Artificial intelligence-based ideological and political teaching customized pushing method

Abstract

The invention discloses an artificial intelligence-based ideological and political teaching customized pushing method, and belongs to the technical field of teaching pushing. The method comprises the following steps of: acquiring behavior and interest data of students, constructing a student-content scoring matrix, calculating the similarity between the students by utilizing a Pearson's correlation coefficient, dividing the students into groups with similar interests and groups with dissimilar interests, and analyzing complexity characteristics and thought depth characteristics of pushed contents for the students with similar interests. Evaluating the understanding accuracy of the recommendation algorithm, dividing the push content into accurate push, incomplete accurate push and inaccurate push, and carrying out classification processing; for students who are not completely and accurately pushed, by predicting understanding accuracy changes of the pushed content in a fixed time period and dynamically adjusting the pushing frequency and the content, the thought depth and personalized matching of the pushed content are effectively improved, it is ensured that the content better serves the ideological and political teaching target, and simplification or deviation of recommendation is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of teaching push, and particularly relates to a customized push method for ideological and political teaching based on artificial intelligence. Background Art

[0002] With the rapid development of information technology and artificial intelligence technology, intelligent push systems are increasingly widely used in the field of education. Especially in ideological and political teaching, customized push according to the personalized characteristics of different students, such as interests, learning progress, knowledge mastery, etc., has become one of the keys to improving teaching effects. Existing push technologies construct a preliminary user portrait based on students' basic information, historical learning data, etc., and use simple content matching or collaborative filtering-based recommendation algorithms to select teaching content that may be relevant to students' interests or needs for push. Finally, the system will make adaptive adjustments according to students' feedback data.

[0003] The existing technologies have the following deficiencies:

[0004] Most existing ideological and political teaching push systems recommend content based on simple interest matching and behavior analysis, usually using collaborative filtering algorithms or push strategies based on users' historical data. However, when faced with complex ideological and political teaching content, these systems often cannot effectively identify the ideological depth and complexity of the pushed content, resulting in a deviation between the pushed content and the teaching objectives. In addition, due to the limitations of personalized recommendation, the system has different accuracies in understanding different students and cannot dynamically adjust the push frequency and content, easily causing the simplification of content recommendation or deviation from the teaching objectives, affecting the educational effect. Summary of the Invention

[0005] The purpose of the present invention is to provide a customized push method for ideological and political teaching based on artificial intelligence to solve the deficiencies in the background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A customized push method for ideological and political teaching based on artificial intelligence, comprising the following steps:

[0007] S1: Collect students' behavior data and interest data, and construct a student-content scoring matrix, where each student in the matrix corresponds to a row and the teaching content corresponds to a column;

[0008] S2: Based on students' behavior data, calculate the similarity between all students through the Pearson correlation coefficient, compare the similarity with a preset similarity threshold, and divide the students into students with similar interests and students with dissimilar interests according to the comparison results;

[0009] S3: Analyze the ideological and political teaching content pushed to students with similar interests. Determine the understanding accuracy of the recommended algorithm for the content pushed to students with similar interests according to the complexity characteristics and ideological depth characteristics of the pushed content.

[0010] S4: Divide the understanding accuracy of the recommended algorithm for the content pushed to students with similar interests into different levels, divide the recommended content into accurate push, incomplete accurate push, and inaccurate push, and perform corresponding processing.

[0011] S5: For students with similar interests with incomplete accurate push, predict the abnormal degree of the understanding accuracy of the recommended algorithm for the pushed content for students with similar interests within a fixed time period, and dynamically adjust the push frequency and push content according to the prediction results.

[0012] Preferably, in S2, mark the rating matrix as R, where: R[i,j] represents the rating of student i for content j, i is the index of the student, j is the index of the content, and the Pearson correlation coefficient formula is used to calculate the interest similarity between two students. The expression is: In the formula, Pearson(A,B) is the Pearson correlation coefficient between student A and student B, R[A,j] represents the rating of student A for content j, R[B,j] represents the rating of student B for content j, and I is the set of contents jointly rated by student A and student B. is the average rating of student A for all contents, and the expression is: is the average rating of student B for all contents, and the expression is:

[0013] Set a similarity threshold T. If Pearson(A,B)≥T, then student A and student B are considered students with similar interests; if Pearson(A,B)<T, then student A and student B are considered students with dissimilar interests.

[0014] Preferably, generate an information density complexity index according to the complexity characteristics of the pushed content. The acquisition method of the information density complexity index is:

[0015] In a graph-based model, each vocabulary in the text is represented as a node, the co-occurrence relationship between words is represented as an edge, and each vocabulary w i serves as a node V i of a graph. If there are h vocabularies after text segmentation, then there are {V1, V2, …, V h}; The edge E in the graph represents the co-occurrence relationship between vocabularies, and the weight w i,j of the edge represents the co-occurrence relationship between two vocabularies w i and w jDefine a window of fixed size for each word to represent the association strength between them, with the expression: Finally, a weighted graph G=(V, E) is formed, where V is the set of nodes, E is the set of edges, and each edge has a weight w i,j ; Use the TextRank algorithm to iteratively calculate the scores of each node, with the expression: In the formula, S(V i ) is the importance score of node V i , representing the contribution degree of the word w i in the entire text. d is the damping coefficient, In(V i ) is the set of nodes pointing to node V i , that is, all the words connected to V i . represents the sum of the weights of all the edges emitted from node V j , representing the out-degree of node V j . Continuously update the importance scores of each node in an iterative manner until the change values of the scores of all nodes are less than the preset threshold, and then stop the iteration. After calculating the importance scores of all nodes, use the scores to calculate the information density complexity index of the entire text, with the expression: In the formula, KM is the information density complexity index.

[0016] Preferably, generate the historical-reality association index according to the ideological depth feature, and the method for obtaining the historical-reality association index is:

[0017] Split the text content into sentences, and extract the sentences related to historical events and real-world problems respectively; Use the BERT model to convert each sentence into a vector representation to capture the semantic information of each sentence. Load the pre-trained BERT model for embedding the sentences into the vector space;

[0018] Input each historical event sentence and real-world problem sentence into the BERT model, and extract the embedding vectors of each sentence. The BERT outputs the embedding vectors of each word, and the formula is expressed as: V 句子 = BERT(S 句子 ); where: V 句子 is the vector representation of the sentence, S 句子 is the input sentence text, and BERT represents vectorizing the sentence through the BERT model; Measure the semantic similarity between the historical event sentence and the real-world problem sentence through cosine similarity, and the calculation expression is: Cosine similarity V 历史 is the embedding vector of the historical event sentence, V 现实is the embedding vector of the real - world problem sentence, and V represents the Euclidean norm of the vector; each historical event sentence is compared pairwise with each real - world problem sentence to calculate their cosine similarity. Based on the similarities between all historical event sentences and real - world problem sentences, the historical - reality correlation index is calculated, and the expression is: Cosine similarity(V 历史,i ,V 现实,j ); where p is the number of historical event sentences, q is the number of real - world problem sentences, and GF is the historical - reality correlation index.

[0019] Preferably, convert the information density complexity index and the historical - reality correlation index into a first - feature vector, and use the first - feature vector as the input of the machine - learning model. The machine - learning model takes the prediction of the understanding accuracy value label of the content pushed to students with similar interests by the recommendation algorithm as the prediction target, and takes minimizing the sum of the prediction errors of the understanding accuracy value labels of the content pushed to students with similar interests by all recommendation algorithms as the training target. Train the machine - learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the understanding accuracy value of the content pushed to students with similar interests by the recommendation algorithm according to the model output result. Among them, the machine - learning model is a polynomial regression model.

[0020] Preferably, in S4, divide the understanding accuracy of the content pushed by the recommendation algorithm to students with similar interests into different levels, and divide the recommended content into accurate push, incompletely accurate push, and inaccurate push. Specifically:

[0021] Compare the obtained understanding accuracy value of the content pushed by the recommendation algorithm to students with similar interests with the gradient standard thresholds. The gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the understanding accuracy value of the content pushed by the recommendation algorithm to students with similar interests with the first standard threshold and the second standard threshold respectively;

[0022] If the understanding accuracy value of the content pushed by the recommendation algorithm to students with similar interests is greater than the second standard threshold, it indicates that the understanding accuracy of the content pushed by the recommendation algorithm to students with similar interests is high. At this time, generate a high - accuracy signal and divide the recommended content into accurate push;

[0023] If the understanding accuracy value of the content pushed by the recommendation algorithm to students with similar interests is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the understanding accuracy of the content pushed by the recommendation algorithm to students with similar interests is average. At this time, generate a medium - accuracy signal and divide the recommended content into incompletely accurate push;

[0024] If the understanding accuracy value of the content pushed to students with similar interests by the recommendation algorithm is less than the first standard threshold, it indicates that the understanding accuracy of the content pushed to students with similar interests by the recommendation algorithm is low. At this time, a low-accuracy signal is generated, and the recommended content is classified as inaccurate push.

[0025] Preferably, in S4, for students with similar interests of the incompletely accurate push, the abnormality degree of the understanding accuracy of the recommendation algorithm for the pushed content for students with similar interests within a fixed time period is predicted. Specifically:

[0026] For students with similar interests of the incompletely accurate push, that is, the understanding accuracy value of the recommendation algorithm for the content pushed to students with similar interests generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold. The understanding accuracy values of the recommendation algorithm for the content pushed to students with similar interests that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold generated in the subsequent fixed time period are collected, a corresponding data set is established, and the mean and standard deviation of the data set are calculated.

[0027] Preferably, if the mean value of the understanding accuracy values of the pushed content in the data set is greater than or equal to the reference threshold of the mean value of the understanding accuracy values of the pushed content, and the standard deviation of the understanding accuracy values of the pushed content is less than the reference threshold of the standard deviation of the understanding accuracy values of the pushed content, the accuracy of the pushed content is relatively high and the fluctuation is small, indicating that the performance of the recommendation algorithm is stable and the quality of the pushed content is high. At this time, no warning signal is generated, and the content is pushed according to the current push frequency;

[0028] If the mean value of the understanding accuracy values of the pushed content is greater than or equal to the reference threshold of the mean value of the understanding accuracy values of the pushed content, and the standard deviation of the understanding accuracy values of the pushed content is greater than or equal to the reference threshold of the standard deviation of the understanding accuracy values of the pushed content, although the understanding accuracy of the pushed content is high but the fluctuation is large, the performance of the recommended content is unstable. At this time, a third-level warning signal is generated, the selection of the pushed content is adjusted, the fluctuation of the content quality is reduced, and the overall push frequency is increased;

[0029] If the mean value of the understanding accuracy values of the pushed content is less than the reference threshold of the mean value of the understanding accuracy values of the pushed content, and the standard deviation of the understanding accuracy values of the pushed content is greater than or equal to the reference threshold of the standard deviation of the understanding accuracy values of the pushed content, the accuracy of the pushed content is low and the fluctuation is large, indicating that the recommendation algorithm's understanding of students is inaccurate, and the quality of the pushed content is low and unstable. At this time, a first-level warning signal is generated, a large-scale adjustment is carried out, and the pushed content and frequency are re-optimized to improve the recommendation accuracy;

[0030] If the average value of the push content understanding accuracy is less than the reference threshold of the average value of the push content understanding accuracy, and the standard deviation of the push content understanding accuracy is less than the reference threshold of the standard deviation of the push content understanding accuracy, the accuracy of the push content is low but the fluctuation is small. At this time, a secondary warning signal is generated and the content is adjusted to improve the overall quality of the content.

[0031] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0032] 1. By comprehensively using the information density complexity index and the historical and realistic correlation index to optimize the personalized push strategy of ideological and political teaching, the present invention effectively solves the problem that the push content in the prior art cannot match the teaching objectives. By calculating the interest similarity based on the scoring matrix of student behavior data and conducting push analysis according to the complexity and ideological depth of the content, the system can accurately evaluate the understanding accuracy of the recommendation algorithm for the push content, classify the push content into different levels according to the accuracy, and ensure that the push content not only meets the students' interests but also can achieve the ideological and political teaching objectives by dynamically adjusting the push frequency and content.

[0033] 2. By introducing a dynamic adjustment mechanism based on warning signals, the present invention continuously optimizes the incompletely accurate push content by analyzing the mean and standard deviation of the understanding accuracy value. When the push content shows stable performance and high quality, the system maintains the current push frequency; if the fluctuation is large or the accuracy is low, the system will make corresponding adjustments according to different levels of warning signals, thereby improving the stability and accuracy of the recommendation system. This mechanism not only improves the effect of personalized recommendation but also enhances the ideological guidance role of the push content and optimizes the overall effect of ideological and political teaching. Description of the Drawings

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0035] Figure 1 It is the method flow chart of the present invention. Detailed Embodiments

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0037] Example, please refer to Figure 1 As shown, a customized push method for ideological and political teaching based on artificial intelligence in this example includes the following steps:

[0038] S1: Collect the behavior data and interest data of students, and construct a student-content rating matrix, where each student in the matrix corresponds to a row and the teaching content corresponds to a column;

[0039] S2: Based on the behavior data of students, calculate the similarity between all students through the Pearson correlation coefficient, compare the similarity with a preset similarity threshold, and divide the students into students with similar interests and students with dissimilar interests according to the comparison results;

[0040] S3: Analyze the ideological and political teaching content pushed to students with similar interests, and determine the understanding accuracy of the recommended algorithm for the content pushed to students with similar interests according to the complexity characteristics and ideological depth characteristics of the pushed content;

[0041] S4: Divide the understanding accuracy of the recommended algorithm for the content pushed to students with similar interests into different levels, divide the recommended content into accurate push, incompletely accurate push, and inaccurate push, and perform corresponding processing;

[0042] S5: For students with similar interests with incompletely accurate push, predict the abnormal degree of the understanding accuracy of the recommended algorithm for the content pushed to students with similar interests within a fixed time period, and dynamically adjust the push frequency and push content according to the prediction results.

[0043] Among them, in S1, collect the behavior data and interest data of students, and construct a student-content rating matrix, where each student in the matrix corresponds to a row and the teaching content corresponds to a column.

[0044] The collection of behavior data and interest data includes: The behavior data is the specific data directly collected by the system from the learning activities of students, usually including: Click behavior: The number of times a student clicks on a certain content on the learning platform. Learning duration: The duration of a student watching a video or reading an article and other content, which reflects the degree of investment in this content. Learning completion status: Whether a student has completed a certain learning task completely, such as watching an entire video or completing a test. Interaction behavior: Such as a student liking, commenting on, sharing, etc. on the content.

[0045] Interest data: Obtained through the historical preferences, personalized settings, or questionnaires of students. These data can include the interest topics actively marked by students, preferred learning forms (such as videos, articles, discussions, etc.), and their interest tendencies in specific aspects (such as history, deeds).

[0046] Matrix Structure: The student-content rating matrix is a matrix in the form of a two-dimensional table, usually denoted as R[m][n], where: Each student (row) is an index of the matrix, identified as m.

[0047] Each teaching content (column) is another index of the matrix, identified as n.

[0048] The matrix cell R[i][j] records the intensity of interaction or preference between student i and teaching content j, usually quantified by a rating. For example, if a student has interacted deeply with an article (such as reading it completely, commenting, and liking), a higher rating can be assigned to this cell.

[0049] The ratings in the matrix are calculated based on students' behavioral data and interest data. Common rating methods include the following: Explicit Rating: If the system allows students to directly rate teaching content (e.g., from 1 to 5 stars), these ratings are directly filled into the matrix to represent the students' preference for the content.

[0050] Implicit Rating: If there is no direct rating, the system infers the rating by analyzing behavioral data. For example: Learning Duration: If a student spends a long time on a certain article or video, it can be inferred that the student has a high interest in the content, so the rating is correspondingly high.

[0051] Completion Rate: Whether a student has completed a certain content is also an important indicator for rating. Completed content usually represents the student's higher interest or learning needs.

[0052] Interaction Behaviors: Such as liking, commenting, sharing, etc. The more interactions, the higher the rating.

[0053] Rating Range: Ratings are generally within a fixed range, such as from 1 to 5 or from 1 to 10. A higher rating indicates that the student has a higher interest or preference for the content, while a lower rating indicates lower interest or preference.

[0054] In most cases, students do not interact with all teaching contents, so many cells in the matrix are empty (i.e., there is no rating data). This is the sparsity problem of the rating matrix. To address this issue, the following methods can be adopted: Filling with Default Values: The system can use a default low score (such as 1 or 0) to represent that a student has not participated in a certain content.

[0055] Collaborative Filtering: By calculating the similarity with other students, fill in the missing ratings. For example, if two students' behaviors are highly similar, the system can infer the possible rating of one student for a certain content based on the rating of the other student.

[0056] The rating matrix is the basis of the collaborative filtering recommendation algorithm. In the subsequent steps, the system will analyze this matrix to identify the students' interest preferences and push appropriate teaching content based on the similarity between students. For example, if multiple students show a high level of interest in a certain type of content (such as history), the system will give priority to recommending similar content to these students. Through this matrix, the system can not only capture the students' current interests but also predict, through data analysis, the content that students may be interested in in the future, improving the personalization and accuracy of the recommended content.

[0057] In this application, the construction of the student-content rating matrix is achieved by collecting students' behavioral data and interest data to quantify the interaction relationship between students and teaching content. The matrix provides key data support for the recommendation system, helping the system identify the interest preferences of each student and providing a basis for subsequent personalized content pushing. By quantitatively describing the interaction intensity between students and content, the rating matrix provides a basis for the recommendation algorithm to analyze the similarity between student groups, thus realizing accurate content pushing.

[0058] The behavioral data has been represented by the student-content rating matrix (R matrix) in the previous step. In this matrix, the behavioral data of each student (such as ratings, viewing durations, completion rates, etc.) has been standardized into a rating, indicating the student's interest preference for a certain teaching content.

[0059] The rating matrix is denoted as R, where: R[i,j] represents the rating or interest value of student i for content j. i is the index of the student (ranging from 1 to m, representing m students), and j is the index of the content (ranging from 1 to m, representing m contents). The Pearson correlation coefficient formula is used to calculate the interest similarity between two students, and the expression is: In the formula, Pearson(A,B) is the Pearson correlation coefficient (similarity) between student A and student B, ranging from -1 to 1. 1 indicates a perfect positive correlation (high similarity), -1 indicates a perfect negative correlation (completely opposite), and 0 indicates no linear correlation. R[A,j] represents the rating of student A for content j. R[B,j] represents the rating of student B for content j. I is the set of contents jointly rated by student A and student B (that is, they have both rated these contents); is the average rating of student A for all contents, and the expression is: is the average rating of student B for all contents, and the expression is: The Pearson correlation coefficient measures the similarity between two students by comparing the differences in their ratings and the fluctuations in their ratings. If the ratings of two students for the same content are both close to their average ratings, their interest similarity is relatively high; otherwise, it is relatively low.

[0060] If the grading patterns of two students are similar, the Pearson correlation coefficient will be close to 1; if the differences are large, it will be close to -1.

[0061] In actual situations, it is impossible for students to rate all content, resulting in a sparse rating matrix. When calculating the Pearson correlation coefficient, only the content that both students have rated (i.e., set III) needs to be considered, and the unrated content is ignored.

[0062] After calculating the Pearson correlation coefficients between all pairs of students, the system will divide the students into students with similar interests and students with dissimilar interests according to a preset similarity threshold.

[0063] Similarity threshold: Set a similarity threshold T, for example, T = 0.5, to divide students into those with similar or dissimilar interests. The threshold T ranges from 0 to 1, and its strictness can be adjusted as needed. If Pearson(A,B) ≥ T, then student A and student B are considered students with similar interests; if Pearson(A,B) < T, then student A and student B are considered students with dissimilar interests.

[0064] S3: Analyze the ideological and political teaching content pushed to students with similar interests. According to the complexity characteristics and ideological depth characteristics of the pushed content, determine the accuracy of the recommendation algorithm's understanding of the content pushed to students with similar interests.

[0065] When analyzing the ideological and political teaching content pushed to students with similar interests, it is necessary to focus on examining the complexity characteristics and ideological depth characteristics of the content. The complexity characteristics include the information volume, theoretical depth, historical background, and understanding difficulty of the content to ensure that the pushed content conforms to the cognitive abilities and learning stages of the students. The ideological depth characteristics measure the effectiveness of the content in conveying core ideas (such as spirit, patriotism, social responsibility, etc.) to ensure that the content can guide students to form correct values. By analyzing these characteristics, it can be evaluated whether the pushed content of the recommendation algorithm not only conforms to the students' interests but also achieves the educational goals of ideological and political teaching.

[0066] Generate an information density complexity index according to the complexity characteristics of the pushed content. The method for obtaining the information density complexity index is as follows:

[0067] Before calculation, it is first necessary to preprocess the text so that subsequent operations can proceed smoothly. Split the text content into independent words or phrases. Remove meaningless common words (such as "de", "shi", "he", etc.) to highlight key information. Restore the words to their basic forms and unify different inflections of verbs and nouns.

[0068] In a graph-based model, each word in the text is represented as a node, and the co-occurrence relationship between words is represented as an edge, and these edges are defined according to the relative positions of the words in the text. Each word wi As a node V of a graph i , if there are h words after text segmentation, then there are {V1, V2, …, V h}; The edges E in the graph represent the co-occurrence relationship between words, that is, the degree to which words are adjacent to each other in a sentence. The weight w i,j of an edge represents the association strength between two words w i and w j . The weight of an edge can be calculated by the co-occurrence times of words in a window, or further adjusted by semantic similarity (such as cosine similarity). Define a window of a fixed size (for example, size 2 or 3) for each word, which represents the words that co-occur with this word within the window range. The size of the window affects the density of the graph. The words within the window are connected to generate edges, and the expression is: Finally, a weighted graph G=(V, E) is formed, where V is the set of nodes (words), E is the set of edges (relationships between words), and each edge has a weight w i,j . Use the TextRank algorithm (based on the PageRank principle) to calculate the importance score of each node (word). The TextRank algorithm iteratively calculates the scores of each node. The higher the score, the greater the contribution of the word to the text information and the higher the information complexity. The expression is: In the formula, S(V i ) is the importance score of node V i , which represents the contribution degree of word w i in the whole text. d is the damping coefficient, usually taking the value of 0.85, which is used to simulate the jump probability and avoid the score being too high caused by local loops. In(V i ) is the set of nodes pointing to node V i , that is, all the words connected to V i . represents the sum of the weights of all edges emitted from node V j , which represents the out-degree of node V j . By iteratively updating the importance scores of each node (word), usually stop iterating until the score change value of all nodes is less than a preset threshold. Finally, the score S(V i ) of each node represents the relative importance of this word in the text. After calculating the importance scores of all nodes, use these scores to calculate the information density complexity index of the whole text. The expression is: In the formula, KM is the information density complexity index.

[0069] A higher information density complexity index indicates that the recommended educational content contains a high volume and complexity of information, with strong vocabulary connections and a focused and profound core message. This is generally suitable for students with strong learning abilities and advanced cognitive abilities, as they are able to process more complex knowledge structures and ideas. However, if the content recommended to students with similar interests is too complex and exceeds their comprehension capabilities, it may make it difficult for them to digest the information, impacting their learning outcomes. In this case, the recommendation algorithm may have a lower accuracy in understanding the content because it fails to effectively match the student's cognitive level with the difficulty of the recommended content.

[0070] Conversely, a lower information density complexity index indicates relatively simple content, less information, and weaker vocabulary connections, making it easier to understand and suitable for students at a basic cognitive stage. However, if the index is too low, the content may be too superficial to meet the learning needs of advanced students and fail to inspire them to think deeply about complex ideas. In this case, the recommendation algorithm may not accurately grasp students' actual learning abilities and needs, causing the content to deviate from educational objectives. Therefore, low information density may also indicate that the recommendation algorithm's understanding accuracy is insufficient.

[0071] The historical reality relevance index is generated based on the depth characteristics of thought. The method for obtaining the historical reality relevance index is:

[0072] Split the text into sentences and extract sentences referring to historical events and current issues. Keyword matching, topic classification, or manual annotation can be used to identify which sentences pertain to historical events and which to current issues. For example, sentences pertaining to historical events might include the word "war," while sentences pertaining to current issues might involve "economy" or "technological innovation."

[0073] Use the BERT model to convert each sentence into a vector representation, capturing the semantic information of each sentence. Use a pre-trained BERT model (such as bert-base-uncased) to embed the sentence into the vector space. BERT's bidirectional encoder can effectively understand the context of the sentence.

[0074] Each historical event sentence and real-world problem sentence is input into the BERT model to extract the embedding vector of each sentence. BERT usually outputs the embedding vector of each word. A common method is to use the vector of the [CLS] tag as the embedding representation of the entire sentence. The formula is: V 句子 =BERTS 句子 ; Where: V 句子 It is the vector representation of the sentence, and the dimension is usually 768 (BERT-base model), S 句子The input is the sentence text. BERT represents vectorizing the sentence through the BERT model; the semantic similarity between the historical event sentence and the real - world problem sentence is measured by cosine similarity, and the calculation formula is: cosine similarity V 历史 is the embedding vector of the historical event sentence, V 现实 is the embedding vector of the real - world problem sentence, and V represents the Euclidean norm of the vector; each historical event sentence is compared pairwise with each real - world problem sentence to calculate their cosine similarity. Based on the similarity between all historical event sentences and real - world problem sentences, the historical - reality correlation index is calculated, and the formula is: cosine similarity(V 历史,i ,V 现实,j ); where p is the number of historical event sentences, q is the number of real - world problem sentences, and GF is the historical - reality correlation index.

[0075] When the historical - reality correlation index is larger, it indicates that the semantic relevance between the historical events and modern real - world problems in the pushed content is stronger, and the content can effectively combine the historical background with current social problems. This kind of content with high relevance helps students understand the reality through history, thereby deepening their understanding of spirit and social responsibility. Therefore, a larger historical - reality correlation index indicates that the recommendation algorithm has a more accurate understanding of the pushed content for students with similar interests and can meet the core goal of ideological and political teaching, that is, guiding students to understand values and a sense of responsibility in the real society through the study of historical events.

[0076] On the contrary, when the historical - reality correlation index is smaller, it indicates that the semantic relevance between the historical events and real - world problems in the pushed content is weaker, and the pushed content is more inclined to a single historical narrative and fails to closely connect it with modern society. Although this kind of pushed content may meet the interests of students, it cannot fully play the role of ideological guidance in ideological and political teaching and cannot help students have in - depth conversations with contemporary social problems. Therefore, a smaller historical - reality correlation index means that the recommendation algorithm has a lower accuracy in understanding the content, and the pushed content fails to fully meet the requirements of ideological and political teaching and may not effectively stimulate students' thinking about real - world problems and their recognition of social responsibility.

[0077] Convert the information density complexity index and the historical reality correlation index into a first feature vector, and use the first feature vector as the input of the machine learning model. The machine learning model takes the prediction target as the accuracy value label of the recommended algorithm's understanding of the content pushed to students with similar interests, and uses minimizing the sum of the prediction errors of the accuracy value labels of all recommended algorithms' understanding of the content pushed to students with similar interests as the training target. Train the machine learning model until the sum of the prediction errors converges and then stop the model training. Determine the accuracy value of the recommended algorithm's understanding of the content pushed to students with similar interests according to the model output result. Among them, the machine learning model is a polynomial regression model.

[0078] The method for obtaining the accuracy value of the recommended algorithm's understanding of the content pushed to students with similar interests is: obtain the corresponding function expression from the first feature vector training data of the trained machine learning model: FD = F(KM, GF); where F is the output function of the model, KM is the information density complexity index, GF is the historical reality correlation index, and FD is the accuracy value of the recommended algorithm's understanding of the content pushed to students with similar interests.

[0079] S4: Divide the accuracy of the recommended algorithm's understanding of the content pushed to students with similar interests into different levels, divide the recommended content into accurate push, incompletely accurate push, and inaccurate push, and perform corresponding processing.

[0080] Compare the obtained accuracy value of the recommended algorithm's understanding of the content pushed to students with similar interests with the gradient standard thresholds. The gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the accuracy value of the recommended algorithm's understanding of the content pushed to students with similar interests with the first standard threshold and the second standard threshold respectively;

[0081] If the accuracy value of the recommended algorithm's understanding of the content pushed to students with similar interests is greater than the second standard threshold, it indicates that the accuracy of the recommended algorithm's understanding of the content pushed to students with similar interests is high. At this time, generate a high-accuracy signal and divide the recommended content into accurate push;

[0082] If the accuracy value of the recommended algorithm's understanding of the content pushed to students with similar interests is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the accuracy of the recommended algorithm's understanding of the content pushed to students with similar interests is average. At this time, generate a medium-accuracy signal and divide the recommended content into incompletely accurate push;

[0083] If the accuracy value of the recommended algorithm's understanding of the content pushed to students with similar interests is less than the first standard threshold, it indicates that the accuracy of the recommended algorithm's understanding of the content pushed to students with similar interests is low. At this time, generate a low-accuracy signal and divide the recommended content into inaccurate push.

[0084] S5: For students with similar interests for whom the push is not completely accurate, predict the degree of abnormality of the understanding accuracy of the push content by the recommendation algorithm for students with similar interests within a fixed time period, and dynamically adjust the push frequency and push content according to the prediction results.

[0085] For students with similar interests for whom the push is not completely accurate, that is, the understanding accuracy value of the push content for students with similar interests generated by the recommendation algorithm within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold. Collect the understanding accuracy values of the push content for students with similar interests generated by the recommendation algorithm that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold in the subsequent fixed time period, establish a corresponding data set, and calculate the mean and standard deviation of the data set.

[0086] If the mean of the understanding accuracy values of the push content in the data set is greater than or equal to the reference threshold of the mean of the understanding accuracy values of the push content, and the standard deviation of the understanding accuracy values of the push content is less than the reference threshold of the standard deviation of the understanding accuracy values of the push content, it indicates that the accuracy of the push content is relatively high and the fluctuation is small, which means the performance of the recommendation algorithm is stable and the quality of the push content is high. At this time, no warning signal is generated, the system does not need to be adjusted frequently, and the content is pushed according to the current push frequency.

[0087] If the mean of the understanding accuracy values of the push content is greater than or equal to the reference threshold of the mean of the understanding accuracy values of the push content, and the standard deviation of the understanding accuracy values of the push content is greater than or equal to the reference threshold of the standard deviation of the understanding accuracy values of the push content, although the understanding accuracy of the push content is high but the fluctuation is large, it means the performance of the content is unstable, and some push content may be suitable while some may not be suitable for students.

[0088] At this time, a level-three warning signal is generated, and the system needs to moderately adjust the selection of the push content, reduce the fluctuation of the content quality, and increase the overall push frequency.

[0089] If the mean of the understanding accuracy values of the push content is less than the reference threshold of the mean of the understanding accuracy values of the push content, and the standard deviation of the understanding accuracy values of the push content is greater than or equal to the reference threshold of the standard deviation of the understanding accuracy values of the push content, it indicates that the accuracy of the push content is low and the fluctuation is large, which means the recommendation algorithm does not accurately understand the students, and the quality of the push content is poor and unstable. At this time, a level-one warning signal is generated, and a large-scale adjustment is made to re-optimize the push content and frequency to improve the recommendation accuracy.

[0090] If the mean of the understanding accuracy values of the push content is less than the reference threshold of the mean of the understanding accuracy values of the push content, and the standard deviation of the understanding accuracy values of the push content is less than the reference threshold of the standard deviation of the understanding accuracy values of the push content, it indicates that the accuracy of the push content is low but the fluctuation is small, which means that although the performance of the recommended content is poor, it is relatively consistent. At this time, a level-two warning signal is generated, and the content is adjusted to improve the overall quality of the content.

[0091] It should be noted here that the importance level of the first-level warning signal is greater than that of the second-level warning signal, and the importance level of the second-level warning signal is greater than that of the third-level warning signal. Relevant personnel can take corresponding adjustment measures according to different warning signal levels respectively.

[0092] In this embodiment, by collecting students' behavior and interest data, a student-content scoring matrix is constructed, the similarity between students is calculated using the Pearson correlation coefficient, and students are divided into groups with similar and dissimilar interests. For the ideological and political teaching content pushed to students with similar interests, its complexity and ideological depth characteristics are analyzed to evaluate the understanding accuracy of the recommendation algorithm for the pushed content. According to the accuracy, the pushed content is divided into accurate, incompletely accurate, and inaccurate pushes, and they are processed separately. For students with incompletely accurate pushes, the system predicts abnormal situations in the understanding accuracy of the pushed content within a fixed time period, and dynamically adjusts the push frequency and content according to the results to optimize the recommendation effect.

[0093] The above formulas are all calculated by removing the dimension and taking their numerical values. The formula is a formula obtained by collecting a large amount of data for software simulation to get the closest to the real situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0094] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but this implementation should not be considered to exceed the scope of this application.

[0095] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. An ideological and political teaching customized push method based on artificial intelligence, characterized in that: It includes the following steps; S1: Collect the behavior data and interest data of students, and construct a student-content scoring matrix, where each student corresponds to a row in the matrix and the teaching content corresponds to a column; S2: Based on the behavior data of students, calculate the similarity between all students through the Pearson correlation coefficient, compare the similarity with a preset similarity threshold, and divide the students into students with similar interests and students with dissimilar interests according to the comparison results; S3: Analyze the ideological and political teaching content pushed to students with similar interests, and determine the understanding accuracy of the recommended algorithm for the content pushed to students with similar interests according to the complexity characteristics and ideological depth characteristics of the pushed content; S4: Divide the understanding accuracy of the recommended algorithm for the content pushed to students with similar interests into different levels, divide the recommended content into accurate push, incomplete accurate push and inaccurate push, and perform corresponding processing; S5: For students with similar interests with incomplete accurate push, predict the abnormal degree of the understanding accuracy of the recommended algorithm for the content pushed to students with similar interests within a fixed time period, and dynamically adjust the push frequency and push content according to the prediction results.

2. The customized push method for ideological and political teaching based on artificial intelligence according to claim 1 is characterized in that: In S2, the rating matrix is denoted as R, where: R[i, j] represents the rating of student i on content j, i is the index of the student, and j is the index of the content. The Pearson correlation coefficient formula is used to calculate the interest similarity between two students, and the expression is: In the formula, Pearson(A, B) is the Pearson correlation coefficient between student A and student B, R[A, j] represents the rating of student A on content j, R[B, j] represents the rating of student B on content j, and I is the set of contents jointly rated by student A and student B; is the average rating of student A on all contents, and the expression is: is the average rating of student B on all contents, and the expression is: Set a similarity threshold T. If Pearson(A,B)≥T, then students A and B are considered students with similar interests; if Pearson(A,B)<T, then students A and B are considered students with dissimilar interests.

3. The customized push method for ideological and political teaching based on artificial intelligence according to claim 1, characterized in that: Generate an information density complexity index according to the complexity characteristics of the pushed content. The acquisition method of the information density complexity index is: In a graph-based model, each word in the text is represented as a node, and the co-occurrence relationship between words is represented as an edge. For each word w i as a node V of a graph i , if there are h words after text tokenization, then there are {V1, V2, …, V h}; The edge E in the graph represents the co-occurrence relationship between words, and the weight w of the edge i,j represents the association strength between two words w i and w j . A fixed-size window is defined for each word, and the expression is: Finally, a weighted graph G = (V, E) is formed, where V is the node set, E is the edge set, and each edge has a weight w i,j ; Use the TextRank algorithm to iteratively calculate the score of each node, and the expression is: In the formula, S(V i ) is the importance score of node V i , representing the contribution of word w i in the entire text. d is the damping coefficient, In(V i ) is the set of nodes pointing to node V i , that is, all words connected to V i . represents the sum of the weights of all edges emitted from node V j , representing the out-degree of node V j . Continuously update the importance score of each node in an iterative manner until the score change value of all nodes is less than the preset threshold to stop the iteration. After calculating the importance scores of all nodes, use the scores to calculate the information density complexity index of the entire text, and the expression is: In the formula, KM is the information density complexity index.

4. The customized push method for ideological and political teaching based on artificial intelligence according to claim 3, characterized in that: Generate a historical-reality correlation index according to the ideological depth characteristics. The acquisition method of the historical-reality correlation index is: Split the text content into sentences, and extract the sentences involving historical events and real problems respectively; use the BERT model to convert each sentence into a vector representation to capture the semantic information of each sentence, and use the pre-trained BERT model to load for embedding the sentences into the vector space; Input each historical event sentence and real-world problem sentence into the BERT model to extract the embedding vector of each sentence. BERT outputs the embedding vector of each word, which is expressed by the formula: V 句子 = BERT(S 句子 ); Where: V 句子 is the vector representation of the sentence, S 句子 is the input sentence text, and the BERT representation vectorizes the sentence through the BERT model; the semantic similarity between the historical event sentence and the real - world problem sentence is measured by cosine similarity, and the calculation formula is: cosine similarity V 历史 is the embedding vector of the historical event sentence, V 现实 is the embedding vector of the real - world problem sentence, and ||V|| represents the Euclidean norm of the vector; each historical event sentence is compared pairwise with each real - world problem sentence to calculate their cosine similarity. Based on the similarity between all historical event sentences and real - world problem sentences, the historical - reality correlation index is calculated, and the formula is: Cosine similarity(V 历史,i , V 现实,j ); In the formula, p is the number of historical event sentences, q is the number of real - world problem sentences, and GF is the historical - reality correlation index.

5. The customized push method for ideological and political teaching based on artificial intelligence according to claim 4, characterized in that: Convert the information density complexity index and the historical-reality correlation index into a first feature vector, use the first feature vector as the input of a machine learning model, use the machine learning model to predict the understanding accuracy value label of the content pushed to students with similar interests by the recommended algorithm as the prediction target, and use minimizing the sum of the prediction errors of the understanding accuracy value labels of the content pushed to students with similar interests by all recommended algorithms as the training target to train the machine learning model until the sum of the prediction errors reaches convergence and stop the model training, and determine the understanding accuracy value of the content pushed to students with similar interests by the recommended algorithm according to the model output result. Among them, the machine learning model is a polynomial regression model.

6. The customized push method for ideological and political teaching based on artificial intelligence according to claim 5, characterized in that: In S4, divide the understanding accuracy of the recommended algorithm for the content pushed to students with similar interests into different levels, divide the recommended content into accurate push, incomplete accurate push and inaccurate push, specifically: Compare the understanding accuracy value of the content pushed to students with similar interests by the obtained recommendation algorithm with the gradient standard threshold. The gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the understanding accuracy value of the content pushed to students with similar interests by the recommendation algorithm with the first standard threshold and the second standard threshold respectively; If the understanding accuracy value of the content pushed to students with similar interests by the recommendation algorithm is greater than the second standard threshold, it indicates that the understanding accuracy of the content pushed to students with similar interests by the recommendation algorithm is high. At this time, generate a high-accuracy signal and classify the recommended content as accurately pushed; If the understanding accuracy value of the content pushed to students with similar interests by the recommendation algorithm is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the understanding accuracy of the content pushed to students with similar interests by the recommendation algorithm is average. At this time, generate a medium-accuracy signal and classify the recommended content as incompletely accurately pushed; If the understanding accuracy value of the content pushed to students with similar interests by the recommendation algorithm is less than the first standard threshold, it indicates that the understanding accuracy of the content pushed to students with similar interests by the recommendation algorithm is low. At this time, generate a low-accuracy signal and classify the recommended content as inaccurately pushed.

7. The customized push method for ideological and political teaching based on artificial intelligence according to claim 1, wherein: In S4, for students with similar interests who are incompletely accurately pushed, predict the abnormality degree of the understanding accuracy of the recommendation algorithm for the pushed content for students with similar interests within a fixed time period. Specifically: For students with similar interests who are incompletely accurately pushed, that is, the understanding accuracy value of the content pushed to students with similar interests generated by the recommendation algorithm within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold. Collect the understanding accuracy values of the content pushed to students with similar interests by the recommendation algorithm that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold generated in the subsequent fixed time period, establish a corresponding data set, and calculate the mean and standard deviation of the data set.

8. An AI-based customized push method for ideological and political teaching according to claim 7, characterized in that: If the mean value of the understanding accuracy values of the pushed content in the data set is greater than or equal to the reference threshold of the mean value of the understanding accuracy values of the pushed content, and the standard deviation of the understanding accuracy values of the pushed content is less than the reference threshold of the standard deviation of the understanding accuracy values of the pushed content, the accuracy of the pushed content is relatively high and the fluctuation is small, indicating that the performance of the recommendation algorithm is stable and the quality of the pushed content is high. At this time, do not generate a warning signal and continue to push content according to the current push frequency; If the mean value of the understanding accuracy values of the pushed content is greater than or equal to the reference threshold of the mean value of the understanding accuracy values of the pushed content, and the standard deviation of the understanding accuracy values of the pushed content is greater than or equal to the reference threshold of the standard deviation of the understanding accuracy values of the pushed content, although the understanding accuracy of the pushed content is high but the fluctuation is large, the performance of the recommended content is unstable. At this time, generate a level-three warning signal, adjust the selection of the pushed content, reduce the fluctuation of the content quality, and increase the overall push frequency; If the average value of the push content understanding accuracy is less than the reference threshold of the average value of the push content understanding accuracy, and the standard deviation of the push content understanding accuracy is greater than or equal to the reference threshold of the standard deviation of the push content understanding accuracy, the accuracy of the push content is low and the fluctuation is large, indicating that the recommendation algorithm does not accurately understand the students, and the quality of the push content is low and unstable. At this time, a first-level warning signal is generated, and major adjustments are made to re-optimize the push content and frequency to improve the recommendation accuracy; If the average value of the push content understanding accuracy is less than the reference threshold of the average value of the push content understanding accuracy, and the standard deviation of the push content understanding accuracy is less than the reference threshold of the standard deviation of the push content understanding accuracy, the accuracy of the push content is low but the fluctuation is small. At this time, a second-level warning signal is generated, and content adjustments are made to improve the overall quality of the content.

Images