Solution matching method and system for artificial intelligence large model

Abstract

The present invention relates to the field of artificial intelligence, and in particular to a method and system for matching answers to artificial intelligence large models, the method comprising: obtaining the question text asked by the user for the nth time, using natural language processing technology to perform feature extraction processing on the question text, and constructing a question feature set; inputting the question feature set into a difficulty assessment model, and outputting the difficulty score of the question text; based on the difficulty score of the question text, using a large model matching degree formula to calculate the matching degree of all artificial intelligence large models in the large model library, and obtaining the matching degree of each artificial intelligence large model; analyzing the matching degree of each artificial intelligence large model according to a preset large model matching strategy, and obtaining a selection plan. This method can accurately and quantitatively measure the matching degree between questions and various large models, so as to improve the professionalism and accuracy of answering questions; at the same time, it can also improve the ability of the large model to process continuous dialogues.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method and system for matching solutions to large artificial intelligence models. Background Art

[0002] Currently, AI big models (hereafter referred to as big models), such as open-source free big models, GPT, Wenxin Yiyan, and iFlytek Spark, demonstrate tremendous potential in providing technical knowledge solutions, guiding innovation hotspots, and analyzing industrial policies. However, different big models vary in their ability to handle continuous conversations and the quality of their answers. These differences arise from varying training data, optimization methods, and differences between general-purpose and specialized big models.

[0003] Therefore, matching the optimal big model to the user's question is crucial for improving the quality of question answering. Furthermore, when matching big models to user questions, we also need to consider the fact that some open-source big models are free, while others require a token-based fee. Appropriately matching free open-source big models with paid big models to answer user questions can help reduce costs.

[0004] However, existing large models generally have problems such as difficulty in accurately and quantitatively measuring the matching degree between problems and various large models, insufficient ability of a single large model to handle continuous questions, and lack of switching strategies for large models.

[0005] To this end, the present invention provides a solution matching method and system for an artificial intelligence large model. Summary of the Invention

[0006] Based on this, it is necessary to provide an artificial intelligence large model solution matching method and system to address the above technical issues.

[0007] According to a first aspect of the present invention, a method for matching answers to artificial intelligence big models is provided, comprising: S1, obtaining a question text asked by a user for the nth time, performing feature extraction processing on the question text using natural language processing technology, and constructing a question feature set, where n is an integer greater than 0; S2, inputting the question feature set into a difficulty assessment model, and outputting a difficulty score of the question text, wherein the difficulty assessment model is generated after training based on a quantized historical question feature set; S3, based on the difficulty score of the question text, calculating the matching degree of all artificial intelligence big models in a big model library using a big model matching formula to obtain the matching degree of each artificial intelligence big model; S4, analyzing the matching degree of each artificial intelligence big model according to a preset big model matching strategy to obtain a selection scheme, wherein the selection scheme is to select a single artificial intelligence big model or multiple artificial intelligence big models; wherein, when a single artificial intelligence big model is selected, the selected single artificial intelligence big model is used as the optimal matching result of the answer; when multiple artificial intelligence big models are selected, the selected multiple artificial intelligence big models are used as the optimal matching results of the answer.

[0008] In some optional implementations of some embodiments, the step S1 specifically includes: S101, obtaining the question text asked by the user, performing word segmentation processing on the question text based on the word segmentation model and performing part-of-speech tagging on each of the question participles, to obtain multiple question participles and part-of-speech tagging information corresponding to the question participles; S102, using a neural network model to perform entity recognition on the question participles based on the part-of-speech tagging information corresponding to the question participles, and determine the entity of at least one key technology; S103, performing semantic analysis processing on the entity of the at least one key technology based on the semantic analysis model, and determining the technical field information to which the entity of the key technology belongs; S104, constructing a question feature set based on the entity of the key technology and the technical field information to which it belongs.

[0009] In some optional implementations of some embodiments, step S1 also includes: S105, quantifying the entities of the key technologies contained in the problem feature set and the information of the technical fields to which they belong, to obtain technical depth information and technical concept breadth information, wherein the technical depth information is determined by the hierarchical relationship of the entities of multiple key technologies, and the technical concept breadth is determined by counting the number of information of the technical fields to which they belong.

[0010] In some optional implementations of some embodiments, in step S3, the calculation formula of the large model matching algorithm is as follows: Where: represents the matching degree of the jth artificial intelligence large model in the large model library, D represents the difficulty score of the first question text, represents the difficulty processing capability score of the jth artificial intelligence large model in the large model library, and is the known difficulty processing ability score data, and | | represents the absolute value.

[0011] In some optional implementations of some embodiments, the preset large model matching strategy in step S4 includes: selecting a single artificial intelligence large model with the best matching degree; and selecting multiple artificial intelligence large models with the same or similar matching degrees that are ranked first, and the similar matching degree means that the difference in matching degrees is less than or equal to a preset interval; step S4 specifically includes: sorting the matching degrees of the artificial intelligence large models from high to low; if the matching degree of the artificial intelligence large model ranked first is significantly greater than that of the artificial intelligence large model ranked second, the artificial intelligence large model ranked first is used as the selection scheme, wherein the matching degree is significantly greater than that means that the difference in matching degrees is greater than the preset interval; if the matching degrees of the multiple artificial intelligence large models ranked first are the same or similar, the multiple artificial intelligence large models ranked first are used as the selection scheme.

[0012] In some optional implementations of some embodiments, the method further includes: when the single artificial intelligence big model is the optimal matching result for answering the question text, the answer of the single artificial intelligence big model to the question text is used as the answer result of the question text; when the multiple artificial intelligence big models are the optimal matching results for answering the question text, the pre-answer result of each artificial intelligence big model in the multiple artificial intelligence big models is obtained, and all the pre-answer results are fused using the answer fusion mechanism to obtain the answer result of the question text.

[0013] In some optional implementations of some embodiments, the answer fusion mechanism includes the following steps: obtaining the pre-answer results of each artificial intelligence large model, and preprocessing the pre-answer results to obtain feature information of the pre-answer results; extracting key information from the feature information of the pre-answer results; analyzing the commonalities and differences of the key information, and using natural language processing technology to generate the answer results of the question text.

[0014] In some optional implementations of some embodiments, the method further includes: obtaining the question text asked by the user for the n+1th time in real time; judging whether there is a correlation between the question text asked for the n+1th time and the question text asked for the nth time based on the keyword library judgment strategy; if there is a correlation, the selection scheme in the question text asked for the nth time is used as the optimal matching result of the answer; if there is no correlation, the selection scheme is dynamically adjusted according to the difficulty score of the question text asked for the n+1th time, wherein the dynamic adjustment of the selection scheme is reflected in the dynamic switching of a single artificial intelligence large model, the increase or decrease in the number of multiple artificial intelligence large models, and the replacement between a single artificial intelligence large model and multiple artificial intelligence large models.

[0015] In some optional implementations of some embodiments, before executing the dynamic adjustment selection plan and outputting the answer result, the method also includes: obtaining the artificial intelligence big model selected in the dynamic adjustment selection plan, and taking the previous n-th question text as the input item, and the output item is the background content of the n+1-th question text; taking the background content as an intervention item of the answer result, and using the selected artificial intelligence big model to answer the n+1-th question text to obtain the answer result.

[0016] In some optional implementations of some embodiments, the method further includes: automatically matching human characters for interaction according to the optimal matching result.

[0017] According to a second aspect of the present invention, there is provided an artificial intelligence large model question-answer matching system, which is applied to the artificial intelligence large model answer matching method, comprising: a question processing module, configured to obtain the question text asked by the user for the nth time, perform feature extraction processing on the question text using natural language processing technology, and construct a question feature set, where n is an integer greater than 0; a question difficulty score calculation module, configured to input the question feature set into a difficulty assessment model and output a difficulty score for the question text, wherein the difficulty assessment model is generated after training based on a quantized historical question feature set; a model matching degree calculation module, configured to calculate the matching degree of all artificial intelligence large models in the large model library based on the difficulty score of the question text using a large model matching degree formula, to obtain the matching degree of each artificial intelligence large model; and a matching result output module, configured to analyze the matching degree of each artificial intelligence large model according to a preset large model matching strategy to obtain a selection scheme, wherein the selection scheme is to select a single artificial intelligence large model or multiple artificial intelligence large models, wherein when a single artificial intelligence large model is selected, the selected single artificial intelligence large model is used as the optimal matching result for the answer; when multiple artificial intelligence large models are selected, the selected multiple artificial intelligence large models are used as the optimal matching results for the answer.

[0018] The advantages and beneficial effects of the present invention are as follows: the method and system for matching answers to artificial intelligence big models provided by the present invention can perform feature extraction processing on the question text asked by the user for the nth time, extract the entities of key technologies and the corresponding technical fields, and then construct a question feature set, and use the difficulty assessment model to evaluate and analyze the question feature set to obtain the difficulty score of the corresponding question text. By calculating the matching degree of the artificial intelligence big model, it is convenient to compare the difficulty processing capabilities of the question text asked for the nth time with the difficulty processing capabilities of each artificial intelligence big model in the big model library. At the same time, a big model matching strategy is adopted to selectively and intelligently select a single artificial intelligence big model or multiple artificial intelligence big models as the optimal matching results for the question text to answer, thereby improving the professionalism and accuracy of answering questions. In addition, by obtaining the question text asked by the user for the n+1th time in real time, according to whether there is a correlation between the question text asked for the n+1th time and the question text asked for the nth time and the difficulty score of the question text asked for the n+1th time, the artificial intelligence big model involved in the answer can be dynamically adjusted in a targeted manner, so as to solve the problem in the prior art that a single big model has insufficient processing capabilities for continuous questions and there is no switching strategy for the big model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is an exemplary data transmission and reception scenario diagram in which the present application can be applied.

[0020] Figure 2 It is a flowchart of an embodiment of the solution matching method of the artificial intelligence large model according to the present application.

[0021] Figure 3 It is a flowchart of another embodiment of the solution matching method of the artificial intelligence large model according to the present application.

[0022] Figure 4 It is a flowchart of another embodiment of the solution matching method of the artificial intelligence large model according to the present application.

[0023] Figure 5 It is a structural diagram of an embodiment of the solution matching system of the artificial intelligence large model according to the present application. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below through specific embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0025] While big models demonstrate significant potential for providing technical insights, guiding innovation hotspots, and analyzing industrial policies, variations in training data or subsequent optimization methods can impact their ability to handle continuous conversations and the quality of their responses. Therefore, matching the optimal big model to the content of a user's question and providing high-quality answers is crucial for user experience. Currently, when users use big models to ask questions, they typically rely on a single model to answer. Consequently, there are widespread issues such as a poor match between question content and the big model's response, and insufficient ability to handle continuous questioning. While some big model platforms support multiple AI big models working together to provide responses, these platforms also determine the AI ​​big model that responds based on the user's subjective choice or a fixed order.

[0026] Based on the above problems, this application proposes an answer matching method of an artificial intelligence big model. This method can intelligently select the most matching big model based on the relevant features of the question content, and identify the subtle differences in the relevant features of each question content in continuous questioning. By dynamically switching between multiple big models, the big model used to answer the question each time is the most preferred big model, thereby improving the accuracy of the answer and enhancing the user experience.

[0027] Figure 1 The diagram schematically shows a data transmission and reception scenario diagram between a terminal device and a server when a user uses a large model on a terminal device according to an embodiment of the present disclosure.

[0028] like Figure 1 As shown, the network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0029] Users can use terminal devices 101, 102, 103 to interact with server 105 via network 104 to receive or send messages, etc. Terminal devices 101, 102, 103 can be installed with various communication client applications, such as web browser applications, search applications, etc. (only for example).

[0030] Terminal devices 101, 102, and 103 may be various electronic devices with display screens that support web browsing, including but not limited to smartphones, tablet computers, laptop computers, and desktop computers. Server 105 may be a server that provides various services, such as a backend management server (for example only) that supports the large model websites browsed by users using terminal devices 101, 102, and 103. The backend management server may analyze and process received data such as user question requests, and provide feedback to terminal devices 101, 102, and 103 on the processing results (e.g., obtaining the most matching large model based on the user's question, or generating relevant information or data related to the user's initial or subsequent questions).

[0031] It should be noted that the solution matching method of the artificial intelligence large model provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the solution matching system of the artificial intelligence large model provided in the embodiment of the present disclosure can generally be set in the server 105. The solution matching method of the artificial intelligence large model provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the solution matching system of the artificial intelligence large model provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0032] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0033] The following will be based on Figure 1 The scene described by Figure 2 The solution matching method of the artificial intelligence large model of the disclosed embodiment is described in detail.

[0034] like Figure 2 FIG2 is a flowchart of an embodiment of a method for matching the solution of an artificial intelligence large model according to the present application. The method for matching the solution of an artificial intelligence large model includes the following steps.

[0035] S1. Obtain the question text asked by the user for the nth time, use natural language processing technology to extract features from the question text, and construct a question feature set, where n is an integer greater than 0.

[0036] In this step, in response to a user's question, after the user enters a question, the question text of the user's question is obtained. This question can be the nth question entered by the user in this round of questioning, or it can be all consecutive questions entered from the first to the nth in this round of questioning. Feature extraction is performed on the question text to construct the question feature set.

[0037] It should be noted that this application does not limit the input method, which can be voice input, mechanical keyboard input, behavioral input, etc. For non-text direct input methods, further recognition needs to be added. For example, when the user uses voice input, after obtaining the user's voice, the voice question is directly translated into text form through voice recognition technology, and the text is used as the question text for the next step of feature extraction; for another example, when the user uses behavioral input, after obtaining the user's action behavior, the user's action behavior is recognized by the action recognition model, and then translated into text form, and the text is used as the question text for the next step of feature extraction. It can be understood that users can input questions through a variety of input methods, which can not only provide convenience for users, but also expand the group of people who can use this method, for example, people with disabilities can also use it.

[0038] S2. Input the question feature set into a difficulty assessment model and output a difficulty score of the question text, wherein the difficulty assessment model is generated after training based on the quantized historical question feature set.

[0039] The difficulty assessment model is trained using a quantized set of historical question features, outputting a difficulty score for the corresponding question text. This difficulty score can be used as a reference when matching a large model, and a more optimal large model is matched based on this difficulty score for question answering.

[0040] S3. Based on the difficulty score of the question text, the matching degree of all artificial intelligence big models in the big model library is calculated using the big model matching formula to obtain the matching degree of each artificial intelligence big model.

[0041] According to one embodiment of the present application, the artificial intelligence large model matching formula is as follows: Where: represents the matching degree of the jth artificial intelligence model in the large model library, D represents the difficulty score of the question text, represents the difficulty processing capability score of the jth artificial intelligence large model in the large model library and is the known difficulty processing ability score data, and | | represents the absolute value.

[0042] Substitute the difficulty score in the second step into the above formula to obtain the matching degree of each artificial intelligence large model in the large model library. According to this formula, it can be concluded that The result must be greater than 0 and less than 1, that is, the matching degree of the artificial intelligence model is between 0 and 1. It can be understood that the closer the matching value is to 1, the closer the difficulty score of the question text is to the difficulty processing capability of the artificial intelligence model.

[0043] According to one embodiment of the present application, the difficulty score of the question text ranges from 0 to 1, with a higher difficulty score indicating a more complex question and a lower difficulty score indicating a simpler question. Correspondingly, the difficulty handling capability score of the large AI model also ranges from 0 to 1, with a higher difficulty handling capability score indicating a stronger problem handling capability.

[0044] S4. Analyze the matching degree of each artificial intelligence big model according to a preset big model matching strategy to obtain a selection scheme, which may be a single artificial intelligence big model or multiple artificial intelligence big models. According to one embodiment of the present application, when a single artificial intelligence big model is selected, the selected single artificial intelligence big model is used as the optimal matching result of the solution; when multiple artificial intelligence big models are selected, the selected multiple artificial intelligence big models are used as the optimal matching results of the solution.

[0045] In other words, the AI ​​big model that answers the user's nth question may be one or more, depending on the pre-set big model matching strategy and the matching degree of each AI big model. Furthermore, after the selected solution is determined, the selected AI big model is used to answer the user's nth question. The pre-set big model matching strategy may include the following.

[0046] According to one embodiment of the present application, the large model matching strategy preset in step S4 includes: selecting a single artificial intelligence large model with the best matching degree; and selecting multiple artificial intelligence large models with the same or similar matching degrees that are ranked first. Similar matching degrees means that the difference in matching degrees is less than or equal to a preset interval.

[0047] Therefore, further, step S4 specifically includes: sorting the matching degrees of the artificial intelligence large models from high to low; if the matching degree of the artificial intelligence large model ranked first is significantly greater than that of the artificial intelligence large model ranked second, the artificial intelligence large model ranked first is selected as the selection scheme, wherein the matching degree is significantly greater means that the difference in matching degree is greater than a preset interval; if the matching degrees of multiple artificial intelligence large models ranked first are the same or similar, the multiple artificial intelligence large models ranked first are selected as the selection scheme.

[0048] According to one embodiment of the present application, for example, the large model library contains three artificial intelligence large models. After the matching degree of each artificial intelligence large model is calculated, the matching degrees obtained are 0.9, 0.85, and 0.98, respectively. The matching degrees of each artificial intelligence large model are sorted from high to low. If the preset interval is 0.05, the difference in matching degree between the first-ranked artificial intelligence large model and the first-ranked artificial intelligence large model is 0.08 (that is, 0.98-0.9=0.08), that is, the difference in matching degree between the two is greater than the preset interval, then only the first-ranked artificial intelligence large model (that is, the matching degree is 0.98) is selected as the selection scheme.

[0049] For another example, suppose the large model library contains three artificial intelligence large models. After calculating the matching degree of each artificial intelligence large model, the matching degrees obtained are 0.96, 0.95, and 0.93, respectively. The matching degrees of each artificial intelligence large model are sorted from high to low. If the preset interval is 0.05, the difference between the first-ranked artificial intelligence large model and the other two artificial intelligence large models is within 0.05 (that is, 0.96-0.95=0.01, 0.96-0.93=0.03), then it can be said that the matching degrees of these three artificial intelligence large models are similar, and these three artificial intelligence large models are selected as the selection plan.

[0050] For another example, suppose the large model library contains three artificial intelligence large models. After calculating the matching degree of each artificial intelligence large model, the matching degrees obtained are 0.96, 0.95, and 0.8, respectively. The matching degrees of each artificial intelligence large model are sorted from high to low. If the preset interval is 0.05, the difference between the first-ranked artificial intelligence large model and the other two artificial intelligence large models is 0.01 and 0.16, respectively (that is, 0.96-0.95=0.01, 0.96-0.8=0.16), which means that the matching degree of the first-ranked artificial intelligence large model is similar to that of the artificial intelligence large model with a matching degree of 0.95, but the matching degree gap with the artificial intelligence large model with a matching degree of 0.8 is significant. Therefore, the artificial intelligence large model ranked first and with a matching degree of 0.95 is selected as the selection plan.

[0051] According to the artificial intelligence big model matching method of the present application, by extracting problem features and analyzing the difficulty of the problem, and then calculating the matching degree between the problem and the big model, the accuracy and efficiency of the artificial intelligence big model's answer can be improved. At the same time, by selecting solutions, this method also provides multiple artificial intelligence big models to answer problems, which can enhance the comprehensiveness and accuracy of the answers.

[0052] Figure 3 It is a flowchart for constructing question feature set through question text.

[0053] like Figure 3As shown, in some optional implementations of the present application, in step S1, the question text asked by the user for the nth time is obtained, natural language processing technology is used to perform feature extraction processing on the question text, and a question feature set is constructed, including steps S101-S104.

[0054] S101. Obtain a question text asked by a user, perform word segmentation processing on the question text based on a word segmentation model, and perform part-of-speech tagging on each question participle to obtain multiple question participles and part-of-speech tagging information corresponding to the question participles.

[0055] According to one embodiment of the present application, in this step, the word segmentation model that can be used is a word segmentation model based on deep learning, including but not limited to a BERT model or an LSTM model, which performs word segmentation processing on the user's question text and performs part-of-speech tagging on each question segmentation to distinguish parts of speech such as nouns and verbs, laying the foundation for subsequent entity recognition.

[0056] S102: Perform entity recognition on the question segmentation using a neural network model according to part-of-speech tagging information corresponding to the question segmentation, and determine an entity of at least one key technology.

[0057] According to one embodiment of the present application, in this step, entity recognition is performed on the question segmentation based on the question segmentation and the corresponding part-of-speech tagging information by utilizing a neural network model, so as to identify the key technical entity recognition contained in the question text, for example, the key technical entity is the algorithm name, equipment model, etc.

[0058] The above neural network model uses a bidirectional long short-term memory conditional random field model (BiLSTM-CRF). The BiLSTM-CRF model is pre-trained on a training set of labeled input word segments and their context features. Entity features are extracted for each question segment, including the word itself, context words, part-of-speech tags, position features, etc. The i-th question segment is represented as , its eigenvector is expressed as ; The feature vector is composed of the word vector of the word segmentation itself, the word vector of the context word, the part-of-speech vector, and the position feature sub-feature vector; the vector of the entity feature of the question word segmentation Input BiLSTM layer to form question segmentation Contextual representation of In the BiLSTM layer, the context representation of each time step is obtained by concatenating the outputs of the forward and backward LSTMs: this context representation is provided to the CRF layer; the CRF layer receives the output from the BiLSTM and considers the label sequence Under the constraint of , the Viterbi algorithm is used to decode and find the most likely label sequence as output. This label sequence is the annotation result of the question word segmentation. The probability of the label sequence is expressed as:

[0059] ;in, is the characteristic function, is the set of all possible label sequences; the most likely label sequence is expressed as: ; The constraints of the label sequence can adopt a BIO annotation protocol, where B represents the beginning of a technical noun, I represents internal, and O represents non-technical noun. According to the annotation results, words marked as B (beginning) are extracted, and the words marked with I (internal) that are continuous with it are collected to form a complete key technical noun.

[0060] S103: Perform semantic analysis on the entity of the at least one key technology based on the semantic analysis model to determine the technical field information to which the entity of the key technology belongs. According to one embodiment of the present application, in this step, the identified key technology term is matched with a predefined technical field dictionary to identify the technical field information to which the question text belongs.

[0061] S104. Construct a problem feature set based on the entities of key technologies and the information of the technical fields to which they belong.

[0062] According to one embodiment of the present application, it is assumed that the question raised by the user is: "How to implement an image recognition system based on deep learning to detect multi-category targets in real time?" The operating system of this method receives the user's question, first splits the question into multiple question word segmentations, and performs part-of-speech tagging on each question word, that is, splits the above question into: "how", "implementation", "based on", "deep learning", "image recognition system", "to", "multi-category targets", "of", "real-time", "detection" and other question word segmentations, and performs part-of-speech tagging on each question word, that is, marks "how" as a pronoun, marks "implementation" as a verb, marks "based on" as a preposition, and marks "deep learning" as a verb. " is marked as an adjective, "image recognition system" is marked as a noun, "to" is marked as a preposition, "multi-category target" is marked as a noun, "of" is marked as an auxiliary word, "real-time" is marked as an adverb, and "detection" is marked as a verb. Secondly, nouns and adjectives before nouns are identified, and key technical entity recognition results such as "deep learning", "image recognition system" and "multi-category target" are identified. The key technical entity recognition results are analyzed again to determine the technical field involved in the problem. According to the above key technical entity recognition results, the above key technical nouns are matched with the predefined technical field dictionary, and the analysis determines that the technical field to which the problem belongs is the field of computer vision technology.

[0063] Through the above steps, we can accurately understand the user's problem and evaluate the technical field to which the problem belongs, so as to provide a basis for subsequent evaluation of the problem difficulty and selection of an appropriate large model.

[0064] Furthermore, step S1 may also include step S105, quantifying the entities of the key technologies included in the problem feature set and the information of the technical fields to which they belong, to obtain technical depth information and technical concept breadth information. According to one embodiment of the present application, the technical depth information is determined by the hierarchical relationship between the entities of multiple key technologies, and the technical concept breadth is determined by counting the number of pieces of information in the technical fields to which they belong.

[0065] That is to say, the extracted entities of key technologies and the information of the technical fields to which they belong are quantified. The results of quantification can be reflected in two aspects: technical depth and conceptual breadth. Among them, the technical depth can be evaluated by the hierarchical relationship of the entities of multiple key technologies, and the conceptual breadth can be measured by counting the number of information in the technical fields to which they belong, thereby obtaining technical depth information and technical concept breadth information respectively.

[0066] Specifically, the prompt word engineering can be used to use a large model for quantitative evaluation of problem feature sets, and to perform quantitative processing using the entities of key technologies and the information of the technical fields to which they belong contained in the problem feature sets, so as to obtain quantitative information on the depth of technology and the breadth of technology concepts.

[0067] Furthermore, step S2 includes steps S201-S202.

[0068] S201: Input the problem feature set containing quantitative technical depth information and technical concept breadth information into the difficulty assessment model. The difficulty assessment model is a large model that is pre-trained and fine-tuned using the quantified historical problem feature set.

[0069] S202, using the prompt word engineering to enable the difficulty assessment model to output the difficulty score of the question text based on the quantitative technical depth information and technical concept breadth information in the input question feature set; wherein the prompt words included in the prompt word engineering define the value range of the difficulty score of the output question text, and define the rules for determining the difficulty score based on the quantitative technical depth information and technical concept breadth information.

[0070] Figure 4 Flowchart of another embodiment of the solution matching method of the artificial intelligence large model according to the present application; Figure 4 As shown, in some optional implementations of the embodiments of the present application, after step S4, the following steps may also be performed.

[0071] S6. When a single artificial intelligence big model is the optimal matching result for answering the question text, the answer of the single artificial intelligence big model to the question text is used as the answer result of the question text; when multiple artificial intelligence big models are the optimal matching results for answering the question text, the pre-answer results of each artificial intelligence big model in the multiple artificial intelligence big models are obtained, and the answer fusion mechanism is used to fuse all the pre-answer results to obtain the answer result of the question text.

[0072] For example, if the artificial intelligence big model with a matching degree of 0.98 is selected as the selection scheme, the artificial intelligence big model is used to answer the question text to obtain the answer to the question text; for example, if three artificial intelligence big models with matching degrees of 0.96, 0.95, and 0.93 are selected as the selection scheme, the three artificial intelligence big models are used to answer the question text respectively to obtain three preliminary answer results. Based on the answer fusion mechanism, the three preliminary answer results are fused to finally obtain the answer to the question text.

[0073] For the case where multiple large models are selected, the present application designs a solution fusion mechanism. According to an embodiment of the present application, the solution fusion mechanism in step S6 may include the following steps.

[0074] Obtain the pre-answer results of each artificial intelligence large model, and pre-process the pre-answer results to obtain feature information of the pre-answer results; extract key information from the feature information of the pre-answer results; analyze the similarities and differences of the key information, and use natural language processing technology to generate the answer results of the question text.

[0075] It is understandable that after extracting and comparing key information, the commonalities among the AI ​​models that answer the question can serve as the core content of the answer, while the differences can be further evaluated. The reliability and innovation of these differences can be evaluated through algorithms, and the better-evaluated differences can be integrated into the core content using natural language processing technology to obtain a complete answer, namely the answer to the question text. It should be noted that when the evaluation results of the differences are all good, the answer can be supplemented by an "or" method; when the evaluation results differ significantly, the answer will be supplemented by the better-evaluated results.

[0076] For example, consider the user question "How to optimize the convergence speed of a deep learning network?" The two AI models with matching degrees of 0.96 and 0.95 selected in step S4 are selected as the solution, and then step S6 is executed. After obtaining the preliminary answers of the two large artificial intelligence models, the two preliminary answers are first preprocessed to obtain feature information of the preliminary answers. The feature information includes "by adjusting the learning rate and using a more efficient optimization algorithm and introducing a regularization term to obtain...", "by adjusting the learning rate, using a more efficient optimization algorithm and batch normalization to obtain...", and information extraction is performed on the feature information to obtain key information associated with the question text. The key information includes "introducing a regularization term", "using batch normalization", "adjusting the learning rate", and "using a more efficient optimization algorithm". By analyzing the commonalities and differences of the key information, it is found that both of the above-mentioned large artificial intelligence models mentioned "adjusting the learning rate" and "using a more efficient optimization algorithm". Therefore, these two key information are taken as commonalities. As for the differences, one of the large artificial intelligence models mentioned "introducing a regularization term" and the other mentioned "using batch normalization". After analyzing and evaluating the differences, it was found that both differences were reliable and innovative. Therefore, after combining with natural language processing technology, a supplementary result of "or" was generated. The final answer to the generated question text at least includes the following content: "In order to optimize the convergence speed of the deep learning network, it is recommended to adjust the learning rate, use a more efficient optimization algorithm, and consider introducing regularization terms or using batch normalization."

[0077] According to one embodiment of the present application, the matching solution of the present application is not only applicable to the matching process of one question, but also applicable to matching in multiple rounds of questions, that is, a scenario of continuous questioning.

[0078] In view of the possible changes in the difficulty of questions in the embodiment of continuous questioning, the present application also provides a dynamic switching mechanism, which dynamically adjusts the large model involved in answering based on the difficulty of the continuous questions monitored in real time. This can be achieved in the following way.

[0079] S7. Obtain the question text asked by the user for the n+1th time in real time; based on the keyword library judgment strategy, determine whether there is any correlation between the question text asked for the n+1th time and the question text asked for the nth time; if there is any correlation, use the selection plan in the question text asked for the nth time as the optimal matching result of the answer; if there is no correlation, dynamically adjust the selection plan according to the difficulty score of the question text asked for the n+1th time.

[0080] It can be understood that n+1 represents the next question based on n, that is, when n is 1, that is, the user's question is the first question. After steps S1-S6 complete the answer to the first question, if it is detected that the user continues to ask a second question, step S7 can be further executed. The difficulty of the question asked for the first time and the question asked for the second time may be different, so it is necessary to adjust the artificial intelligence model for answering the question, and use the adjusted artificial intelligence model to answer the user's second question. Furthermore, after detecting that the user has asked a third question, step S7 is continued, that is, the difficulty of each question is obtained in real time during the continuous question-answering session, and the artificial intelligence model involved in the answer is automatically and dynamically adjusted based on the previous question.

[0081] Because each AI model can answer different technical questions to varying degrees, correlation can be used to determine whether the questions asked by the user are consecutive and have similar difficulty levels. For consecutive questions, the question is likely a further extension of the previous question's technical point. Therefore, the same AI model can be used to answer the question, eliminating the need to repeat steps S1-S4, saving computational space and improving speed.

[0082] Furthermore, the dynamic adjustment and selection scheme is reflected in the dynamic switching of a single artificial intelligence large model, the increase or decrease in the number of multiple artificial intelligence large models, and the replacement between a single artificial intelligence large model and multiple artificial intelligence large models.

[0083] According to one embodiment of the present application, in order to improve the accuracy of answers to users' continuous questions, before executing the dynamic adjustment selection scheme and outputting the answer result, the matching method also includes: obtaining the artificial intelligence big model selected in the dynamic adjustment selection scheme, and taking the previous n-th question text as the input item, and the output item is the background content of the n+1-th question text; taking the background content as an intervention item of the answer result, and using the selected artificial intelligence big model to answer the n+1-th question text to obtain the answer result.

[0084] In this embodiment, due to the lack of relevance to the previous question, the artificial intelligence model for answering questions needs to be readjusted. Directly using the dynamically adjusted new large model for direct answers may result in a fault situation, that is, the answer to the current round of questions is separated from the basis of the answers in the previous rounds, making the answer incoherent, difficult to understand, or biased, and even some content contradictory. In order to prevent the above problems, all questions before the current round of questions are input into the dynamically adjusted new large model, that is, the text of the previous n questions is used as the background content of the text of the n+1th question, and the answer to the n+1th question is based on the previous n questions.

[0085] According to one embodiment of the present application, the keyword library judgment strategy involved in the present application specifically includes: constructing a keyword library based on preset keywords, obtaining the question text asked by the user for the n+1th time, using natural language processing technology, and splitting the question text asked for the n+1th time according to sentence granularity based on the comma symbol to obtain several split sentences, and traversing the keyword library to obtain the result of whether the split sentence hits the keyword. If the result is that the split sentence hits the keyword, it is determined that there is a correlation between the question text asked for the n+1th time and the question text asked for the nth time. If the result is that the split sentence does not hit the keyword, it is determined that there is no correlation between the question text asked for the n+1th time and the question text asked for the nth time.

[0086] According to one embodiment of the present application, for example, the question text asked by the user for the nth time is: "What is reinforcement learning". After the system answers the nth question, the user continues to ask questions, and the question text asked for the n+1th time is: "In combination with the above problems, how to balance exploration and utilization". Based on the keyword library judgment strategy, the question text asked for the n+1th time is split according to the comma symbol to obtain multiple split sentences, such as: "In combination with the above problems" and "How to balance exploration and utilization". The keyword library is traversed. If the keyword library contains the keyword "in combination with the above problems", the result of the split sentence hitting the keyword is obtained, that is, it is determined that there is a correlation between the question text asked for the n+1th time and the question text asked for the nth time.

[0087] In order to improve the user experience of interactive answers, this application can also automatically match characters for interaction based on the optimal matching results from the user's perspective. Each artificial intelligence big model has a technology innovation assistant role with unique personality and visual image. Each role has a specific knowledge field, professional background and language style based on the characteristics of its corresponding big model.

[0088] In Example 1, when a user asks a question for the first time, the question difficulty is assessed as easy to answer and is a common question in this field. Therefore, the artificial intelligence model that answers this question is given an assistant a role through the display interface, and the user can conduct a human-computer dialogue with assistant a through the display interface.

[0089] Example 2, following Example 1, after answering the current question, the user continues to ask questions. Assuming that when it is determined that the user's subsequent question is still in this field, the optimal matching result changes and the artificial intelligence model involved in the answer needs to be dynamically adjusted, the assistant role at this time needs to change according to the dynamic adjustment of the artificial intelligence model involved in the answer, such as becoming assistant B.

[0090] Example 3, following Example 1, after answering the current question, the user continues to ask questions. Assuming that when it is determined that the user's subsequent questions are still in this field but the difficulty of the questions increases, the optimal matching result changes and the artificial intelligence model involved in the answer needs to be dynamically adjusted. At this time, the assistant role needs to change according to the dynamic adjustment of the artificial intelligence model involved in the answer, such as becoming expert a.

[0091] As can be understood from Examples 1-3 above, the dynamic transformation of the AI ​​models will result in different roles. When two AI models can answer questions of similar difficulty, Assistant A and Assistant B can be identified due to the changes in the models. When the difficulty of the two AI models differs significantly, the more specialized Expert A can be included. By giving the AI ​​models an anthropomorphic identity and utilizing this anthropomorphic interaction design, users not only receive professional answers but also enjoy a more humane and personalized service experience.

[0092] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0093] Further references Figure 5 , as a response to the above Figure 2 The present application provides an embodiment of an artificial intelligence large model solution matching system, which is similar to Figure 2 Corresponding to the method embodiment shown, the system can be specifically applied to various electronic devices.

[0094] like Figure 5 As shown, the answer matching system 300 of the artificial intelligence large model described in this embodiment includes: a question processing module 301, a question difficulty score calculation module 302, a model matching degree calculation module 303 and a matching result output module 304.

[0095] The question processing module 301 is used to obtain the question text asked by the user for the nth time, perform feature extraction processing on the question text using natural language processing technology, and construct a question feature set, where n is an integer greater than 0.

[0096] The question difficulty score calculation module 302 is used to input the question feature set into the difficulty assessment model and output the difficulty score of the question text, wherein the difficulty assessment model is generated after training based on the quantized historical question feature set.

[0097] The model matching degree calculation module 303 is used to calculate the matching degree of all artificial intelligence large models in the large model library based on the difficulty score of the question text using the large model matching degree formula to obtain the matching degree of each artificial intelligence large model.

[0098] The matching result output module 304 is used to analyze the matching degree of each artificial intelligence big model according to the preset big model matching strategy to obtain a selection scheme, wherein the selection scheme is to select a single artificial intelligence big model or to select multiple artificial intelligence big models. When a single artificial intelligence big model is selected, the selected single artificial intelligence big model is used as the optimal matching result of the solution; when multiple artificial intelligence big models are selected, the selected multiple artificial intelligence big models are used as the optimal matching results of the solution.

[0099] According to the matching system of the present application, the matching method of the artificial intelligence big model of the present application can be applied to perform feature extraction processing on the question text asked by the user for the nth time, extract the entities of the key technologies and the corresponding technical field information, and then construct a question feature set, and use the difficulty assessment model to evaluate and analyze the question feature set to obtain the difficulty score of the corresponding question text. By calculating the matching degree of the artificial intelligence big model, it is convenient to compare the difficulty processing capabilities of the question text asked for the nth time with the difficulty processing capabilities of each artificial intelligence big model in the big model library; at the same time, a big model matching strategy is adopted to selectively and intelligently select a single artificial intelligence big model or multiple artificial intelligence big models as the optimal matching results for the question text to answer, thereby improving the professionalism and accuracy of answering questions; in addition, by obtaining the question text asked by the user for the n+1th time in real time, according to whether there is a correlation between the question text asked for the n+1th time and the question text asked for the nth time and the difficulty score of the question text asked for the n+1th time, the artificial intelligence big model involved in the answer can be dynamically adjusted in a targeted manner to solve the problem in the prior art that a single big model has insufficient processing capabilities for continuous questions and the big model has no switching strategy.

[0100] According to embodiments of the present disclosure, any multiple modules of the question processing module 301, the question difficulty score calculation module 302, the model matching degree calculation module 303, and the matching result output module 304 can be combined into a single module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module. According to embodiments of the present disclosure, at least one of the question processing module 301, the question difficulty score calculation module 302, the model matching degree calculation module 303, and the matching result output module 304 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware through any other reasonable means of circuit integration or packaging, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or any appropriate combination of any of these. Alternatively, at least one of the question processing module 301, the question difficulty score calculation module 302, the model matching degree calculation module 303 and the matching result output module 304 can be at least partially implemented as a computer program module, which can perform corresponding functions when executed.

[0101] The electronic device according to an embodiment of the present application includes a processor, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage portion into a random access memory (RAM). The processor may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present application.

[0102] The RAM stores various programs and data required for the operation of the electronic device. The processor, ROM, and RAM are connected to each other via a bus. The processor executes the programs in the ROM and / or RAM to perform the various operations of the method flow according to the embodiments of the present application. It should be noted that the programs may also be stored in one or more memories other than ROM and RAM. The processor may also execute the programs stored in the one or more memories to perform the various operations of the method flow according to the embodiments of the present application.

[0103] According to an embodiment of the present application, the electronic device may further include an input / output (I / O) interface, which is also connected to the bus. The electronic device may further include one or more of the following components connected to the I / O interface: an input portion including a keyboard, a mouse, etc.; an output portion including a cathode ray tube (CRT), a liquid crystal display (LCD), a speaker, etc.; a storage portion including a hard disk, etc.; and a communication portion including a network interface card such as a LAN card or a modem. The communication portion performs communication processing via a network such as the Internet. A drive is also connected to the I / O interface as needed. Removable media, such as magnetic disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed in the drive as needed, so that computer programs read therefrom can be installed into the storage portion as needed.

[0104] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when the one or more programs are executed, the method according to the embodiments of this application is implemented.

[0105] According to embodiments of the present application, a computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but not limited to, 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), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present application, a computer-readable storage medium may include the ROM and / or RAM described above, and / or one or more memories other than ROM and RAM.

[0106] The present application also includes a computer program product comprising a computer program containing program code for executing the method shown in the flowchart. When the computer program product is executed in a computer system, the program code is used to cause the computer system to implement the item recommendation method provided in the present application.

[0107] When the computer program is executed by the processor, the above functions defined in the system / device of the embodiment of the present application are performed. According to the embodiment of the present application, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0108] In one embodiment, the computer program may be stored on a tangible storage medium, such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal over a network medium, downloaded and installed via a communication component, and / or installed from a removable medium. The program code contained in the computer program may be transmitted using any suitable network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0109] In such an embodiment, the computer program can be downloaded and installed from a network via the communication portion, and / or installed from a removable medium. When the computer program is executed by the processor, the above-mentioned functions defined in the system of the embodiment of the present application are performed. According to the embodiment of the present application, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0110] According to an embodiment of the present application, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0111] Obviously, the embodiments described above are only some of the embodiments of the present application, rather than all of the embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned specific embodiments, or to make equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the present application specification and the accompanying drawings, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present application.

Claims

1. The solution matching method of the artificial intelligence large model is characterized by: include: S1. Obtain the question text asked by the user for the nth time, use natural language processing technology to extract features from the question text, and construct a question feature set, where n is an integer greater than 0; S2, inputting the question feature set into a difficulty assessment model, and outputting a difficulty score of the question text, wherein the difficulty assessment model is generated after training based on the quantized historical question feature set; S3. Based on the difficulty score of the question text, the matching degree of all artificial intelligence big models in the big model library is calculated using the big model matching degree formula to obtain the matching degree of each artificial intelligence big model; S4. Analyze the matching degree of each artificial intelligence big model according to the preset big model matching strategy to obtain a selection scheme, wherein the selection scheme is to select a single artificial intelligence big model or multiple artificial intelligence big models; Wherein, when a single artificial intelligence big model is selected, the selected single artificial intelligence big model is used as the optimal matching result of the solution; when multiple artificial intelligence big models are selected, the selected multiple artificial intelligence big models are used as the optimal matching result of the solution; Also includes: Get the text of the question asked by the user for the n+1th time in real time; Based on the keyword library judgment strategy, it is judged whether there is relevance between the question text asked for the n+1th time and the question text asked for the nth time. If there is relevance, the selection scheme in the question text asked for the nth time is taken as the optimal matching result of the answer. If there is no relevance, the selection scheme is dynamically adjusted according to the difficulty score of the question text asked for the n+1th time. The dynamic adjustment selection scheme includes the dynamic switching of a single artificial intelligence large model, the increase or decrease of the number of multiple artificial intelligence large models, and the replacement of a single artificial intelligence large model with multiple artificial intelligence large models. Before executing the dynamic adjustment selection scheme and outputting the solution result, the method further includes: Obtain the artificial intelligence large model selected in the dynamic adjustment selection scheme, and use the previous n-th question text as an input item, and the output item is the background content of the n+1-th question text; The background content is used as an intervention item of the solution result, and the selected artificial intelligence large model is used to solve the n+1th question text to obtain the solution result.

2. The method for matching the solution of the artificial intelligence large model according to claim 1, characterized in that: The step S1 specifically includes: S101, obtaining a question text asked by a user, performing word segmentation processing on the question text based on a word segmentation model and performing part-of-speech tagging on each question participle, to obtain multiple question participles and part-of-speech tagging information corresponding to the question participles; S102, performing entity recognition on the problem segmentation using a neural network model according to part-of-speech tagging information corresponding to the problem segmentation, and determining an entity of at least one key technology; S103, performing semantic analysis processing on the entity of the at least one key technology based on the semantic analysis model to determine the technical field information to which the entity of the key technology belongs; S104. Construct a problem feature set based on the entities of key technologies and the information of the technical fields to which they belong.

3. The method for matching the solution of the artificial intelligence large model according to claim 2, characterized in that: The step S1 further comprises: S105. Quantify the entities of the key technologies contained in the problem feature set and the information on the technical fields to which they belong, to obtain technical depth information and technical concept breadth information, wherein the technical depth information is determined by the hierarchical relationship of the entities of multiple key technologies, and the technical concept breadth is determined by counting the number of pieces of information on the technical fields to which they belong.

4. The method for matching the solution of the artificial intelligence large model according to claim 1, characterized in that: In step S3, the large model matching degree formula is as follows: Where: represents the matching degree of the jth artificial intelligence large model in the large model library, D represents the difficulty score of the question text, represents the difficulty processing capability score of the jth artificial intelligence large model in the large model library and is the known difficulty processing ability scoring data, and | | represents the absolute value.

5. The method for matching the solution of the artificial intelligence large model according to claim 4 is characterized in that: The preset large model matching strategy in step S4 includes: Selecting a single AI model with the best match; and Selecting multiple artificial intelligence large models with the same or similar matching degree that are ranked first, wherein the similar matching degree means that the difference in matching degree is less than or equal to a preset interval; The step S4 specifically includes: Sort the matching degree of the AI ​​big models from high to low; If the matching degree of the first-ranked artificial intelligence large model is significantly greater than that of the second-ranked artificial intelligence large model, the first-ranked artificial intelligence large model will be used as the selection plan, wherein the matching degree is significantly greater means that the difference in matching degrees is greater than a preset interval; if the matching degrees of multiple artificial intelligence large models ranked first are the same or similar, the multiple artificial intelligence large models ranked first will be used as the selection plan.

6. The method for matching the answer of the artificial intelligence big model according to any one of claims 1 to 5, characterized in that: The method further comprises: When the single artificial intelligence big model is used as the optimal matching result for answering the question text, the answer of the single artificial intelligence big model to the question text is used as the answer result of the question text; When the multiple artificial intelligence big models are used as the optimal matching results for answering the question text, the preliminary answer results of each artificial intelligence big model in the multiple artificial intelligence big models are obtained, and the answer fusion mechanism is used to fuse all the preliminary answer results to obtain the answer result of the question text.

7. The method for matching the solution of the artificial intelligence large model according to claim 6, characterized in that: The solution fusion mechanism comprises the following steps: Obtaining the pre-answer results of each artificial intelligence large model, and pre-processing the pre-answer results to obtain feature information of the pre-answer results; Extracting key information from the feature information of the preliminary answer result; The commonalities and differences of the key information are analyzed, and the answer to the question text is generated using natural language processing technology.

8. The method for matching the solution of the artificial intelligence large model according to claim 1, characterized in that: The method further comprises: automatically matching the characters for interaction according to the optimal matching result.

9. An artificial intelligence large model solution matching system, applied to an artificial intelligence large model solution matching method as claimed in any one of claims 1 to 8, characterized in that: include: The question processing module is used to obtain the question text asked by the user for the nth time, perform feature extraction processing on the question text using natural language processing technology, and construct a question feature set, where n is an integer greater than 0; A question difficulty score calculation module, used to input the question feature set into a difficulty assessment model and output a difficulty score of the question text, wherein the difficulty assessment model is generated after training based on the quantified historical question feature set; The model matching degree calculation module is used to score the difficulty of the question text, and uses the large model matching degree formula to calculate the matching degree of all artificial intelligence large models in the large model library to obtain the matching degree of each artificial intelligence large model; A matching result output module is used to analyze the matching degree of each artificial intelligence big model according to a preset big model matching strategy to obtain a selection scheme, wherein the selection scheme is to select a single artificial intelligence big model or to select multiple artificial intelligence big models, wherein when a single artificial intelligence big model is selected, the selected single artificial intelligence big model is used as the optimal matching result of the solution; when multiple artificial intelligence big models are selected, the selected multiple artificial intelligence big models are used as the optimal matching results of the solution; Also includes: Get the text of the question asked by the user for the n+1th time in real time; Based on the keyword library judgment strategy, it is judged whether there is relevance between the question text asked for the n+1th time and the question text asked for the nth time. If there is relevance, the selection scheme in the question text asked for the nth time is taken as the optimal matching result of the answer. If there is no relevance, the selection scheme is dynamically adjusted according to the difficulty score of the question text asked for the n+1th time. The dynamic adjustment selection scheme includes the dynamic switching of a single artificial intelligence large model, the increase or decrease of the number of multiple artificial intelligence large models, and the replacement of a single artificial intelligence large model with multiple artificial intelligence large models. Before executing the dynamic adjustment selection scheme and outputting the solution result, the method further includes: Obtain the artificial intelligence large model selected in the dynamic adjustment selection scheme, and use the previous n-th question text as an input item, and the output item is the background content of the n+1-th question text; The background content is used as an intervention item of the solution result, and the selected artificial intelligence large model is used to solve the n+1th question text to obtain the solution result.

Images