Intelligent teaching content generation method for red culture transmission of courses in colleges and universities

Abstract

The invention discloses an intelligent teaching content generation method for red culture transmission of college courses, and particularly relates to the technical field of education, and the method comprises the following steps: obtaining current red culture teaching content and questions of all students, summarizing the questions of target students into a question set, and matching the question set with the teaching content to obtain a preliminary matching content set; carrying out matching degree analysis on the content set, and judging whether a teaching content curriculation hidden danger exists or not; if the hidden danger exists, generating a score audit index and a measurement index, dividing the problems into high-matching-degree problems and low-matching-degree problems according to the indexes, and eliminating the low-matching-degree problems; and finally, based on a customized content generation strategy, regenerating red culture teaching content suitable for the target student. According to the method, customized red culture teaching content can be regenerated, the learning interest and participation degree of students can be effectively improved, and the key points of the teaching content can be dynamically adjusted based on the questioning question set and the fuzzy logic reasoning result of the students.

Description

Technical Field

[0001] The present invention relates to the field of educational technology, and more specifically, to a method for generating intelligent teaching content for red culture dissemination in college courses. Background Art

[0002] . With the continuous development of modern education, the role of red culture in college courses has become increasingly prominent. Especially in the field of ideological and political education, the teaching content of red culture is not only the teaching of historical events, but also an important part of educating students on patriotism and collectivism.

[0003] However, under the traditional teaching model, the dissemination of red culture is often too single and cannot fully mobilize students' enthusiasm and sense of participation, especially when students have questions about the details in the teaching content, they cannot get timely answers, which may cause some students to misinterpret or misunderstand the red cultural content. With the development of modern educational technology, it is possible to generate intelligent teaching content based on students' personalized needs by using artificial intelligence and big data technology. By analyzing students' questions and customizing the teaching content suitable for their understanding ability and interests, the quality of red culture teaching can be effectively improved, students' understanding and recognition of red culture can be enhanced, and a good value orientation can be established. Summary of the invention

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A method for generating intelligent teaching content for red culture dissemination in college courses, comprising the following steps:

[0006] The current teaching content of red culture and the questions asked by all students are obtained respectively, and then one of the students is set as the target student, the questions asked by the target student are summarized to obtain a question set, and the questions are matched with the current teaching content of red culture to obtain a preliminary matching content set;

[0007] Based on the preliminary matching content set, a matching degree analysis is conducted, and the matching degree analysis results are used to identify whether there is a hidden danger of misinterpretation of the teaching content;

[0008] If there is a hidden danger of misinterpretation of the teaching content, the preliminary matching content set is subjected to misinterpretation review and measurement operations, and a misinterpretation review index and a misinterpretation measurement index are generated respectively. Then, each question in the preliminary matching content set is divided into a high-matching question and a low-matching question based on the misinterpretation review index and the misinterpretation measurement index.

[0009] Eliminate low-matching questions and include all high-matching questions in the customized content generation strategy. If there is no risk of misinterpretation of teaching content, identify each question in the preliminary matching content set as a high-matching question and include them in the customized content generation strategy.

[0010] Based on customized content generation strategies, the current teaching content of red culture is regenerated with target students as the object.

[0011] In a preferred embodiment, performing matching analysis refers to:

[0012] Through the semantic deviation detection model, semantic deviation detection is performed on each judgment question in the preliminary matching content set to generate a semantic deviation value, and then the average and standard deviation of all semantic deviation values ​​are calculated.

[0013] In a preferred embodiment, identifying whether there is a hidden danger of misinterpretation of teaching content according to the matching analysis result means:

[0014] Compare the mean value and standard deviation with the preset standard value one and standard value two respectively. If the mean value meets the conditions that the mean value is less than or equal to the preset standard value one and the standard deviation meets the conditions that the mean value is less than or equal to the preset standard value one and the standard deviation meets the conditions that the mean value is less than or equal to the preset standard value one and the standard deviation meets the conditions that the teaching content is misinterpreted. If the mean value meets the conditions that the mean value is less than or equal to the preset standard value one and the standard deviation meets the conditions that the teaching content is misinterpreted.

[0015] In a preferred embodiment, the semantic deviation detection model refers to:

[0016] S(Q i ) represents the i-th question Q i The norm of the semantic representation vector of the current red culture teaching content C, S(C) represents the norm of the semantic representation vector of the current red culture teaching content C, R(Q i ,C) represents the i-th question Q i The semantic deviation value between the current teaching content C of red culture.

[0017] In a preferred embodiment, the logic for obtaining the distortion audit index is:

[0018] Match each question Qi with the teaching content C respectively, and introduce irrelevant teaching content fragments as negative samples, namely negative sample content fragments C neg , and then use a two-layer nested semantic vector model to calculate the similarity. The first layer uses a context-based word embedding model, and the second layer uses a sentence embedding model. The semantic vector model is:

[0019] Indicates the i-th question Q i The k-th word embedding vector, E(C j ) represents the jth word embedding vector of all words in the current red culture teaching content C, E(Cneg j ) represents the negative sample content segment C neg The jth word embedding vector of all words in the question Qi and the teaching content C are respectively the number of words in the question Qi and the teaching content C, and the negative sample content segment C is neg The number of words in the content is the same as the number of words in the teaching content C. Sdev(Q i ,C) represents the i-th question Q i The similarity deviation with the current red culture teaching content C;

[0020] Calculate the keyword matching rate deviation between the question and the teaching content. The formula is:

[0021] kj represents the jth keyword, wj represents the TF-IDF weight preset for the jth keyword kj in the teaching content C, kj∈Q i Indicates whether the keyword kj appears in the question. If it exists, the value is 1, otherwise it is 0. u is the total number of keywords. Kdev(Q i ,C) indicates asking question Q i The keyword matching rate deviation with the current red culture teaching content C;

[0022] The calculation formula of the misinterpretation audit index is:

[0023] QS=log e (α1*Sdev(Q i ,C)+α2*Kdev(Q i ,C)+1); α1 and α2 are both preset audit coefficients, and QS represents the distortion audit index.

[0024] In a preferred embodiment, the logic for obtaining the distortion measurement index is:

[0025] Get the questions Q respectively i The positive, negative and neutral emotional scores of the current red culture teaching content C are calculated, and the same type of emotional scores are processed by absolute value after difference calculation. Then the absolute value processing results of the three types of emotional scores are averaged to obtain the emotional tendency deviation value Tdev (Q i ,C);

[0026] Get Question Q i The first context window C of the current red culture teaching content C l The correlation Rel(Q i,C), and then perform the following calculation:

[0027] dl represents the context window C l Ask questions i The distance in the teaching content, wl represents the context window C l The influence coefficient in the teaching content, λ is the attenuation coefficient, which is used to control the distance to the correlation Rel(Q i ,C), p is the influence degree of the context window C l The total number of Cdev(Q i ,C) represents the contextual value;

[0028] The calculation formula of the distortion index is:

[0029] QH=β1*Tdev(Q i ,C)+β2*(1-Cdev(Q i ,C)); β1 and β2 are both preset measurement coefficients, and QH represents the distortion measurement index.

[0030] In a preferred embodiment, dividing each question in the preliminary matching content set into high-matching questions and low-matching questions based on the distortion review index and the distortion measurement index means using fuzzy logic, treating the distortion review index and the distortion measurement index of each question as a set of input variables, and the matching type of the question and the content as the output variable, fuzzifying the input variables, converting the values ​​of the input variables into fuzzy sets, fuzzifying the output variables, converting the output variables into fuzzy sets, formulating fuzzy rules to describe the matching suitability of questions and contents under different data type combinations, and reasoning the fuzzified input variables through fuzzy rules to obtain the matching type of the question and the content.

[0031] Technical effects and advantages of the present invention:

[0032] The present invention analyzes the questions asked by target students and generates intelligent content based on the matching degree between the questions and the teaching content and the hidden dangers of misinterpretation. The present invention regenerates customized red culture teaching content according to the needs, interests and understanding ability of each student, making the teaching more targeted and personalized, and can effectively improve students' learning interest and participation.

[0033] Through the misinterpretation review index and the misinterpretation measurement index, the present invention can automatically identify potential misinterpretations in students' questions, and analyze them according to fuzzy logic to eliminate low-matching questions. The present invention reviews and measures questions that may be misinterpreted to ensure that the final generated teaching content can accurately convey the core values ​​of red culture and avoid misunderstandings and deviations in students' learning process.

[0034] The present invention can dynamically adjust the focus of teaching content based on the students' question sets and fuzzy logic reasoning results. For questions with a high degree of matching, the present invention will provide a more detailed explanation and allocate data volume based on the fuzzy logic representation of the question. This dynamic adjustment strategy makes the teaching content more in line with students' learning needs and can more flexibly respond to different students' different levels of understanding of red culture.

[0035] Through intelligent analysis and customized content generation, the present invention can automatically generate personalized teaching plans for each student without additional intervention by teachers, thereby improving teaching efficiency. At the same time, in-depth explanation of students' specific problems can effectively help students understand the connotation of red culture more deeply and enhance their mastery and recognition of teaching content.

[0036] Through more targeted and intelligent teaching content generation, the present invention can better promote the dissemination of red culture in colleges and universities. Under the guidance of more accurate teaching content, students can more comprehensively understand the historical and practical significance of red culture, thereby enhancing their sense of identity with red culture and their awareness of consciously inheriting the red spirit. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings;

[0038] Figure 1 This is a schematic diagram of the principle of an intelligent teaching content generation method for the dissemination of red culture in college courses in the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Reference Figure 1 The following embodiments are obtained:

[0041] Embodiment 1: A method for generating intelligent teaching content for red culture dissemination in college courses, comprising the following steps:

[0042] The current teaching content of red culture and all students' questions are obtained respectively, and then one of the students is set as the target student. The questions of the target students are summarized to obtain a question set, and matched with the current teaching content of red culture to obtain a preliminary matching content set; basic data is obtained from the current teaching content and the questions raised by students. The teaching content usually includes historical events, figures, classic stories, etc. of red culture, and the questions asked by students can reflect their confusion and interests in the learning process. By collecting teaching content and student questions, it can ensure that the subsequent matching and analysis are based on the content actually learned by students, and provide a data basis for the subsequent customized generation of teaching content. Summarizing the questions asked by the target students can analyze the students' questions and confusions in the study of red culture in a quantitative way, and match them with the teaching content to ensure that each question has corresponding reference teaching content. Summarize the questions asked by the target students to form a question set, and match them with the current teaching content through natural language processing (NLP) and other technologies to generate a preliminary matching content set. This process is to ensure that each question has corresponding teaching content support and provide a basis for subsequent analysis and customized generation.

[0043] Based on the preliminary matching content set, a matching degree analysis is performed to identify whether there is a hidden danger of misinterpretation of the teaching content according to the matching degree analysis results; the matching degree analysis aims to evaluate the relevance of the questions asked and the teaching content, and identify whether there is a hidden danger of misinterpretation. Misinterpretation risks may cause students to misunderstand the content of red culture, so it is very important to discover and correct them in time. Through technologies such as semantic deviation detection, the degree of match between the questions asked and the teaching content is calculated. If there is a large semantic deviation between the question and the content, it may mean that the question misinterprets the teaching content. According to the results of the matching degree analysis, it can be judged whether there is a risk of misinterpretation of the teaching content.

[0044] If there is a hidden danger of misinterpretation of teaching content, the preliminary matching content set is subjected to misinterpretation review and measurement operations, and the misinterpretation review index and misinterpretation measurement index are generated respectively. Then, each question in the preliminary matching content set is divided into high-matching questions and low-matching questions based on the misinterpretation review index and misinterpretation measurement index. When there is a hidden danger of misinterpretation, it is necessary to deeply analyze the difference between the question and the teaching content, and generate the misinterpretation review index and misinterpretation measurement index. The misinterpretation review index quantifies the semantic deviation between the question and the content by calculating the similarity between the question and the teaching content, the keyword matching rate, and the negative sample similarity. The misinterpretation measurement index evaluates the deviation between the question and the content in terms of emotion and context through sentiment analysis and context relevance calculation. Through these two indexes, it is possible to further evaluate whether each question misinterprets the teaching content. The misinterpretation review index and misinterpretation measurement index are used to classify the question into high-matching questions and low-matching questions. High-matching questions indicate that the question is highly relevant to the teaching content, while low-matching questions may be less relevant to the teaching content or misinterpreted. The misinterpretation audit index and misinterpretation measurement index are analyzed using fuzzy logic, and the matching type of questions is obtained through fuzzy rule reasoning. High-matching questions are those that have been verified and are highly consistent with the teaching content, while low-matching questions may be irrelevant or misinterpreted. Low-matching questions need to be eliminated to ensure that the final teaching content generated is accurate and relevant.

[0045] Low-matching questions are eliminated, and all high-matching questions are included in the customized content generation strategy. If there is no risk of misinterpretation of the teaching content, each question in the preliminary matching content set is identified as a high-matching question and included in the customized content generation strategy; low-matching questions are eliminated to ensure that the customized teaching content generated subsequently can focus on questions that are highly relevant to the teaching content, avoiding unnecessary misinterpretation or misleading. After eliminating low-matching questions, all high-matching questions are included in the customized content generation strategy. The customized content generation strategy ensures that the teaching content generated subsequently can accurately answer the questions of the target students, and further in-depth analysis and supplementation are carried out based on the questions.

[0046] Based on the customized content generation strategy, the current red culture teaching content is regenerated with the target students as the object. The goal is to regenerate personalized teaching content according to the needs of the target students, so that the teaching is more targeted and effectively solves the students' confusion. According to the customized content generation strategy, teaching content suitable for the target students will be automatically generated. These contents not only include answers to the questions asked by students, but also in-depth explanations of the red culture content raised by them to ensure that students can fully understand the teaching content.

[0047] Matching degree analysis means: using a semantic deviation detection model, semantic deviation detection is performed on each judgment question in the preliminary matching content set to generate a semantic deviation value. The semantic deviation detection model refers to: S(Q i ) represents the i-th question Q i The norm of the semantic representation vector of the current red culture teaching content C, S(C) represents the norm of the semantic representation vector of the current red culture teaching content C, R(Q i ,C) represents the i-th question Q i The semantic deviation value between the current red culture teaching content C. The semantic similarity between each question and the teaching content is evaluated using the semantic deviation detection model. Based on natural language processing technology, the model can calculate the difference between the semantic representation of each question and the semantic representation of the current red culture teaching content, and generate a semantic deviation value. The natural language processing technology will not be described here. Semantic deviation detection is performed on each question in the preliminary matching content set in turn, and the difference between the semantic representation vector of the question and the semantic representation vector of the current red culture teaching content is calculated. The semantic deviation value indicates the degree of semantic deviation between the question and the teaching content. The larger the value, the lower the match between the question and the content. The mean and standard deviation of all semantic deviation values ​​are calculated. The mean reflects the overall semantic match between the entire question set and the teaching content. The standard deviation measures the degree of discreteness of the semantic match between different questions and the teaching content in the question set. A larger standard deviation means that the semantic deviation values ​​of different questions are more different.

[0048] Identifying whether there is a risk of misinterpretation of teaching content based on the matching analysis results means:

[0049] Compare the average value and the standard deviation with the preset standard value 1 and standard value 2 respectively. If it is satisfied that the average value is less than or equal to the preset standard value 1 and the standard deviation is less than or equal to the standard value 2, it is determined that there is no hidden danger of teaching content distortion. If it does not satisfy that the average value is less than or equal to the preset standard value 1 and the standard deviation is less than or equal to the standard value 2, it is determined that there is a hidden danger of teaching content distortion. The average value and the standard deviation are statistical indicators for evaluating the matching degree between all the questioned questions and the teaching content. Through these two indicators, it is possible to evaluate as a whole whether the students' questions are highly consistent with the teaching content. The average value reflects the overall matching degree between the set of questioned questions and the teaching content. If the average semantic deviation value is low, it indicates that most of the questions have a high semantic matching degree with the teaching content, and the question content is relatively consistent with the teaching content. If the average value is high, it indicates that there is a risk of overall distortion. The standard deviation reflects the degree of dispersion of the matching degree between different questioned questions. A smaller standard deviation means that the semantic matching degrees of most questions are similar and the deviations are consistent, indicating that the understanding consistency between the questions and the teaching content is good. On the contrary, a larger standard deviation means that the semantic deviation values of some questions deviate significantly from the average value, and there may be significant distortion.

[0050] By comparing the calculated average value and the standard deviation with the preset standard values, it is judged whether the set of students' questioned questions is within a reasonable range. If the conditions are met, it is considered that there is no problem with the understanding of the questioned questions; if not, it is considered that there is a hidden danger of distortion. The standard value 1 is used to limit the average level of the semantic deviation value to ensure a relatively high overall semantic matching degree between the questioned questions and the teaching content. If the average value exceeds the standard value 1, it indicates that the overall matching degree of the questioned questions is insufficient and there may be distortion. The standard value 2 is used to limit the degree of dispersion of the semantic deviation value to ensure that the matching degrees of different questions do not fluctuate much. If the standard deviation exceeds the standard value 2, it indicates that the matching degrees between the questions vary greatly, which may mean that some questions have significant distortion.

[0051] Based on the comparison results of the average value and the standard deviation, it is possible to automatically judge whether there is a hidden danger of teaching content distortion in the set of questioned questions. This step can be used as a quality control measure during the content generation process to ensure the accuracy of the teaching content and the consistency of students' understanding. If the conditions are met: when the average value of the set of questioned questions is less than or equal to the standard value 1 and the standard deviation is less than or equal to the standard value 2, it is considered that the students' questioned questions have a relatively high matching degree with the teaching content and the understanding is relatively accurate, and no further intervention or review is required. It will enter the subsequent content generation stage. If the conditions are not met: if any one of the average value or the standard deviation exceeds the preset standard, it is determined that there is a hidden danger of distortion. This means that the semantic deviation values of some questions are relatively large or the overall semantic deviation is relatively high, and further distortion review and measurement are required to ensure the accuracy of the finally generated teaching content.

[0052] Through such quantitative analysis, we can ensure the accurate delivery of teaching content and prevent students from misunderstanding or misinterpreting the red culture during the learning process. This is of great significance to improving the quality of teaching and the accuracy of customized teaching content. If misinterpretation risk detection is not performed, the wrong information may be further amplified in the subsequent customized content generation process. Through this step, misinterpretation risks can be discovered and handled in time before the problem content is generated, ensuring the accuracy and effectiveness of the generated teaching content.

[0053] The logic for obtaining the misinterpreted audit index is:

[0054] Match each question Qi with the teaching content C respectively, and introduce irrelevant teaching content fragments as negative samples, namely negative sample content fragments C neg , and then use a two-layer nested semantic vector model to calculate the similarity. The first layer uses a context-based word embedding model, and the second layer uses a sentence embedding model. The semantic vector model is:

[0055] Indicates the i-th question Q i The k-th word embedding vector, E(C j ) represents the jth word embedding vector of all words in the current red culture teaching content C, E(Cneg j ) represents the negative sample content segment C neg The jth word embedding vector of all words in Q i and C is m and n, respectively. i )‖ represents the vector E(Q i ), ‖E(C)‖ represents the norm of the vector E(C), and the negative sample content fragment C neg The number of words in the content is the same as the number of words in the teaching content C. Sdev(Q i ,C) represents the i-th question Q i The similarity deviation with the current red culture teaching content C; by calculating the similarity between each question word and all words in the teaching content and taking the maximum similarity value, the present invention can ensure that each question word finds the most relevant part of the teaching content. This method ensures that even if some words in the question have a strong semantic match with some words in the teaching content, the present invention can correctly identify them, thereby avoiding unnecessary misinterpretation.

[0056] Each question is matched with the teaching content C and the semantic similarity between them is calculated. First, a two-layer nested semantic vector model is used. The first layer uses a context-based word embedding model, and the second layer uses a sentence embedding model to ensure that the similarity calculation can capture detailed semantic relationships. In order to further measure whether the question misinterprets the teaching content, irrelevant teaching content fragments are introduced as negative samples. The purpose of introducing negative samples is to enhance the judgment of whether the question deviates from the teaching content by calculating the similarity between the question and the irrelevant content.

[0057] Calculate the keyword matching rate deviation between the question and the teaching content. The formula is:

[0058] kj represents the jth keyword, wj represents the TF-IDF weight preset for the jth keyword kj in the teaching content C, kj∈Q i Indicates whether the keyword kj appears in the question. If it exists, the value is 1, otherwise it is 0. u is the total number of keywords. Kdev(Q i ,C) indicates asking question Q i The keyword matching rate deviation with the current red culture teaching content C; first calculate the matching rate of the keywords in the question and the keywords in the teaching content, and then ensure that the matching degree of the core vocabulary can be effectively measured by weighting the word frequency. The keyword matching rate deviation is used to measure the matching degree between the key terms in the question and the key terms in the teaching content. It is based on the TF-IDF weight (word frequency-inverse document frequency) of each keyword to ensure that the high-importance words in the teaching content are accurately reflected in the question. If these keywords fail to appear in the question, it means that the question is inconsistent with the core information of the teaching content and there is a possibility of misinterpretation. This parameter captures the matching degree between the question and the important information in the teaching content. Keywords are often the core part of the teaching content, especially in the field of red culture, the correct use of certain keywords (such as historical events, characters, etc.) is crucial. Keyword matching is one of the important means to evaluate semantic understanding, because if the question lacks core keywords, it means that the students' understanding of the content may be incomplete or wrong. Therefore, it is included in the calculation of the misinterpretation review index to make the index more representative.

[0059] The calculation formula of the misinterpretation audit index is:

[0060] QS=log e (α1*Sdev(Q i ,C)+α2*Kdev(Q i,C)+1); α1 and α2 are both preset audit coefficients, which are used to adjust the relative weights of semantic deviation and keyword matching rate. Semantic similarity deviation captures deeper semantic differences, while keyword matching rate deviation focuses on the matching of specific words. The combination of the two can comprehensively evaluate whether there is any misinterpretation of the question from both the semantic level and the surface vocabulary usage. The use of the logarithmic function can compress the numerical range and avoid the imbalance of the misinterpretation audit index caused by excessive semantic deviation and keyword deviation. It can also increase the stability of the formula, especially making the numerical changes smoother under large differences. The logarithmic function smoothes the large differences so that the index is within a reasonable range, avoiding some deviation values ​​that are too large and affect the rationality of the overall index. At the same time, through the logarithmic function, the formula can handle input values ​​of different ranges more stably. The constant 1 is added to avoid unreasonable results when both semantic deviation and keyword deviation are 0, such as the inability to calculate the logarithm. This ensures that when the question is highly matched with the teaching content, the formula can still return a reasonable value. QS stands for the distortion review index. The index reflects the overall match between the questions and the teaching content, as well as the possibility of distortion. The distortion review index is generated to determine whether there is a semantic deviation between the questions and the red culture teaching content, especially the potential risk of causing students to misunderstand the teaching content. By calculating the similarity between the questions and the teaching content, the keyword matching rate, and the comparison with negative samples (i.e., irrelevant content), the present invention can automatically identify and quantify the degree of distortion of the questions, thereby ensuring that the teaching content finally generated has no obvious distortion. Moreover, the distortion review index helps to screen out questions that may lead to misunderstandings, and correct them before generating customized content, thereby ensuring the accuracy and consistency of teaching quality.

[0061] The larger the distortion review index, the greater the semantic deviation between the question and the current red culture teaching content, that is, the higher the possibility of distortion. If the distortion review index is large, it means that the semantic similarity between the question and the teaching content is low, and there is a large semantic deviation between the two. The question may deviate from the core idea or key information of the teaching content. A large distortion review index also indicates that the degree of match between the keywords in the question and the keywords in the teaching content is low. The question may not accurately capture the key information in the teaching content, reflecting that there may be errors in the students' understanding of the content. The increase in the distortion review index means that the risk of the question distorting the teaching content is higher. The present invention needs to further evaluate and deal with these problems to avoid incorporating these distortion problems into the subsequent teaching content generation process, which causes students to misunderstand the content.

[0062] The logic of obtaining the distorted measurement index is:

[0063] Get the questions Q respectively iThe positive, negative and neutral emotional scores of the current red culture teaching content C are calculated, and the same type of emotional scores are processed by absolute value after difference calculation. Then the absolute value processing results of the three types of emotional scores are averaged to obtain the emotional tendency deviation value Tdev (Q i ,C); By calculating the difference in the positive, negative, and neutral sentiment scores between the questions and the teaching content, we can measure whether there is a deviation between the questions and the teaching content in terms of sentiment orientation. The pre-trained sentiment analysis model obtains these sentiment scores by analyzing the tone and expression of the text, and the formula can quantify whether the questions are emotionally consistent with the teaching content by calculating the difference in sentiment scores. By processing the absolute values ​​of the differences in the three sentiment scores, the formula ensures that both positive and negative sentiment deviations can be accurately captured. The average of the absolute value processing results of the three types of sentiment scores further smooths the differences in different sentiment types, making the overall results more robust and quantifying the consistency of the sentiment orientation between the questions and the teaching content. The greater the sentiment deviation, the greater the difference in sentiment orientation between the questions and the teaching content, which may lead to students' emotional misunderstanding of the content. Therefore, by calculating the sentiment deviation, we can detect whether the questions may distort the sentiment orientation of the teaching content.

[0064] Get Question Q i The first context window C of the current red culture teaching content C l The correlation Rel(Q i ,C), and then perform the following calculation:

[0065] dl represents the context window C l Ask questions i The distance in the teaching content, wl represents the context window C l The influence coefficient in the teaching content, λ is the attenuation coefficient, which is used to control the distance to the correlation Rel(Q i ,C), p is the context window C l The total number of Cdev(Q i , C) represents the context association value; each context window represents a semantic unit of the teaching content, and the association between the question and each context window is calculated by Rel(Q i,C), it can be judged whether the question is consistent with the local semantics of the teaching content. The distance decay function exp(-λdl) is introduced to take into account the distance between the question and the teaching content. The influence of the context window with a longer distance on the relevance will gradually weaken. This design makes the formula more effective in long texts. The influence coefficient wl reflects the importance of each context window and can be weighted according to the length of the window or the information density. In order to measure the consistency of the local semantics between the question and the teaching content, by taking a weighted average of the relevance of each context window, it can be judged whether the question is consistent with the teaching content in terms of local semantics. The large deviation of the context relevance indicates that the question does not match the teaching content in the local context, and there may be a semantic distortion.

[0066] The calculation formula of the distortion index is:

[0067] QH=β1*Tdev(Q i ,C)+β2*(1-Cdev(Q i ,C)); β1 and β2 are both preset measurement coefficients, and QH represents the misinterpretation measurement index. The emotional deviation value and the context relevance value are combined to comprehensively measure whether there is a risk of misinterpretation in the question. The emotional deviation is used to detect whether the question deviates from the teaching content in terms of emotion, while the context relevance deviation is used to detect the consistency of the local semantics of the question. By subtracting the context relevance deviation 1-Cdev(Q i ,C) is introduced, the formula can ensure that the lower the context relevance, the higher the distortion index. β1 and β2 are used to adjust the relative importance of emotional bias and context relevance bias. The specific application scenario can adjust these two coefficients as needed to achieve optimized processing of different types of teaching content. The distortion index is a comprehensive evaluation result of whether there is distortion in the question. By combining emotional bias and context relevance bias, the index can comprehensively evaluate the degree of match between the question and the teaching content. The larger the distortion index, the greater the difference between the question and the teaching content, and the greater the possibility of distortion, which should be further processed by the present invention.

[0068] It should be noted that: dl represents the context window C l Ask questions i The distance in the teaching content can be obtained in the following ways:

[0069] 1. Distance based on word position: This distance is calculated based on the position of the word in the question and the position of the context window in the teaching content. The unit of distance is usually the number of words, that is, the distance between the word in the question and the starting word or a keyword in a context window in the teaching content. For example:

[0070] 2. Distance based on paragraph position: If the teaching content is divided into multiple paragraphs or chapters, paragraph numbers can be used to indicate the distance between questions and teaching content.

[0071] 3. Distance based on time (applicable to dynamic content): If the teaching content is based on video, audio or lecture, time can be used as a measure of distance. The distance can indicate the gap between the content involved in the question and the teaching content on the timeline.

[0083] The time distance reflects the time sequence between the question and the discussion of the teaching content. The greater the distance, the weaker the correlation between the question and the teaching content.

[0084] Distance is the relative position between the question and the context window in the teaching content, which can be measured based on vocabulary position, paragraph position or time axis. This distance is used to quantify the degree of relevance between the question and the teaching content. The larger the distance, the farther the question and the content are physically or temporally separated, and the semantic relevance may be weaker. By introducing the distance decay function, the formula ensures that as the distance increases, the impact of the question on the relevance gradually weakens, thereby reasonably reflecting the impact of different context windows on the semantic relevance of the question.

[0085] Dividing each question in the preliminary matching content set into high-matching questions and low-matching questions based on the distortion review index and the distortion measurement index means using fuzzy logic, treating each question distortion review index and the distortion measurement index as a set of input variables, and the matching type of the question and the content as the output variable, fuzzifying the input variables, converting the values ​​of the input variables into fuzzy sets, fuzzifying the output variables, converting the output variables into fuzzy sets, formulating fuzzy rules, describing the matching suitability of questions and contents under different data type combinations, and reasoning the fuzzified input variables through fuzzy rules to obtain the matching type of the question and the content.

[0086] The misinterpretation audit index and the misinterpretation measurement index are two input variables. Each question has these two indexes, which reflect the matching situation between the question and the teaching content and the possibility of misinterpretation. The numerical range of the misinterpretation audit index and the misinterpretation measurement index is divided into different fuzzy sets, such as high, medium and low levels. This can handle the matching situation of the question more flexibly, rather than just strict numerical judgment. For example, the value of the misinterpretation audit index can be fuzzified into "low audit index", "medium audit index" and "high audit index". The value of the misinterpretation measurement index can be fuzzified into "low measurement index", "medium measurement index" and "high measurement index". The matching type between the question and the content is the output variable and also needs to be fuzzified. The matching type can be divided into different fuzzy sets, such as "high matching degree", "medium matching degree" and "low matching degree". The purpose of fuzzification of output variables is to be able to flexibly define different matching levels according to different combinations of input variables, so as to distinguish which questions match the teaching content well and which questions may be misinterpreted. For example, the fuzzy set of the output variable can be "high matching", "medium matching" and "low matching".

[0087] The fuzzy rule is the core of the logic processing. It describes how to judge the matching degree between the question and the content under different combinations of the distortion review index and the distortion measurement index. For example, when the distortion review index is high but the distortion measurement index is low, the present invention can judge the question as a "medium match".

[0088] Fuzzy rules are formulated based on expert experience or actual data, which allows the present invention to flexibly judge the matching type according to different combinations of input values. For example: Rule 1: If the distortion audit index is "high audit index" and the distortion measurement index is "high measurement index", the matching type is "low matching degree". Rule 2: If the distortion audit index is "low audit index" and the distortion measurement index is "medium measurement index", the matching type is "medium matching degree". Rule 3: If the distortion audit index is "low audit index" and the distortion measurement index is "low measurement index", the matching type is "high matching degree".

[0089] Fuzzy reasoning is to infer the matching type of the question according to the pre-established fuzzy rules, combined with the input distortion audit index and distortion measurement index. This process will combine the fuzzified input variables through fuzzy rules, and finally get the matching type result. For example: if the distortion audit index of a question is medium and the distortion measurement index is low, then according to the rules, the present invention may classify it as "medium matching". After inferring the matching type of the question, the fuzzy output set needs to be converted into an actual numerical result, that is, defuzzification is performed. The purpose of this step is to convert the result of fuzzy reasoning (such as "high matching degree" or "low matching degree") into a specific judgment, which helps the present invention to divide the question into a high matching degree question or a low matching degree question. Defuzzification can adopt different methods, such as taking the maximum membership method, and taking the fuzzy set with the highest membership as the final output. According to the fuzzy reasoning results, the present invention will divide each question in the preliminary matching content set into a high matching degree question or a low matching degree question. High-match questions are those that, after review, are well matched with the teaching content, while low-match questions are those that have the potential for misinterpretation or are poorly matched with the teaching content.

[0090] Based on the customized content generation strategy, the current red culture teaching content is regenerated with the target students as the object. The implementation method can be as follows: based on the customized content generation strategy, with the target students as the object, the various questions in the customized content generation strategy are sorted according to their performance in fuzzy logic, and then a percentage value is assigned according to the sequence number. Then, in the final regenerated red culture teaching content, when the high matching degree problem in the original red culture teaching content is explained in more detail, the percentage value is used to determine the amount of data explained for each problem in the regenerated teaching content. The problem with the front sequence number will be assigned a higher percentage, so as to obtain a more detailed explanation. Percentage allocation method: The percentage allocation can be gradient decreased according to the order of the problems. For example, the problem with the first sequence number can be assigned 30% of the explanation amount, the problem with the second sequence number can be assigned 25%, and so on, until the problem with the last sequence number is assigned a relatively small amount of explanation. For example, for a certain problem, the present invention may explain in detail the specific historical events, details or add supplementary materials in the red culture; while for another problem, the present invention may give a brief explanation.

[0091] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0092] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0093] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0094] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0095] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for generating intelligent teaching content for red culture dissemination in college courses, characterized in that: The following steps are involved: The current teaching content of red culture and the questions asked by all students are obtained respectively, and then one of the students is set as the target student, the questions asked by the target student are summarized to obtain a question set, and the questions are matched with the current teaching content of red culture to obtain a preliminary matching content set; Based on the preliminary matching content set, a matching degree analysis is conducted, and the matching degree analysis results are used to identify whether there is a hidden danger of misinterpretation of the teaching content; If there is a hidden danger of misinterpretation of the teaching content, the preliminary matching content set is subjected to misinterpretation review and measurement operations, and a misinterpretation review index and a misinterpretation measurement index are generated respectively. Then, each question in the preliminary matching content set is divided into a high-matching question and a low-matching question based on the misinterpretation review index and the misinterpretation measurement index. Eliminate low-matching questions and include all high-matching questions in the customized content generation strategy. If there is no risk of misinterpretation of teaching content, identify each question in the preliminary matching content set as a high-matching question and include them in the customized content generation strategy. Based on customized content generation strategies, the current teaching content of red culture is regenerated with target students as the object.

2. According to claim 1, the intelligent teaching content generation method for the red culture dissemination of college courses is characterized by: Performing a matching analysis means: Through the semantic deviation detection model, semantic deviation detection is performed on each judgment question in the preliminary matching content set to generate a semantic deviation value, and then the average and standard deviation of all semantic deviation values ​​are calculated.

3. According to claim 2, the intelligent teaching content generation method for the red culture dissemination of college courses is characterized by: Identifying whether there is a risk of misinterpretation of teaching content based on the matching analysis results means: Compare the mean value and standard deviation with the preset standard value one and standard value two respectively. If the mean value meets the conditions that the mean value is less than or equal to the preset standard value one and the standard deviation meets the conditions that the mean value is less than or equal to the preset standard value one and the standard deviation meets the conditions that the mean value is less than or equal to the preset standard value one and the standard deviation meets the conditions that the teaching content is misinterpreted. If the mean value meets the conditions that the mean value is less than or equal to the preset standard value one and the standard deviation meets the conditions that the teaching content is misinterpreted.

4. According to claim 3, the intelligent teaching content generation method for the red culture dissemination of college courses is characterized by: The semantic deviation detection model refers to: S(Q i ) represents the i-th question Q i The norm of the semantic representation vector of the current red culture teaching content C, S(C) represents the norm of the semantic representation vector of the current red culture teaching content C, R(Q i ,C) represents the i-th question Q i The semantic deviation value between the current teaching content C of red culture.

5. According to claim 4, the intelligent teaching content generation method for the red culture dissemination of college courses is characterized by: The logic for obtaining the misinterpreted audit index is: Match each question Qi with the teaching content C respectively, and introduce irrelevant teaching content fragments as negative samples, namely negative sample content fragments C neg , and then use a two-layer nested semantic vector model to calculate the similarity. The first layer uses a context-based word embedding model, and the second layer uses a sentence embedding model. The semantic vector model is: The i-th question Q i The k-th word embedding vector, E(C j ) represents the jth word embedding vector of all words in the current red culture teaching content C, E(Cneg j ) represents the negative sample content segment C neg The jth word embedding vector of all words in the question Qi and the teaching content C are respectively the number of words in the question Qi and the teaching content C, and the negative sample content segment C is neg The number of words in the content is the same as the number of words in the teaching content C. Sdev(Q i ,C) represents the i-th question Q i The similarity deviation with the current red culture teaching content C; Calculate the keyword matching rate deviation between the question and the teaching content. The formula is: kj represents the jth keyword, wj represents the TF-IDF weight preset for the jth keyword kj in the teaching content C, kj∈Q i Indicates whether the keyword kj appears in the question. If it exists, the value is 1, otherwise it is 0. u is the total number of keywords. Kdev(Q i ,C) indicates asking question Q i The keyword matching rate deviation with the current red culture teaching content C; The calculation formula of the misinterpretation audit index is: QS=log e (α1*Sdev(Q i ,C)+α2*Kdev(Q i ,C)+1); α1 and α2 are both preset audit coefficients, and QS represents the distortion audit index.

6. According to claim 5, the intelligent teaching content generation method for the red culture dissemination of college courses is characterized by: The logic of obtaining the distorted measurement index is: Get the questions Q respectively i The positive, negative and neutral emotional scores of the current red culture teaching content C are calculated, and the same type of emotional scores are processed by absolute value after difference calculation. Then the absolute value processing results of the three types of emotional scores are averaged to obtain the emotional tendency deviation value Tdev (Q i ,C); Get Question Q i The first context window C of the current red culture teaching content C l The correlation Rel(Q i ,C), and then perform the following calculation: dl represents the context window C l Ask questions i The distance in the teaching content, wl represents the context window C l The influence coefficient in the teaching content, λ is the attenuation coefficient, which is used to control the distance to the correlation Rel(Q i ,C), p is the context window C l The total number of Cdev(Q i ,C) represents the contextual value; The calculation formula of the distortion index is: QH=β1*Tdev(Q i ,C)+β2*(1-Cdev(Q i ,C)); β1 and β2 are both preset measurement coefficients, and QH represents the distortion measurement index.

7. According to claim 6, the intelligent teaching content generation method for the red culture dissemination of college courses is characterized by: Dividing each question in the preliminary matching content set into high-matching questions and low-matching questions based on the distortion review index and the distortion measurement index means using fuzzy logic, treating each question distortion review index and the distortion measurement index as a set of input variables, and the matching type of the question and the content as the output variable, fuzzifying the input variables, converting the values ​​of the input variables into fuzzy sets, fuzzifying the output variables, converting the output variables into fuzzy sets, formulating fuzzy rules, describing the matching suitability of questions and contents under different data type combinations, and reasoning the fuzzified input variables through fuzzy rules to obtain the matching type of the question and the content.

Images