] The present invention also discloses B11. A question answering device based on multi-model collaboration, the question answering device based on multi-model collaboration comprising:

[0224] a generating module, configured to generate, in response to input question information, first conclusion information corresponding to the question information by using a first large language model;

[0225] The generating module is further configured to generate correction information corresponding to the first conclusion information using a second language model based on the first conclusion information and the question information;

[0226] A display module is used to generate and display second conclusion information of the question information through a third language model based on the first conclusion information, the question information and the correction information.

[0227] B12. The apparatus as described in B11, wherein the generating module is further configured to perform semantic fusion analysis on the first conclusion information and the question information to obtain ambiguous content in the first conclusion information;

[0228] Based on the ambiguous content, correction information corresponding to the first conclusion information is generated through a second language model.

[0229] B13. The apparatus as described in B12, wherein the generating module is further configured to extract search keywords from the question information;

[0230] Performing semantic relevance analysis based on the search keywords to determine the information relevance between each of the search keywords and the first conclusion information;

[0231] Determining target search keywords based on the information relevance;

[0232] Performing a deep semantic analysis on the first conclusion information according to the target search keyword to obtain lexical ambiguity information, syntactic ambiguity information, and pragmatic ambiguity information;

[0233] The ambiguous content in the first conclusion information is determined according to the lexical ambiguity information, the syntactic ambiguity information, and the pragmatic ambiguity information.

[0234] B14. The apparatus as described in B12, wherein the generating module is further configured to construct an ambiguity resolution model, and parse the ambiguous content through the ambiguity resolution model to generate a correction strategy corresponding to the ambiguous content;

[0235] Correction information corresponding to the first conclusion information is generated through a second large language model according to the ambiguous content and the corresponding correction strategy.

[0236] B15. The apparatus of B11, and the generating module is further configured to perform content analysis on the first conclusion information and the question information to obtain error content in the first conclusion information.

[0237] generate correction information corresponding to the first conclusion information based on the error content through a second large language model.

[0238] B16. The apparatus of B11, and the displaying module is further configured to correct the first conclusion information based on the question information and the correction information through a third large language model to generate second conclusion information of the question information.

[0239] display the second conclusion information of the question information as an answer to the question information, and display the answer.

[0240] B17. The apparatus of B11, and the question answering device based on multi-model cooperation further comprises a determining module configured to perform quality evaluation based on the first conclusion information to obtain a quality evaluation index.

[0241] determine a quality evaluation score of the first conclusion information based on the quality evaluation index.

[0242] if the quality evaluation score does not reach a preset score threshold, perform the step of generating correction information corresponding to the first conclusion information based on the first conclusion information and the question information through a second large language model.

[0243] B18. The apparatus of B17, and the determining module is further configured to perform content quality evaluation based on the first conclusion information to obtain content quality information.

[0244] perform text quality evaluation based on the first conclusion information to obtain text quality information.

[0245] determine a quality evaluation index according to the content quality information and the text quality information.

[0246] The application also discloses C19. A question answering device based on multi-model cooperation, the question answering device based on multi-model cooperation comprises a memory, a processor and a question answering program based on multi-model cooperation stored on the memory and executable on the processor, and the question answering program based on multi-model cooperation is configured to implement the question answering method based on multi-model cooperation.

[0247] The application further discloses D20. A storage medium, wherein a question answering program based on multi-model cooperation is stored on the storage medium, and the question answering program based on multi-model cooperation realizes the question answering method based on multi-model cooperation when being executed by a processor.

Claims

1. A question answering method based on multi-model collaboration, characterized in that: The method comprises: In response to the input question information, generating first conclusion information corresponding to the question information through the first large language model; generating correction information corresponding to the first conclusion information through a second language model based on the first conclusion information and the question information; The second conclusion information of the question information is generated and displayed through a third language model according to the first conclusion information, the question information and the correction information.

2. The method according to claim 1, wherein Generating correction information corresponding to the first conclusion information by using a second language model based on the first conclusion information and the question information includes: Performing semantic fusion analysis on the first conclusion information and the question information to obtain ambiguous content in the first conclusion information; Based on the ambiguous content, correction information corresponding to the first conclusion information is generated through a second language model.

3. The method according to claim 2, wherein The performing semantic fusion analysis on the first conclusion information and the question information to obtain ambiguous content in the first conclusion information includes: Extracting search keywords from the question information; Performing semantic relevance analysis based on the search keywords to determine the information relevance between each of the search keywords and the first conclusion information; Determining target search keywords based on the information relevance; Performing a deep semantic analysis on the first conclusion information according to the target search keyword to obtain lexical ambiguity information, syntactic ambiguity information, and pragmatic ambiguity information; The ambiguous content in the first conclusion information is determined according to the lexical ambiguity information, the syntactic ambiguity information, and the pragmatic ambiguity information.

4. The method according to claim 2, wherein Generating correction information corresponding to the first conclusion information based on the ambiguous content by using a second language model includes: Constructing an ambiguous resolution model, and parsing the ambiguous content using the ambiguous resolution model to generate a correction strategy corresponding to the ambiguous content; Correction information corresponding to the first conclusion information is generated through a second large language model according to the ambiguous content and the corresponding correction strategy.

5. The method according to claim 1, wherein The step of generating correction information corresponding to the first conclusion information by using a second language model based on the first conclusion information and the question information further includes: Performing content analysis on the first conclusion information and the question information to obtain error content in the first conclusion information; Based on the error content, correction information corresponding to the first conclusion information is generated through a second language model.

6. The method according to claim 1, wherein The generating and displaying the second conclusion information of the question information by the third language model based on the first conclusion information, the question information, and the correction information includes: Based on the question information and the correction information, the first conclusion information is corrected by a third language model to generate second conclusion information of the question information; The second conclusion information of the question information is used as an answer to the question information, and the answer is displayed.

7. The method according to claim 1, wherein Before generating correction information corresponding to the first conclusion information by using a second language model based on the first conclusion information and the question information, the method further includes: Performing quality evaluation based on the first conclusion information to obtain a quality evaluation index; Determining a quality evaluation score for the first conclusion information based on the quality evaluation indicator; If the quality evaluation score does not reach a preset score threshold, a step of generating correction information corresponding to the first conclusion information through a second large language model based on the first conclusion information and the question information is performed.

8. A question answering device based on multi-model collaboration, characterized in that: The question answering device based on multi-model collaboration includes: a generating module, configured to generate, in response to input question information, first conclusion information corresponding to the question information by using a first large language model; The generating module is further configured to generate correction information corresponding to the first conclusion information using a second language model based on the first conclusion information and the question information; A display module is used to generate and display second conclusion information of the question information through a third language model based on the first conclusion information, the question information and the correction information.

9. A question answering device based on multi-model collaboration, characterized in that: The question answering device based on multi-model collaboration includes: a memory, a processor, and a question answering program based on multi-model collaboration stored on the memory and executable on the processor, wherein the question answering program based on multi-model collaboration is configured to implement the question answering method based on multi-model collaboration as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores a question answering program based on multi-model collaboration, and when the question answering program based on multi-model collaboration is executed by the processor, the question answering method based on multi-model collaboration as described in any one of claims 1 to 7 is implemented.

Images