Content analysis method and device, equipment and medium

Abstract

The invention discloses a content analysis method and device, equipment and a medium. The content analysis method comprises the steps of obtaining a target question and a plurality of candidate answers corresponding to the target question; inputting the target question and the plurality of candidate answers into a first analysis model to obtain a first analysis result of each candidate answer; wherein the first analysis model is trained by the following steps: determining a plurality of sample questions and a plurality of sample candidate answers corresponding to each sample question; analyzing the plurality of sample candidate answers corresponding to each sample question by using an AI large model to obtain a sample analysis result of each sample candidate answer; and performing model training by using the plurality of sample questions, the plurality of sample candidate answers corresponding to each sample question and the sample analysis result corresponding to each sample candidate answer to obtain a first analysis model. According to the analysis model provided by the invention, the user can be helped to quickly select the optimal reply from a plurality of different replies, and the adaptability of the analysis model is relatively high.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a content analysis method, apparatus, device, and medium. Background Art

[0002] With the development of artificial intelligence (AI) technology, various large AI models have emerged, and users can interact with large AI models, such as having conversations, asking questions, etc.

[0003] During the interaction between users and large AI models, due to certain deviations in the process of large AI models generating responses, the responses given by large AI models are diverse. For example, for the same question, large AI models may give multiple different responses.

[0004] In the face of multiple different responses, it is difficult for users to quickly select the optimal response. Summary of the Invention

[0005] This application provides a content analysis method, apparatus, device, and medium, which can help users quickly select the optimal response from multiple different responses.

[0006] To achieve the above objective, this application adopts the following technical solutions: In a first aspect, this application provides a content analysis method, including: Obtain a target question and multiple candidate answers corresponding to the target question; Input the target question and the multiple candidate answers into a first analysis model to obtain a first analysis result for each candidate answer; Among them, the first analysis model is trained through the following method: Determine multiple sample questions and multiple sample candidate answers corresponding to each sample question; Use a large AI model to analyze the multiple sample candidate answers corresponding to each sample question to obtain a sample analysis result for each sample candidate answer; Use the multiple sample questions, the multiple sample candidate answers corresponding to each sample question, and the sample analysis results corresponding to each sample candidate answer for model training to obtain the first analysis model.

[0007] In some possible implementation manners, the method further includes: Use a large AI model to analyze the target question and the multiple candidate answers corresponding to the target question to obtain a second analysis result for each candidate answer; Compare the first analysis result and the second analysis result of each candidate answer to obtain multiple comparison results; If there is a target comparison result, the first analysis model is updated using the target question and the second analysis results of multiple candidate answers corresponding to the target question to obtain a second analysis model; the target comparison result is a comparison result where the similarity between the first analysis result and the second analysis result of the candidate answer is lower than a preset similarity threshold.

[0008] In some possible implementation manners, the method further includes: Determine a target candidate answer according to the first analysis result of each candidate answer.

[0009] In some possible implementation manners, the sample analysis result includes a sample score and a sample comment.

[0010] In some possible implementation manners, the sample score is determined by the following method: Using an AI large model, score each sample candidate answer of each sample question separately from multiple dimensions to obtain a first separate score for each sample candidate answer; Using an AI large model, score the combined sample candidate answers separately from multiple dimensions to obtain a first combined score for each sample candidate answer; Average the first separate score and the first combined score of each sample candidate answer to obtain a first-stage score; Adjust the first-stage score of the sample candidate answer according to the similarity between multiple sample candidate answers corresponding to the sample question to obtain a second-stage score; Use the average of the first-stage score and the second-stage score of each sample candidate answer as the sample score of each sample candidate answer.

[0011] In some possible implementation manners, adjusting the first-stage score of the sample candidate answer according to the similarity between multiple sample candidate answers corresponding to the sample question to obtain a second-stage score includes: Determine a highest-score sample answer and a lowest-score sample answer from multiple sample candidate answers corresponding to the sample question. The first-stage score of the highest-score sample answer is the highest, and the first-stage score of the lowest-score sample answer is the lowest; Adjust the first-stage score of the first other sample candidate answer according to the similarity between the first other sample candidate answer and the highest-score sample answer among multiple sample candidate answers except the highest-score sample answer to obtain a first adjusted score of the first other sample candidate answer; Adjust the first-stage score of the second other sample candidate answer according to the similarity between the second other sample candidate answer and the lowest-score sample answer among multiple sample candidate answers except the lowest-score sample answer to obtain a second adjusted score of the second other sample candidate answer; Adjust the first-stage scores corresponding to the selected sample candidate answers according to the similarity between any two sample candidate answers among multiple sample candidate answers to obtain the third adjusted scores; In the case where the sample candidate answer is the highest-score sample answer, determine the second-stage score of the highest-score sample answer according to the second adjusted score and the third adjusted score of the highest-score sample answer; In the case where the sample candidate answer is the lowest-score sample answer, determine the second-stage score of the lowest-score sample answer according to the first adjusted score and the third adjusted score of the lowest-score sample answer; In the case where the sample candidate answer is other sample answers other than the highest-score sample answer and the lowest-score sample answer, determine the second-stage scores of the other sample answers according to the first adjusted scores, the second adjusted scores and the third adjusted scores of the other sample answers.

[0012] In some possible implementation manners, the sample comments are determined as follows: Use the AI large model to generate sample comments of the sample candidate answers in multiple dimensions based on the sample scores of the sample candidate answers corresponding to the sample questions.

[0013] In a second aspect, the present application provides a content analysis device, and the device includes: An acquisition module, configured to acquire a target question and multiple candidate answers corresponding to the target question; An analysis module, configured to input the target question and multiple candidate answers into a first analysis model to obtain a first analysis result of each candidate answer; wherein, the first analysis model is trained as follows: determine multiple sample questions and multiple sample candidate answers corresponding to each sample question; use the AI large model to analyze the multiple sample candidate answers corresponding to each sample question to obtain a sample analysis result of each sample candidate answer; use the multiple sample questions, the multiple sample candidate answers corresponding to each sample question, and the sample analysis structures corresponding to each sample candidate answer to perform model training to obtain the first analysis model.

[0014] In a third aspect, the present application provides a computing device, including a memory and a processor; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the computing device is caused to execute the method described in the first aspect.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium, and the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method described in the first aspect.

[0016] As can be seen from the above technical solutions, the present application has at least the following beneficial effects: The present application provides a content analysis method. An analysis model is obtained through model training. The analysis model can analyze a target question and multiple candidate answers corresponding to the target question, obtain the analysis results of each candidate answer, and based on the analysis results of each candidate answer or the candidate answers corresponding to the analysis results output by the analysis model, so that users can quickly obtain the optimal answer. In addition, compared with large AI models, the amount of data called by the analysis model obtained by the content analysis method of the embodiments of the present application is relatively small. The analysis model can be stored in the user's local terminal. Even if the user's local terminal is disconnected from the cloud, the user can use the analysis model in the local terminal to obtain the optimal answer to the target question. Therefore, the analysis model in the embodiments of the present application has strong adaptability.

[0017] It should be understood that the description of technical features, technical solutions, beneficial effects or similar languages in the present application does not imply that all features and advantages can be achieved in any single embodiment. On the contrary, it can be understood that the description of features or beneficial effects means that specific technical features, technical solutions or beneficial effects are included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions and beneficial effects described in this embodiment can be combined in any appropriate manner. Those skilled in the art will understand that an embodiment can be implemented without one or more specific technical features, technical solutions or beneficial effects of a specific embodiment. In other embodiments, additional technical features and beneficial effects can also be identified in specific embodiments that do not embody all embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of obtaining the first analysis result corresponding to the candidate answer by using the first analysis model in the embodiment of the present application; Figure 2 It is a flowchart of the method for establishing the first analysis model in the embodiment of the present application; Figure 3 It is a flowchart of the determination method of the sample score corresponding to the sample candidate answer in the embodiment of the present application; Figure 4 It is a flowchart of the acquisition process of the training database corresponding to the first analysis model in the embodiment of the present application; Figure 5 It is a schematic diagram of a content analysis device provided by the embodiment of the present application; Figure 6 It is a schematic diagram of a computing device provided by the embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The terms "first", "second", "third", etc. in the description and drawings of this application are used to distinguish different objects, rather than to limit a specific order.

[0020] In the embodiments of this application, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of this application should not be construed as more preferred or more advantageous than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0021] For the sake of clear and concise description of the following embodiments, a brief introduction to the related technology is given first: An AI (Artificial Intelligence) large model refers to a deep learning model with a huge number of parameters, usually containing billions or even trillions of parameters. These models are trained through deep learning techniques and can process and generate various types of data such as natural language, images, and audio.

[0022] Users can interact with the AI large model, such as having conversations, asking questions, etc. When a user asks the AI large model the same question, the AI large model will output multiple different answers. However, when faced with multiple different answers, it is difficult for users to quickly judge the best answer from these answers.

[0023] In view of this, the embodiments of this application provide a content analysis method, which can be executed by a processing device. The processing device can be the user's local terminal, such as a mobile phone, a computer, a laptop, etc. Specifically, the method includes: the processing device obtains an analysis model through model training. The analysis model can analyze a target question and multiple candidate answers corresponding to the target question, obtain the analysis results of each candidate answer, and directly output the candidate answer corresponding to the best analysis result according to the analysis results of each candidate answer or the analysis model, so that the user can quickly obtain the optimal answer. In addition, compared with using the AI large model to assist the user in judging the optimal answer from multiple answers, the analysis model obtained by the content analysis method in the embodiments of this application calls a relatively small amount of data, and the analysis model can be stored in the user's local terminal. Even if the user's local terminal is disconnected from the cloud, the user can use the analysis model in the local terminal to obtain the optimal answer to the target question. Therefore, the analysis model in the embodiments of this application has strong adaptability.

[0024] The content analysis method provided by the embodiments of this application is introduced as follows. As Figure 1 shown, the method includes the following steps: S101. The processing device obtains a target question and multiple candidate answers corresponding to the target question.

[0025] The target question(s) can be proposed by the user, and the number of target questions can be one or more. When the user inputs one target question into the AI large model or inputs multiple target questions into the AI large model simultaneously as a single query to the AI large model, the AI large model outputs a candidate answer corresponding to each target question in the single query. The user makes multiple queries to the AI large model, and the AI large model outputs multiple different candidate answers for each target question. Among them, the user can query one AI large model or query multiple AI large models separately.

[0026] In addition, the user can also query a search engine for one or more target questions, and the user filters out multiple different candidate answers based on the search results of the search engine. The above is an illustrative example of the method for obtaining multiple candidate answers corresponding to the target question, and the embodiments of the present application do not limit this.

[0027] S102. The processing device inputs the target question and multiple candidate answers into the first analysis model to obtain the first analysis result of each candidate answer.

[0028] The first analysis result can be the score and comment of the candidate answer. Exemplarily, the first analysis model has an analysis criterion for the candidate answer. When the target question and the multiple candidate answers corresponding to the target question are input into the first analysis model, the first analysis model analyzes each candidate answer from different dimensions according to the analysis criterion and generates the score and comment corresponding to each candidate answer.

[0029] When the processing device obtains the target question and the multiple candidate answers corresponding to the target question, it can input the target question and the multiple candidate answers corresponding to the target question into the first analysis model. When the first analysis model outputs the first analysis result of each candidate answer, the user can compare based on the first analysis result of each candidate answer to obtain the desired candidate answer.

[0030] In some other embodiments, the processing device can compare the first analysis results of each candidate answer, determine the candidate answer corresponding to the optimal first analysis result as the target candidate answer, and the user directly obtains the target candidate answer without the user comparing the first analysis results of each candidate answer, that is, directly outputs the candidate answer corresponding to the best analysis result. Among them, an evaluation rule can be set in the first analysis model, and the first analysis model evaluates the first analysis results of each candidate answer according to the evaluation rule, and determines the first analysis result with the best evaluation as the optimal first analysis result.

[0031] The process of obtaining the first analysis model is introduced as follows. The process of obtaining the first analysis model includes two parts. The first part is the acquisition of sample data, and the second part is the model training using the sample data. As Figure 2 shown, the method includes the following steps: S1. The processing device determines a plurality of sample questions and a plurality of sample candidate answers corresponding to each sample question.

[0032] The sample questions can be the questions collected from users asking the AI large model, or other questions collected from the Internet. The plurality of sample candidate answers corresponding to the sample questions can be obtained by the user asking the sample questions to the AI large model multiple times, and the AI large model outputs a plurality of sample candidate answers, or the user queries professional knowledge books, consults professionals, etc. to obtain the sample questions and the plurality of candidate answers corresponding to the sample questions. The embodiments of the present application do not limit the sources of the sample questions and the plurality of sample candidate answers corresponding to each sample question.

[0033] S2. The processing device uses the AI large model to analyze the plurality of sample candidate answers corresponding to each sample question, and obtains the sample analysis results of each sample candidate answer.

[0034] Among them, the sample analysis results can include sample scores and sample comments.

[0035] Specifically, as Figure 3 shown, the sample score can be determined through the following steps: S21. The processing device uses the AI large model to score each sample candidate answer of each sample question separately from multiple dimensions, and obtains the first separate score of each sample candidate answer.

[0036] The determination method of the first separate score of each candidate answer is illustrated by the following example: Given a sample question and three sample candidate answers corresponding to the sample question, the three sample candidate answers are: the first sample candidate answer, the second sample candidate answer, and the third sample candidate answer.

[0037] In the AI large model, five dimensions are set for scoring each sample candidate answer: the first dimension: integrity, that is, whether the sample candidate answer answers all aspects of the sample question; the second dimension: accuracy, that is, whether the sample candidate answer is answered accurately; the third dimension: professionalism, that is, whether the expression of the sample candidate answer is professional; the fourth dimension: clarity, that is, whether the language of the sample candidate answer is logically clear and easy to understand; the fifth dimension: relevance, that is, whether the sample candidate answer is closely related to the sample question and there is no other redundant or irrelevant answer. Among them, the 5 dimensions are for illustrative introduction.

[0038] The scoring ranges corresponding to each dimension are as follows:

[0039] Sum up the scores of each dimension and calculate the score of each sample candidate answer according to the first calculation rule. The first calculation rule can be, for example, calculating the arithmetic mean or weighted mean of the scores of multiple dimensions, etc.

[0040] For example, the processing device inputs the first sample candidate answer into the AI large model and obtains a completeness score of 5, an accuracy score of 3, a professionalism score of 3, a clarity score of 2, and a relevance score of 4 for the first sample candidate answer. Then the score of the first sample candidate answer is: . Similarly, input the second sample candidate answer and the third sample candidate answer into the AI large model respectively to obtain the scores of the second sample candidate answer and the third sample candidate answer.

[0041] The processing device inputs each sample candidate answer into the AI large model each time, and names the score obtained for this sample candidate answer as the primary score. The processing device names the score calculated from multiple primary scores through the second calculation rule as the advanced score. The second calculation rule can be, for example, calculating the arithmetic mean or weighted mean of multiple primary scores, etc.

[0042] In one embodiment, the processing device inputs the first sample candidate answer into the AI large model only once to obtain the primary score of the first sample candidate answer, and this primary score serves as the first individual score of the first sample candidate answer; inputs the second sample candidate answer into the AI large model only once to obtain the primary score of the second sample candidate answer, and this primary score serves as the first individual score of the second sample candidate answer; inputs the third sample candidate answer into the AI large model only once to obtain the primary score of the third sample candidate answer, and this primary score serves as the first individual score of the third sample candidate answer.

[0043] In another embodiment, the processing device inputs the first sample candidate answer into the AI large model five times, obtaining five preliminary scores corresponding to the first sample candidate answer. The five preliminary scores are the first preliminary score, the second preliminary score, the third preliminary score, the fourth preliminary score, and the fifth preliminary score respectively. The first preliminary score, the second preliminary score, the third preliminary score, the fourth preliminary score, and the fifth preliminary score are compared in terms of score magnitude, and the highest score and the lowest score are removed. The remaining three preliminary scores are calculated according to the third calculation rule to obtain the advanced score of the first sample candidate answer. The third calculation rule can be, for example, calculating the arithmetic mean or weighted mean of multiple preliminary scores, etc. Similarly, the determination process of the advanced score of the second sample candidate answer and the determination process of the advanced score of the third sample candidate answer refer to the determination process of the advanced score of the first sample candidate answer. In this embodiment, the advanced scores corresponding to the first sample candidate answer, the second sample candidate answer, and the third sample candidate answer are used as the first individual scores corresponding to the first sample candidate answer, the second sample candidate answer, and the third sample candidate answer respectively.

[0044] In yet another embodiment, the processing device inputs the first sample candidate answer into three different AI large models respectively. In the case where the first sample candidate answer is input into the three different AI large models only once, three preliminary scores are obtained through the three AI large models. The three preliminary scores are calculated according to the fourth calculation rule to obtain the first individual score of the first sample candidate answer. The fourth calculation rule is, for example, calculating the arithmetic mean or weighted mean of multiple preliminary scores, etc. In the case where the first sample candidate answer is input into the three different AI large models multiple times, each AI large model outputs an advanced score of the first sample candidate answer. The three advanced scores are calculated according to the fifth calculation rule to obtain the first individual score of the first sample candidate answer. The fifth calculation rule is, for example, calculating the arithmetic mean or weighted mean of multiple advanced scores, etc. Similarly, the determination process of obtaining the corresponding first individual score of the second sample candidate answer by using three different AI large models, and the determination process of obtaining the corresponding first individual score of the third sample candidate answer by using three different AI large models, can refer to the determination process of obtaining the corresponding first individual score of the first sample candidate answer by using three different AI large models.

[0045] S22. The processing device uses the AI large model to score the combined sample candidate answers separately from multiple dimensions, obtaining the first combined score of each sample candidate answer.

[0046] Specifically, taking the first sample candidate answer, the second sample candidate answer and the third sample candidate answer as examples, the processing device combines the first sample candidate answer, the second sample candidate answer and the third sample candidate answer and inputs them into the AI ​​big model. The AI ​​big model scores the first sample candidate answer, the second sample candidate answer and the third sample candidate answer in the combination from five dimensions: completeness, accuracy, professionalism, clarity and relevance.

[0047] The first sample candidate answer, the second sample candidate answer and the third sample candidate answer can be combined according to the set rules. For example, the first sample candidate answer, the second sample candidate answer and the third sample candidate answer can be sorted and combined according to the rules of positive order, reverse order and random order, as shown in the following table:

[0048] The positive sequence combination, reverse sequence combination, random sequence combination 1, random sequence combination 2 and random sequence combination 3 are input into the AI ​​big model according to the set number of times. The score of each combination by the AI ​​big model is shown in the following table:

[0049] The setting number can be one or more times.

[0050] For example, when the number of times is set to one, the first sample candidate answer gets five scores. , , , , , the second sample candidate answer received five scores , , , , , the third sample candidate answer received five scores , , , , The processing device then calculates the five scores corresponding to the first sample candidate answer, the five scores corresponding to the second sample candidate answer, and the five scores corresponding to the third sample candidate answer according to the sixth calculation rule to obtain the first combined scores corresponding to the first sample candidate answer, the second sample candidate answer, and the third sample candidate answer. The sixth calculation rule is, for example, calculating the arithmetic mean or weighted mean of multiple scores.

[0051] For example, when the number of times is set to two, taking the first sample candidate answer as an example, the first sample candidate answer in the positive order combination gets two scores and , the processing equipment will and The score of the first sample candidate answer in the forward combination is calculated by summing and averaging ; the first sample candidate answer in the reverse combination gets two scores and , and and are summed and averaged to calculate the score of the first sample candidate answer in the reverse combination ; the first sample candidate answer in the first shuffled combination gets two scores and , and and are summed and averaged to calculate the score of the first sample candidate answer in the first shuffled combination ; the first sample candidate answer in the second shuffled combination gets two scores and , and and are summed and averaged to calculate the score of the first sample candidate answer in the second shuffled combination ; the first sample candidate answer in the third shuffled combination gets two scores and , and and are summed and averaged to calculate the score of the first sample candidate answer in the third shuffled combination ; finally, the processing device averages the five scores , , , , to obtain the first combination score. The determination process of the first combination score of the second sample candidate answer and the determination process of the first combination score of the third sample candidate answer refer to the determination process of the first combination score of the first sample candidate answer.

[0052] In addition, the processing device can also input the forward combination, reverse combination, first shuffled combination, second shuffled combination, and third shuffled combination into multiple large AI models according to the set number of times. Each large AI model scores the first sample candidate answer, second sample candidate answer, and third sample candidate answer respectively. Finally, the scores of the first sample candidate answer in each large AI model are calculated according to the seventh calculation rule to obtain the first combination score of the first sample candidate answer, the scores of the second sample candidate answer in each large AI model are calculated according to the seventh calculation rule to obtain the first combination score of the second sample candidate answer, and the scores of the third sample candidate answer in each large AI model are calculated according to the seventh calculation rule to obtain the first combination score of the third sample candidate answer. The seventh calculation rule is, for example, to calculate the arithmetic mean or weighted mean of multiple scores, etc.

[0053] In this embodiment, the processing device combines multiple sample candidate answers according to the sorting rule, which can reduce the influence of the order of the multiple sample candidate answers on the scoring of each sample candidate answer.

[0054] S23. The processing device averages the first individual score and the first combined score of each sample candidate answer to obtain the first-stage score.

[0055] Specifically, taking the first sample candidate answer, the second sample candidate answer, and the third sample candidate answer as examples, the processing device averages the first individual score and the first combined score of the first sample candidate answer to obtain the first-stage score of the first sample candidate answer; averages the first individual score and the first combined score of the second sample candidate answer to obtain the first-stage score of the second sample candidate answer; averages the first individual score and the first combined score of the third sample candidate answer to obtain the first-stage score of the third sample candidate answer. Among them, averaging the first individual score and the first combined score can be to sum and average the first individual score and the first combined score, or to weight and average the first individual score and the first combined score.

[0056] S24. The processing device adjusts the first-stage score of the sample candidate answer according to the similarity between multiple sample candidate answers corresponding to the sample question to obtain the second-stage score.

[0057] In a specific embodiment, the method for determining the second-stage score includes the following steps: S241. The processing device determines the highest-score sample answer and the lowest-score sample answer from multiple sample candidate answers corresponding to the sample question. The first-stage score of the highest-score sample answer is the highest, and the first-stage score of the lowest-score sample answer is the lowest.

[0058] Specifically, taking the first sample candidate answer, the second sample candidate answer, and the third sample candidate answer as examples, assuming that the first-stage score of the first sample candidate answer is greater than the first-stage score of the second sample candidate answer, and the first-stage score of the second sample candidate answer is greater than the first-stage score of the third sample candidate answer, then the processing device determines the first sample candidate answer as the highest-score sample answer and the third sample candidate answer as the lowest-score sample answer.

[0059] S242. The processing device adjusts the first-stage score of the first other sample candidate answer according to the similarity between the first other sample candidate answer and the highest-score sample answer among multiple sample candidate answers to obtain the first adjusted score of the first other sample candidate answer.

[0060] Specifically, using the sample answer with the highest score, i.e., the first sample candidate answer, as the reference answer, the first-stage scores of the second sample candidate answer and the third sample candidate answer are adjusted respectively to obtain the first adjusted scores of the second sample candidate answer and the third sample candidate answer. The second sample candidate answer and the first sample candidate answer form the first answer pair, and the third sample candidate answer and the first sample candidate answer form the second answer pair. The processing device inputs the first answer pair and the second answer pair into the AI large model respectively; for the first answer pair, the AI large model compares the similarities between the second sample candidate answer and the first sample candidate answer in each dimension of integrity, accuracy, professionalism, clarity, and relevance; in each dimension, if the similarity between the second sample candidate answer and the first sample candidate answer is high, the score of the second sample candidate answer in that dimension is increased, and if the similarity between the second sample candidate answer and the first sample candidate answer is low, the score of the second sample candidate answer in that dimension is decreased; after the scores of the second sample candidate answer in each dimension are adjusted, the adjusted scores in each dimension are aggregated and calculated to obtain the first adjusted score of the second sample candidate answer. For the second answer pair, the method for determining the first adjusted score of the third sample candidate answer refers to the method for determining the first adjusted score of the second sample candidate answer.

[0061] S243. The processing device adjusts the first-stage score of the second other sample candidate answer according to the similarity between the second other sample candidate answer (excluding the sample answer with the lowest score) among the multiple sample candidate answers and the sample answer with the lowest score, to obtain the second adjusted score of the second other sample candidate answer.

[0062] Specifically, taking the lowest-score sample answer, i.e., the third sample candidate answer, as the reference answer, the first-stage scores of the first sample candidate answer and the second sample candidate answer are adjusted respectively to obtain the second adjusted scores of the first sample candidate answer and the second sample candidate answer. The first sample candidate answer and the third sample candidate answer form the third answer pair, and the second sample candidate answer and the third sample candidate answer form the fourth answer pair. The third answer pair and the fourth answer pair are respectively input into the AI large model; for the third answer pair, the AI large model compares the similarities between the first sample candidate answer and the third sample candidate answer in each dimension of integrity, accuracy, professionalism, clarity, and relevance; in each dimension, if the similarity between the first sample candidate answer and the third sample candidate answer is high, the score of the first sample candidate answer in this dimension is lowered, and if the similarity between the first sample candidate answer and the third sample candidate answer is low, the score of the first sample candidate answer in this dimension is raised; after the scores of the first sample candidate answer in each dimension are adjusted, the adjusted scores in each dimension are aggregated and calculated to obtain the second adjusted score of the first sample candidate answer. For the fourth answer pair, the method for determining the second adjusted score of the second sample candidate answer refers to the method for determining the second adjusted score of the first sample candidate answer.

[0063] S244. The processing device adjusts the first-stage scores corresponding to the selected sample candidate answers according to the similarities between any two sample candidate answers among the multiple sample candidate answers to obtain the third adjusted scores.

[0064] Specifically, taking the second sample candidate answer as the reference answer, the first-stage scores of the first sample candidate answer and the third sample candidate answer are adjusted respectively to obtain the third adjusted scores of the first sample candidate answer and the third sample candidate answer.

[0065] The first sample candidate answer and the second sample candidate answer form the fifth answer pair, and the third sample candidate answer and the second sample candidate answer form the sixth answer pair. The processing device inputs the fifth answer pair and the sixth answer pair into the AI large model respectively; for the fifth answer pair, the AI large model compares the scores of the first sample candidate answer and the second sample candidate answer in each dimension of integrity, accuracy, professionalism, clarity, and relevance; in each dimension, if the score of the first sample candidate answer in this dimension is greater than or equal to the score of the second sample candidate answer in this dimension, there is no need to adjust the score of the first sample candidate answer in this dimension, and if the score of the first sample candidate answer in this dimension is less than the score of the second sample candidate answer in this dimension, the score of the first sample candidate answer in this dimension is lowered; after the scores of the first sample candidate answer in each dimension are determined, the determined scores in each dimension are aggregated and calculated to obtain the third adjusted score of the first sample candidate answer.

[0066] For the sixth answer pair, the AI large model compares the scores of the third sample candidate answer and the second sample candidate answer under each dimension of integrity, accuracy, professionalism, clarity, and relevance; under each dimension, if the score of the third sample candidate answer under this dimension is greater than the score of the second sample candidate answer under this dimension, then increase the score of the third sample candidate answer under this dimension, and if the score of the third sample candidate answer under this dimension is less than or equal to the score of the second sample candidate answer under this dimension, then there is no need to adjust the score of the third sample candidate answer under this dimension; after the scores of the third sample candidate answer under each dimension are determined, the determined scores under each dimension are aggregated and calculated to obtain the third adjusted score of the third sample candidate answer.

[0067] According to the above embodiments, the first-stage scores of the first sample candidate answer are adjusted twice to obtain the second adjusted score and the third adjusted score of the first sample candidate answer respectively, the first-stage scores of the second sample candidate answer are adjusted twice to obtain the first adjusted score and the second adjusted score of the second sample candidate answer respectively, and the first-stage scores of the third sample candidate answer are adjusted twice to obtain the first adjusted score and the third adjusted score of the third sample candidate answer respectively.

[0068] In some embodiments, after obtaining the first adjusted score, the second adjusted score, and the third adjusted score of each sample candidate answer, the second-stage score of each sample candidate answer can be determined.

[0069] Among them, in the case where the sample candidate answer is the highest-score sample answer, according to the second adjusted score and the third adjusted score of the highest-score sample answer, the second-stage score of the highest-score sample answer is determined, and the highest-score sample answer does not have a corresponding first adjusted score; by way of example, the first sample candidate answer is the highest-score sample answer, and the second adjusted score and the third adjusted score of the first sample candidate answer are calculated according to the eighth calculation rule to obtain the second-stage score of the first sample candidate answer.

[0070] In the case where the sample candidate answer is other sample answers except the highest-score sample answer and the lowest-score sample answer, according to the first adjusted score, the second adjusted score, and the third adjusted score of the other sample answers, the second-stage score of the other sample answers is determined; by way of example, the second sample candidate answer is other sample answers except the highest-score sample answer and the lowest-score sample answer, and the first adjusted score, the second adjusted score, and the third adjusted score of the second sample candidate answer are calculated according to the eighth calculation rule to obtain the second-stage score of the second sample candidate answer.

[0071] In the case where the sample candidate answer is the lowest-scoring sample answer, the second-stage score of the lowest-scoring sample answer is determined according to the first adjusted score and the third adjusted score of the lowest-scoring sample answer; illustratively, the third sample candidate answer is the lowest-scoring sample answer, and the first adjusted score and the third adjusted score of the third sample candidate answer are calculated according to the eighth calculation rule to obtain the second-stage score of the third sample candidate answer. The eighth calculation rule is, for example, calculating the arithmetic mean or weighted mean of multiple scores.

[0072] In addition, multiple AI large models can be used, and the first answer pair, the second answer pair, the third answer pair, the fourth answer pair, the fifth answer pair, and the sixth answer pair can be input into each AI large model for multiple times. In each AI large model, multiple second adjusted scores and multiple third adjusted scores corresponding to the first sample candidate answer, multiple first adjusted scores and multiple second adjusted scores corresponding to the second sample candidate answer, and multiple first adjusted scores and multiple third adjusted scores corresponding to the third sample candidate answer are obtained respectively; taking the second adjusted score and the third adjusted score corresponding to the first sample candidate answer as an example, the process is as follows: For each AI big model, the multiple second adjusted scores and multiple third adjusted scores of the first sample candidate answer are calculated according to the ninth calculation rule to obtain the final second adjusted score and third adjusted score of the first sample candidate answer in the AI ​​big model; finally, the final second adjusted score and third adjusted score of the first sample candidate answer in each AI big model are summarized, and the corresponding second adjusted score and third adjusted score of the first sample candidate answer are obtained by using multiple AI big models according to the ninth calculation rule. The ninth calculation rule is, for example, to calculate the arithmetic mean or weighted mean of multiple scores.

[0073] Similarly, the second sample candidate answer uses multiple AI large models to obtain the corresponding first adjusted score and second adjusted score, and the third sample candidate answer uses multiple AI large models to obtain the corresponding first adjusted score and third adjusted score, with reference to the process of obtaining the corresponding second adjusted score and third adjusted score of the first candidate answer using multiple AI large models.

[0074] S25. Take the average of the first-stage score and the second-stage score of each sample candidate answer as the sample score of each sample candidate answer.

[0075] Specifically, the average of the first-stage score and the second-stage score may be an arithmetic average or a weighted average of the two.

[0076] After determining the sample scores of each sample candidate answer, the sample comments of each sample candidate answer can be determined in the following manner: Using an AI large model, based on the sample scores of the sample candidate answers corresponding to the sample questions, generate sample comments for the sample candidate answers in multiple dimensions.

[0077] S3. Use multiple sample questions, multiple sample candidate answers corresponding to each sample question, and the sample analysis results corresponding to each sample candidate answer for model training to obtain a first analysis model.

[0078] Specifically, multiple sample questions, multiple sample candidate answers corresponding to each sample question, and the sample analysis results corresponding to each sample candidate answer can be collected and a training database can be created, and the training database can be used for model training to obtain a first analysis model.

[0079] As Figure 4 shown, the process of obtaining the training database corresponding to the first analysis model is as follows: S31. The processing device collects multiple sample candidate answers and the sample scores corresponding to each sample candidate answer and creates a scoring database.

[0080] The processing device collects multiple sample candidate answers and the sample scores corresponding to each sample candidate answer, forms scoring data, and stores the scoring data in the scoring database.

[0081] S32. The processing device uses the AI large model to analyze the sample scores corresponding to each sample candidate answer in the scoring database to obtain the sample comments corresponding to each sample candidate answer.

[0082] The processing device can obtain the sample scores of multiple sample candidate answers corresponding to the sample questions from the scoring database. According to some of the above embodiments, the sample score of each sample candidate answer is the AI large model scoring each sample candidate answer from multiple dimensions and then fusing and calculating the scores of multiple dimensions. Therefore, after knowing the sample score of each sample candidate answer, the score of the sample candidate answer in each dimension can be deduced reversely. According to the score of each dimension, the highlights and deficiencies of the sample candidate answer in each dimension can be identified, and then the sample comments of the sample candidate answer in each dimension can be obtained. For example, if the first sample candidate answer has a high score in the integrity dimension, it is considered that the first sample candidate answer covers all key information; if the first sample candidate answer has a low score in the clarity dimension, it is considered that the language expression of the first sample candidate answer has defects such as logical confusion and difficulty in understanding.

[0083] According to some of the above embodiments, in the process of obtaining the sample scores of each sample candidate answer, it is necessary to calculate the first individual score and the first combined score of each sample candidate answer, then fuse and calculate the first individual score and the first combined score to obtain the first-stage score, and then adjust the first-stage score of each sample candidate answer according to the similarity between multiple sample candidate answers to obtain the second-stage score. The first-stage score and the second-stage score are fused and calculated to obtain the sample score of each sample candidate answer. Therefore, the reliability of the sample score of each sample candidate answer is relatively high, and the reliability of the corresponding sample comment is relatively high.

[0084] S33. The processing device collects multiple sample questions, multiple sample candidate answers corresponding to each sample question, sample scores and sample comments corresponding to each sample candidate answer, and creates a training database.

[0085] The processing device collects multiple sample questions, multiple sample candidate answers corresponding to each sample question, sample scores and sample comments corresponding to each sample candidate answer, and forms a training database.

[0086] In order to improve the accuracy of the first analysis result output by the first analysis model for each candidate answer, the content analysis method provided in the embodiments of the present application further includes: using an AI large model to analyze the target question and multiple candidate answers corresponding to the target question to obtain the second analysis result of each candidate answer; comparing the first analysis result and the second analysis result of each candidate answer to obtain multiple comparison results; if there is a target comparison result, then using the second analysis result of the target question and multiple candidate answers corresponding to the target question to update the first analysis model to obtain a second analysis model; the target comparison result is a comparison result in which the similarity between the first analysis result and the second analysis result of the candidate answer is lower than a preset similarity threshold.

[0087] Specifically, the first analysis result includes a first score, and the second analysis result includes a second score. The AI large model subtracts the second score of each candidate answer from the first score. If the difference between the two is within the threshold range, there is no target comparison result, and the first analysis model does not need to be updated. If the difference between the two is outside the threshold range, there is a target comparison result. Analyze in which aspects the first analysis result has defects, and formulate corresponding solutions for the existing defects to update the first analysis model to obtain the second analysis model. When deploying the first analysis model on the user's local terminal in the case of the same set of data (questions and corresponding multiple answers), the similarity between the first analysis result given by the first analysis model and the second analysis result given by the cloud AI large model is poor, indicating that the accuracy of the first analysis model deployed on the user's local terminal is no longer reliable. That is, the first analysis model needs to be updated. At this time, the second analysis result output by the AI large model for the target question can be used to optimize and train the first model to obtain the second analysis model.

[0088] Based on the above description, the embodiments of the present application have the following beneficial effects: The content analysis method provided by the embodiments of the present application obtains the first analysis model through model training. The first analysis model can analyze the target question and multiple candidate answers corresponding to the target question, obtain the first analysis result of each candidate answer, and directly output the candidate answer corresponding to the best first analysis result according to the first analysis results of each candidate answer or the first analysis model, so that the user can quickly obtain the optimal answer. In addition, compared with using the AI large model to assist the user in judging the optimal answer from multiple answers, the first analysis model obtained by the content analysis method of the embodiments of the present application calls a relatively small amount of data, and the first analysis model can be stored in the user's local terminal. Even if the user's local terminal is disconnected from the cloud, the user can use the first analysis model in the local terminal to obtain the optimal answer to the target question. Therefore, the first analysis model in the embodiments of the present application has strong adaptability.

[0089] As described above in conjunction with Figures 1 to 4 the content analysis method provided by the embodiments of the present application has been introduced in detail. Next, the content analysis device provided by the embodiments of the present application will be introduced in conjunction with the attached Figure 5 drawings.

[0090] Such as Figure 5As shown in the figure, an embodiment of the present application provides a content analysis device, which includes an acquisition module 201 and an analysis module 202. The acquisition module 201 is used to acquire a target question and multiple candidate answers corresponding to the target question. The analysis module 202 is used to input the target question and multiple candidate answers into a first analysis model to obtain a first analysis result for each candidate answer. Among them, the first analysis model is trained in the following way: determine multiple sample questions and multiple sample candidate answers corresponding to each sample question; use the AI large model to analyze multiple sample candidate answers corresponding to each sample question to obtain a sample analysis result for each sample candidate answer; use multiple sample questions, multiple sample candidate answers corresponding to each sample question, and the sample analysis structure corresponding to each sample candidate answer to perform model training to obtain the first analysis model.

[0091] Optionally, the analysis module 202 is further used to input the target question and multiple candidate answers into the AI large model to obtain a second analysis result for each candidate answer; compare the first analysis result and the second analysis result of each candidate answer to obtain multiple comparison results. The content analysis device further includes an update module. The update module is used to update the first analysis model with the second analysis result of the target question and the multiple candidate answers corresponding to the target question when there is a target comparison result to obtain a second analysis model. The target comparison result is a comparison result where the similarity between the first analysis result and the second analysis result of the candidate answer is lower than the preset similarity threshold.

[0092] Optionally, the content analysis device further includes a determination module, and the determination module is used to determine a target candidate answer according to the first analysis result of each candidate answer.

[0093] Optionally, the sample analysis result of each sample candidate answer obtained by the analysis module 202 includes a sample score and a sample comment.

[0094] Optionally, the analysis module 202 is specifically used to: use the AI large model to score each sample candidate answer of each sample question separately from multiple dimensions to obtain a first separate score for each sample candidate answer; use the AI large model to score the combined sample candidate answers separately from multiple dimensions to obtain a first combined score for each sample candidate answer; average the first separate score and the first combined score of each sample candidate answer to obtain a first-stage score; adjust the first-stage score of the sample candidate answer according to the similarity between multiple sample candidate answers corresponding to the sample question to obtain a second-stage score; use the average of the first-stage score and the second-stage score of each sample candidate answer as the sample score of each sample candidate answer.

[0095] Optionally, the analysis module 202 is further configured to: determine the highest-scoring sample answer and the lowest-scoring sample answer from multiple sample candidate answers corresponding to a sample question, where the first-stage score of the highest-scoring sample answer is the highest and the first-stage score of the lowest-scoring sample answer is the lowest; adjust the first-stage score of the first other sample candidate answer based on the similarity between the first other sample candidate answer and the highest-scoring sample answer among the multiple sample candidate answers, to obtain the first adjusted score of the first other sample candidate answer; adjust the first-stage score of the second other sample candidate answer based on the similarity between the second other sample candidate answer and the lowest-scoring sample answer among the multiple sample candidate answers, to obtain the second adjusted score of the second other sample candidate answer; adjust the first-stage score corresponding to the selected sample candidate answer based on the similarity between any two selected sample candidate answers among the multiple sample candidate answers, to obtain the third adjusted score; in the case where the sample candidate answer is the highest-scoring sample answer, determine the second-stage score of the highest-scoring sample answer based on the second adjusted score and the third adjusted score of the highest-scoring sample answer; in the case where the sample candidate answer is the lowest-scoring sample answer, determine the second-stage score of the lowest-scoring sample answer based on the first adjusted score and the third adjusted score of the lowest-scoring sample answer; and in the case where the sample candidate answer is an other sample answer other than the highest-scoring sample answer and the lowest-scoring sample answer, determine the second-stage score of the other sample answer based on the first adjusted score, the second adjusted score, and the third adjusted score of the other sample answer.

[0096] Optionally, the analysis module 202 is further configured to: use an AI large model to generate sample comments for the sample candidate answers in multiple dimensions based on the sample scores of the sample candidate answers corresponding to the sample question.

[0097] The content analysis device according to the embodiments of the present application can correspond to execute the methods described in the embodiments of the present application, and the above other operations and / or functions of each module / unit of the content analysis device are respectively for implementing Figures 1 to 4 the corresponding processes of the respective methods in the illustrated embodiments. For the sake of brevity, they are not described herein again.

[0098] Embodiments of the present application further provide a computing device. As Figure 6 shown, this figure is a schematic diagram of a computing device provided by an embodiment of the present application. The computing device 300 includes a bus 301, a processor 302, a communication interface 303, and a memory 304. The processor 302, the memory 304, and the communication interface 303 communicate with each other through the bus 301.

[0099] The bus 301 can be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 only a thick line is used in Figure 6 , but it does not mean that there is only one bus or one type of bus.

[0100] The processor 302 can be any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0101] The communication interface 303 is used for external communication.

[0102] The memory 304 can include volatile memory, such as random access memory (RAM). The memory 304 can also include non-volatile memory, such as read-only memory (ROM), flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0103] Executable code is stored in the memory 304, and the processor 302 executes the executable code to perform the foregoing content analysis method.

[0104] Specifically, in the case of implementing Figure 5 the embodiments shown, and Figure 5 when each module or unit of the content analysis device described in the embodiments is implemented by software, the software or program code required to execute the functions of each module / unit in Figure 5 can be partially or fully stored in the memory 304. The processor 302 executes the program code corresponding to each unit stored in the memory 304 to perform the foregoing content analysis method.

[0105] ​​The embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid-state hard disk). The computer-readable storage medium includes instructions that instruct the computing device to perform the above-mentioned content analysis method.

[0106] The embodiment of the present application further provides a computer program product, which includes one or more computer instructions. When the computer instructions are loaded and executed on a computing device, the process or function described in the embodiment of the present application is generated in whole or in part.

[0107] The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer or data center to another website, computer or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0108] When the computer program product is executed by a computer, the computer executes any of the aforementioned content analysis methods. The computer program product may be a software installation package, and when any of the aforementioned content analysis methods is needed, the computer program product may be downloaded and executed on a computer.

[0109] The descriptions of the processes or structures corresponding to the above-mentioned figures have different emphases. For parts that are not described in detail in a certain process or structure, please refer to the relevant descriptions of other processes or structures.

[0110] The above description is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application should be included in the protection scope of the present application.

Claims

1. A content analysis method, characterized in that, The method includes: Obtaining a target question and multiple candidate answers corresponding to the target question; Inputting the target question and the multiple candidate answers into a first analysis model to obtain a first analysis result for each candidate answer; Wherein, the first analysis model is trained in the following manner: Determining multiple sample questions and multiple sample candidate answers corresponding to each sample question; Using an AI large model to analyze the multiple sample candidate answers corresponding to each sample question to obtain a sample analysis result for each sample candidate answer; Using the multiple sample questions, the multiple sample candidate answers corresponding to each sample question, and the sample analysis results corresponding to each sample candidate answer for model training to obtain the first analysis model.

2. The method according to claim 1, wherein The method further includes: Using the AI large model to analyze the target question and the multiple candidate answers corresponding to the target question to obtain a second analysis result for each candidate answer; If there is a target comparison result, then using the target question and the second analysis results of the multiple candidate answers corresponding to the target question to update the first analysis model to obtain a second analysis model; the target comparison result is a comparison result where the similarity between the first analysis result and the second analysis result of the candidate answer is lower than a preset similarity threshold.

3. The method according to claim 1, characterized in that, The method further includes: Determining a target candidate answer according to the first analysis result of each candidate answer.

4. The method according to claim 1, characterized in that, The sample analysis result includes a sample score and a sample comment.

5. The method according to claim 4, characterized in that, The sample score is determined in the following manner: Using the AI large model to score each sample candidate answer of each sample question separately from multiple dimensions to obtain a first separate score for each sample candidate answer; Using the AI large model to score the combined sample candidate answers separately from multiple dimensions to obtain a first combined score for each sample candidate answer; Averaging the first separate score and the first combined score of each sample candidate answer to obtain a first-stage score; Adjusting the first-stage score of the sample candidate answer according to the similarity between the multiple sample candidate answers corresponding to the sample question to obtain a second-stage score; Taking the average of the first-stage score and the second-stage score of each sample candidate answer as the sample score of each sample candidate answer.

6. The method according to claim 5, characterized in that The adjusting the first-stage score of the sample candidate answer according to the similarity between the multiple sample candidate answers corresponding to the sample question to obtain a second-stage score includes: Determining a highest-score sample answer and a lowest-score sample answer from the multiple sample candidate answers corresponding to the sample question, where the first-stage score of the highest-score sample answer is the highest and the first-stage score of the lowest-score sample answer is the lowest; Adjusting the first-stage score of the first other sample candidate answer according to the similarity between the first other sample candidate answer and the highest-score sample answer among the multiple sample candidate answers to obtain a first adjusted score of the first other sample candidate answer; Adjust the first-stage score of the second other sample candidate answer among the multiple sample candidate answers except the lowest-score sample answer according to the similarity between the second other sample candidate answer and the lowest-score sample answer, to obtain the second adjusted score of the second other sample candidate answer; Adjust the first-stage score corresponding to the selected sample candidate answer according to the similarity between any two sample candidate answers among the multiple sample candidate answers, to obtain the third adjusted score; In the case where the sample candidate answer is the highest-score sample answer, determine the second-stage score of the highest-score sample answer according to the second adjusted score and the third adjusted score of the highest-score sample answer; In the case where the sample candidate answer is the lowest-score sample answer, determine the second-stage score of the lowest-score sample answer according to the first adjusted score and the third adjusted score of the lowest-score sample answer; In the case where the sample candidate answer is an other sample answer other than the highest-score sample answer and the lowest-score sample answer, determine the second-stage score of the other sample answer according to the first adjusted score, the second adjusted score and the third adjusted score of the other sample answer.

7. The method according to claim 4, wherein The sample comment is determined by the following method: Using the AI large model, based on the sample scores of the sample candidate answers corresponding to the sample questions, generate sample comments of the sample candidate answers in multiple dimensions.

8. A content analysis device, characterized in that, The device includes: An acquisition module, configured to acquire a target question and multiple candidate answers corresponding to the target question; An analysis module, configured to input the target question and multiple candidate answers into a first analysis model to obtain a first analysis result of each candidate answer; wherein, the first analysis model is trained by the following method: determine multiple sample questions and multiple sample candidate answers corresponding to each sample question; use the AI large model to analyze the multiple sample candidate answers corresponding to each sample question to obtain a sample analysis result of each sample candidate answer; use the multiple sample questions, the multiple sample candidate answers corresponding to each sample question, and the sample analysis structure corresponding to each sample candidate answer to perform model training to obtain the first analysis model.

9. A computing device, characterized in that, It includes a memory and a processor; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the computing device is caused to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.

Images