Higher education textbook quality evaluation method and system based on particle swarm big language model

Abstract

The invention provides a higher education textbook quality evaluation method and system based on a particle swarm big language model, and relates to the technical field of education evaluation. The method comprises the steps of collecting text data of higher education textbooks; preprocessing the text data by using a particle swarm optimization algorithm to obtain preprocessed data; constructing a large language model based on a particle swarm, and performing feature extraction on the preprocessed data through the large language model to obtain feature data; the feature data comprises semantic features, structural features and quality features; performing comprehensive evaluation on the feature data through the evaluation function to obtain a comprehensive evaluation score; and comparing the comprehensive evaluation score with a preset standard score to determine the quality grade of the textbook. By comprehensively evaluating the textbook features, flexibly optimizing feature selection and scientifically setting the quality standard, the accuracy and practicability of textbook quality evaluation are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of educational assessment technology, and in particular to a method and system for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model. Background Technology

[0002] In modern higher education, the quality of textbooks directly impacts teaching effectiveness and students' learning experience. Traditional textbook evaluation methods primarily rely on expert review. While expert review incorporates professional knowledge, this method has several drawbacks. First, professional review depends on the reviewer's subjective judgment, leading to poor objectivity, significant individual differences and biases, and difficulties in reaching consensus among different experts. Second, the expert review process is complex and time-consuming, especially with a large number of textbooks, requiring weeks or even months and failing to meet the need for rapid feedback. Finally, traditional methods struggle with large-scale textbook data analysis, thus overlooking potentially excellent textbooks and rapidly evolving educational needs.

[0003] With the rapid development of artificial intelligence technology, especially the maturity of Natural Language Processing (NLP) technology, text analysis-based textbook evaluation methods have gradually become a research hotspot. These methods can automate and deeply analyze textbook content, providing more accurate evaluations. However, existing NLP-based evaluation systems still have shortcomings, mainly in the following aspects: First, most existing models rely heavily on static text feature extraction, lacking dynamic learning and self-optimization capabilities, and cannot adapt to the rapid updates and changes in textbook content. Second, existing models often neglect in-depth analysis of the textbook's structure and logical relationships, focusing only on single semantic content, leading to insufficient multi-dimensional understanding of textbook quality during evaluation.

[0004] Particle swarm optimization (PSO), as a swarm intelligence-based optimization algorithm, has demonstrated excellent performance in data analysis and processing in recent years. Combining PSO with large language models enables in-depth analysis and evaluation of higher education textbooks. By dynamically adjusting the optimization algorithm and enhancing the model's adaptability, the shortcomings of existing methods can be effectively overcome. Therefore, a method for evaluating the quality of higher education textbooks based on PSO and large language models has emerged, becoming an important research direction for improving the accuracy and efficiency of textbook evaluation. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model. By comprehensively evaluating textbook characteristics, flexibly optimizing feature selection, and scientifically setting quality standards, the method significantly improves the accuracy and practicality of textbook quality evaluation.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model includes:

[0008] Collect textual data from higher education textbooks;

[0009] The text data is preprocessed using the particle swarm optimization algorithm to obtain preprocessed data;

[0010] A large language model based on particle swarm optimization is constructed, and feature extraction is performed on the preprocessed data using the large language model to obtain feature data; the feature data includes semantic features, structural features, and quality features.

[0011] The feature data are comprehensively evaluated using an evaluation function to obtain a comprehensive evaluation score;

[0012] The comprehensive evaluation score is compared with the preset standard score to determine the quality level of the teaching materials.

[0013] Preferably, the text data includes chapters, content summaries, examples, and exercises from the higher education textbook.

[0014] Preferably, the text data is preprocessed using a particle swarm optimization algorithm to obtain preprocessed data, including:

[0015] The text data is cleaned to remove irrelevant information, noise, and redundant data, resulting in cleaned data.

[0016] The cleaned data is segmented into words or phrases to obtain segmented data.

[0017] The particle swarm optimization algorithm is used to select features from the segmented data to identify the most representative words and features.

[0018] The selected words and feature terms are vectorized to generate corresponding numerical representations, thus obtaining vectorized features.

[0019] The vectorized features are standardized to ensure that each vectorized feature is within the same range, resulting in preprocessed data.

[0020] Preferably, the word segmentation data is subjected to feature selection using a particle swarm optimization algorithm to identify the most representative words and feature terms, including:

[0021] According to formula F i =w×f i +α×I i +β×Gi Calculate the initial fitness value for each word; where F i Let I be the initial fitness value for segmented data i, w be the weighting coefficient controlling the influence of segmentation frequency on the fitness value, and α be the weighting coefficient controlling the influence of information gain on the fitness value. i To measure the amount of information contributed by segmented data i to the classification result, β is a weighting coefficient that controls the influence of mutual information on the fitness value, and G... i The value of the mutual information of the segmented data i;

[0022] Initialize the particle swarm and assign each particle a random subset of features to represent the vocabulary to be selected;

[0023] By iteratively updating the positions of particles, the feature subset is adjusted using the particle swarm optimization algorithm to maximize the overall fitness function;

[0024] Evaluate the real-time fitness of the current particle position;

[0025] The best-performing feature subset is selected based on the real-time fitness, and this best-performing feature subset is used as the most representative vocabulary and feature terms to be selected in the end.

[0026] Preferably, a large language model based on particle swarm optimization is constructed, and feature extraction is performed on the preprocessed data using the large language model to obtain feature data, including:

[0027] The structure and parameters of the large language model are defined; wherein, the number of layers of the large language model is determined to be 4, namely one input layer, two hidden layers and one output layer; the number of neurons in the input layer is 500; the number of neurons in the first hidden layer is 256; the number of neurons in the second hidden layer is 128; the activation function of the hidden layer is ReLU; and the activation function of the output layer is Softmax.

[0028] Design the training dataset and corresponding labels based on the preprocessed data;

[0029] The parameters of the large language model are initialized using a particle swarm optimization algorithm to set the number of particles and the initialization position.

[0030] Based on the training dataset and the corresponding labels, the weights of the large language model are updated through an iterative optimization algorithm, and the model performance is evaluated using a cross-validation method to obtain a well-evaluated feature extraction model.

[0031] The preprocessed data is used to extract features using the feature extraction model to obtain the semantic features, the structural features, and the quality features. The semantic features are the meaning and contextual information extracted from the text. The structural features are used to describe the attributes of the text structure. The structural features include chapter division, paragraph length, and sentence complexity. The quality features are indicators used to measure the quality of the teaching materials. The quality features include readability index and content consistency score.

[0032] Preferably, the calculation formula for the evaluation function is:

[0033] S=γ·S semantic +δ·S structural +θ·S quality

[0034] Where S is the comprehensive evaluation score, representing the overall assessment of the textbook quality. semantic The scoring of semantic features reflects the depth and relevance of the text content. structural The scoring for structural features reflects the organization and logic of the teaching materials. quality The quality features are scored to assess readability, error rate, or consistency metrics. γ is the weighting coefficient of semantic features, which determines the importance of semantic features in the overall score. δ is the weighting coefficient of structural features, which determines the importance of structural features in the overall score. θ is the weighting coefficient of quality features, which determines the importance of quality features in the overall score.

[0035] The formula for calculating the score of the semantic feature is as follows:

[0036]

[0037] Where N is the number of features, TFIDFF i It is the TF-IDF value of the segmented data i;

[0038] The formula for calculating the score of the structural feature is as follows:

[0039]

[0040] Where M is the number of structural features, L j It is the length of chapter j, L max It is the maximum chapter length, P j It refers to the number of paragraphs;

[0041] The formula for calculating the score of the quality characteristic is as follows:

[0042]

[0043] Where R is the readability score, D is the document length, and C is the content consistency score.

[0044] Preferably, the comprehensive evaluation score is compared with a preset standard score to determine the quality level of the teaching material, including:

[0045] A preset standard score is determined, which includes a score range for different quality levels;

[0046] The comprehensive evaluation score is standardized to obtain a standardized comprehensive score.

[0047] Based on the range of the standard scores, the standardized composite score is mapped to a specific quality level.

[0048] Preferably, the standardized score S normalized The formula is:

[0049]

[0050] Where S is the comprehensive evaluation score, S min S is the lowest score in the sample. max The highest score in the sample.

[0051] Preferably, mapping the standardized composite score to a specific quality level based on the range of the standard scores includes:

[0052] If S normalized If the score is ≥0.9, the standardized composite score will be marked as "excellent".

[0053] If 0.75≤S normalized If the score is less than 0.9, the standardized composite score will be marked as "good".

[0054] If 0.5≤S normalized If the score is less than 0.75, the standardized composite score will be marked as "moderate".

[0055] If 0.3≤S normalized If the score is less than 0.5, the standardized composite score will be marked as "passing".

[0056] If S normalized If the score is less than 0.3, the standardized composite score will be marked as "failing".

[0057] A quality evaluation system for higher education textbooks based on a particle swarm optimization large language model includes:

[0058] The data collection unit is used to collect textual data from higher education textbooks;

[0059] The data preprocessing unit is used to preprocess the text data using the particle swarm optimization algorithm to obtain preprocessed data;

[0060] The feature extraction unit is used to construct a large language model based on particle swarm optimization, and to extract features from the preprocessed data using the large language model to obtain feature data; the feature data includes semantic features, structural features, and quality features.

[0061] The comprehensive evaluation unit is used to comprehensively evaluate the feature data through an evaluation function to obtain a comprehensive evaluation score;

[0062] The quality determination unit is used to compare the comprehensive evaluation score with a preset standard score to determine the quality level of the teaching materials.

[0063] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0064] This invention provides a method and system for evaluating the quality of higher education textbooks based on a particle swarm optimization (PSO) large language model. The method includes: collecting text data of higher education textbooks; preprocessing the text data using a particle swarm optimization algorithm to obtain preprocessed data; constructing a large language model based on PSO, and extracting features from the preprocessed data using the large language model to obtain feature data; the feature data includes semantic features, structural features, and quality features; comprehensively evaluating the feature data using an evaluation function to obtain a comprehensive evaluation score; and comparing the comprehensive evaluation score with a preset standard score to determine the quality level of the textbook. This invention significantly improves the accuracy and practicality of textbook quality evaluation by comprehensively evaluating textbook features, flexibly optimizing feature selection, and scientifically setting quality standards. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0067] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] The purpose of this invention is to provide a method and system for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model. By comprehensively evaluating textbook characteristics, flexibly optimizing feature selection, and scientifically setting quality standards, the method significantly improves the accuracy and practicality of textbook quality evaluation.

[0070] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0071] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model, including:

[0072] Step 100: Collect textual data of higher education textbooks;

[0073] Step 200: Use the particle swarm optimization algorithm to preprocess the text data to obtain preprocessed data;

[0074] Step 300: Construct a large language model based on particle swarm optimization, and extract features from the preprocessed data using the large language model to obtain feature data; the feature data includes semantic features, structural features, and quality features;

[0075] Step 400: Perform a comprehensive evaluation of the feature data using an evaluation function to obtain a comprehensive evaluation score;

[0076] Step 500: Compare the comprehensive evaluation score with the preset standard score to determine the quality level of the teaching material.

[0077] Preferably, the text data includes chapters, content summaries, examples, and exercises from the higher education textbook.

[0078] Specifically, in step 100, this embodiment establishes a data acquisition framework, clearly defining the target textbook category, applicable scope, and required information type. Targeted data collection and downloading are implemented using major online education platforms, university textbook procurement websites, and library digital resources. The collected text data includes each chapter of the textbook, along with corresponding content summaries, examples, and exercises. This content is obtained through batch downloading, API (Application Programming Interface) calls, etc., ensuring data integrity and diversity. For each type of textbook, corresponding data formats and structures are developed for subsequent processing and analysis. Furthermore, this embodiment involves regularly updating the data source to maintain the timeliness and relevance of the textbook information.

[0079] Secondly, during data processing, this embodiment performs preliminary cleaning and organization of the collected text data. Specifically, it is necessary to identify and remove redundant information, advertisements, copyright notices, and irrelevant text content to ensure data purity and quality. Furthermore, this embodiment establishes a standardized template to unify the formatting of chapters, content summaries, examples, and exercises, thereby facilitating subsequent data analysis and feature extraction. These operations can be automated to improve efficiency and reduce the risk of human intervention. Simultaneously, to ensure compliance with relevant intellectual property laws and regulations, the data source must be acknowledged and necessary authorization obtained during data collection to avoid copyright infringement. The acquisition and organization of data will provide reliable foundational data support for the subsequent construction of a large language model based on particle swarm optimization.

[0080] Preferably, the text data is preprocessed using a particle swarm optimization algorithm to obtain preprocessed data, including:

[0081] The text data is cleaned to remove irrelevant information, noise, and redundant data, resulting in cleaned data.

[0082] The cleaned data is segmented into words or phrases to obtain segmented data.

[0083] The particle swarm optimization algorithm is used to select features from the segmented data to identify the most representative words and features.

[0084] The selected words and feature terms are vectorized to generate corresponding numerical representations, thus obtaining vectorized features.

[0085] The vectorized features are standardized to ensure that each vectorized feature is within the same range, resulting in preprocessed data.

[0086] Specifically, in step 200, this embodiment first performs data cleaning. A cleaning program is written to automatically identify and remove irrelevant information, noise, and redundant data. Specific operations include using regular expressions and text analysis tools to traverse the collected text data, removing specific symbols, HTML tags, duplicate content, and unnecessary spaces. This process ensures that only content relevant to the analysis is retained, such as chapter titles, paragraph text, examples, and exercises. The cleaned text should exhibit a uniform format to facilitate subsequent processing and analysis.

[0087] The cleaned data undergoes word segmentation, breaking the text down into individual words or phrases. Word segmentation is performed using Natural Language Processing (NLP) libraries (such as NLTK, spaCy, or Jieba). The selected segmenter splits words according to predefined language rules and a dictionary, ensuring accuracy and effectiveness. From the resulting segmented data, a particle swarm optimization algorithm is used for feature selection. By initializing the particle swarm, each particle represents a subset of features from the segmented data. The particles iteratively adjust their positions to find feature combinations that maximize the fitness function. Finally, the most representative words and features are obtained from the feature selection, and vectorization techniques (such as Word2Vec or TF-IDF) are used to convert the selected words into numerical representations. These vectorized features are further standardized to ensure a consistent data range for all features. This process in this embodiment ensures that all features are within the same range, and the resulting preprocessed data lays a solid foundation for subsequent model building and feature extraction.

[0088] Preferably, the word segmentation data is subjected to feature selection using a particle swarm optimization algorithm to identify the most representative words and feature terms, including:

[0089] According to formula F i =w×f i +α×I i +β×G i Calculate the initial fitness value for each word; where F i Let I be the initial fitness value for segmented data i, w be the weighting coefficient controlling the influence of segmentation frequency on the fitness value, and α be the weighting coefficient controlling the influence of information gain on the fitness value. i To measure the amount of information contributed by segmented data i to the classification result, β is a weighting coefficient that controls the influence of mutual information on the fitness value, and G... i The value of the mutual information of the segmented data i;

[0090] Initialize the particle swarm and assign each particle a random subset of features to represent the vocabulary to be selected;

[0091] By iteratively updating the positions of particles, the feature subset is adjusted using the particle swarm optimization algorithm to maximize the overall fitness function;

[0092] Evaluate the real-time fitness of the current particle position;

[0093] The best-performing feature subset is selected based on the real-time fitness, and this best-performing feature subset is used as the most representative vocabulary and feature terms to be selected in the end.

[0094] Furthermore, in this embodiment, each particle represents a random subset of features from the segmented data. Each particle is randomly assigned a set of features based on an initial fitness value, reflecting the quality of the current feature combination. The position of each particle is updated through an iterative process, and the feature subset is adjusted using a particle swarm optimization algorithm to find the feature combination that maximizes the overall fitness function. In each iteration, the real-time fitness of the current particle's position is evaluated, and the best-performing feature subset is recorded. After multiple iterations, the best-performing feature subset is finally selected. These features are identified as the most representative words and feature terms, which can be used for subsequent model training and data analysis.

[0095] Preferably, a large language model based on particle swarm optimization is constructed, and feature extraction is performed on the preprocessed data using the large language model to obtain feature data, including:

[0096] The structure and parameters of the large language model are defined; wherein, the number of layers of the large language model is determined to be 4, namely one input layer, two hidden layers and one output layer; the number of neurons in the input layer is 500; the number of neurons in the first hidden layer is 256; the number of neurons in the second hidden layer is 128; the activation function of the hidden layer is ReLU; and the activation function of the output layer is Softmax.

[0097] Design the training dataset and corresponding labels based on the preprocessed data;

[0098] The parameters of the large language model are initialized using a particle swarm optimization algorithm to set the number of particles and the initialization position.

[0099] Based on the training dataset and the corresponding labels, the weights of the large language model are updated through an iterative optimization algorithm, and the model performance is evaluated using a cross-validation method to obtain a well-evaluated feature extraction model.

[0100] The preprocessed data is used to extract features using the feature extraction model to obtain the semantic features, the structural features, and the quality features. The semantic features are the meaning and contextual information extracted from the text. The structural features are used to describe the attributes of the text structure. The structural features include chapter division, paragraph length, and sentence complexity. The quality features are indicators used to measure the quality of the teaching materials. The quality features include readability index and content consistency score.

[0101] Optionally, in this embodiment, when constructing a large language model based on particle swarm optimization, the model's structure and parameters need to be defined first. The large language model is set to a four-layer structure, including one input layer, two hidden layers, and one output layer. The input layer consists of 500 neurons and is responsible for receiving vectorized features from the preprocessed data. The first hidden layer has 256 neurons, and the second hidden layer has 128 neurons. All hidden layers use ReLU (Rectified Linear Unit) as the activation function to introduce non-linearity, enabling the model to learn complex features. The output layer uses Softmax as the activation function to transform the final output into a probability distribution, suitable for classification tasks.

[0102] Secondly, a training dataset and its corresponding labels are designed based on the preprocessed data to ensure the diversity and representativeness of the dataset. The training dataset will contain samples from each value category, and the labels will identify the category information corresponding to these samples. During training, the parameters of the large language model are initialized using a particle swarm optimization algorithm, setting the number of particles and their initial positions. Each particle is randomly distributed throughout the search space, representing a specific combination of model parameters. Then, the velocity and position of the particles are updated iteratively to maximize the fitness function, thereby finding the optimal parameter combination and ensuring improved model performance.

[0103] Finally, after parameter optimization, the weights of the large language model are continuously updated using an iterative optimization algorithm based on the training dataset and corresponding labels. Cross-validation is employed to evaluate model performance and ensure its generalization ability. After training and evaluation, the feature extraction model is used to extract features from the preprocessed data, ultimately extracting semantic, structural, and quality features. Specifically, semantic features include deep meaning and contextual information extracted from the text; structural features describe the text's structural attributes, such as chapter divisions, paragraph length, and sentence complexity; and quality features are used to measure textbook quality, including readability index and content consistency score. This series of explicit steps ensures the effectiveness and accuracy of feature extraction, providing a reliable basis for subsequent textbook evaluation.

[0104] Preferably, the calculation formula for the evaluation function is:

[0105] S=γ·Ssemantic +δ·S structural +θ·S quality

[0106] Where S is the comprehensive evaluation score, representing the overall assessment of the textbook quality. semantic The scoring of semantic features reflects the depth and relevance of the text content. structural The scoring for structural features reflects the organization and logic of the teaching materials. quality The quality features are scored to assess readability, error rate, or consistency metrics. γ is the weighting coefficient of semantic features, which determines the importance of semantic features in the overall score. δ is the weighting coefficient of structural features, which determines the importance of structural features in the overall score. θ is the weighting coefficient of quality features, which determines the importance of quality features in the overall score.

[0107] As an example, the weighting coefficients γ, δ, and θ can be set through empirical testing or based on the opinions of domain experts. Generally, their initial values ​​can be set as γ = 0.5, δ = 0.3, and θ = 0.2, and then adjusted according to the feedback from the training data.

[0108] The formula for calculating the score of the semantic feature is as follows:

[0109]

[0110] Where N is the number of features, TFIDFF i It is the TF-IDF value of the segmented data i;

[0111] The formula for calculating the score of the structural feature is as follows:

[0112]

[0113] Where M is the number of structural features, L j It is the length of chapter j, L max It is the maximum chapter length, P j It refers to the number of paragraphs;

[0114] The formula for calculating the score of the quality characteristic is as follows:

[0115]

[0116] Where R is the readability score, D is the document length, and C is the content consistency score.

[0117] Specifically, this embodiment comprehensively considers features across multiple dimensions, making the evaluation of textbook quality more comprehensive and detailed. By combining semantic, structural, and quality features, the formula effectively reflects the depth and relevance, organizational logic, readability, and consistency of textbook content, thus providing a multi-level dynamic evaluation mechanism. Furthermore, the use of specific indicators such as TF-IDF values, chapter length, and readability scores enables data-driven quantitative analysis, avoiding the influence of subjective evaluations. This comprehensive approach not only improves the accuracy of the evaluation but also provides more guiding feedback for different types of textbooks, promoting continuous improvement in textbook quality.

[0118] Furthermore, in the evaluation function, the relationship between the number of N features and the number of M structural features reflects the complexity and organization of the teaching materials at different levels. Specifically, the number of N features typically refers to the total number of feature items extracted from text analysis, including keywords and sentences, while the number of M structural features focuses on the structural attributes of the teaching materials, such as the number of chapters and paragraphs. R is calculated by weighted averaging the TF-IDF values ​​of each semantic feature, thereby assessing the depth and relevance of the text content. C is calculated by comparing the length of each chapter (L) with the maximum chapter length (L_max) and standardizing it in conjunction with the number of paragraphs (P), forming a comprehensive score for the logical organization of the teaching materials. This relationship and calculation method ensure the systematic and comprehensive nature of the evaluation process, better reflecting the overall quality of the teaching materials.

[0119] Preferably, the comprehensive evaluation score is compared with a preset standard score to determine the quality level of the teaching material, including:

[0120] A preset standard score is determined, which includes a score range for different quality levels;

[0121] The comprehensive evaluation score is standardized to obtain a standardized comprehensive score.

[0122] Based on the range of the standard scores, the standardized composite score is mapped to a specific quality level.

[0123] Preferably, the standardized score S normalized The formula is:

[0124]

[0125] Where S is the comprehensive evaluation score, S min S is the lowest score in the sample. max The highest score in the sample.

[0126] Preferably, mapping the standardized composite score to a specific quality level based on the range of the standard scores includes:

[0127] If S normalized If the score is ≥0.9, the standardized composite score will be marked as "excellent".

[0128] If 0.75≤S normalized If the score is less than 0.9, the standardized composite score will be marked as "good".

[0129] If 0.5≤S normalized If the score is less than 0.75, the standardized composite score will be marked as "moderate".

[0130] If 0.3≤S normalized If the score is less than 0.5, the standardized composite score will be marked as "passing".

[0131] If S normalized If the score is less than 0.3, the standardized composite score will be marked as "failing".

[0132] Optionally, in determining the quality level of teaching materials, it is first necessary to set a preset standard score, which will be divided into multiple score ranges according to the requirements of different quality levels. Specific implementation steps include collecting a large number of comprehensive evaluation scores for teaching materials to analyze their quality performance and setting reasonable scoring ranges based on educational standards and practical experience. For example, a comprehensive evaluation score in the range of 90-100 can be marked as "Excellent," 80-89 as "Good," 70-79 as "Average," 60-69 as "Pass," and below 60 as "Fail." This step ensures that the division of quality levels has a certain degree of scientific validity and applicability.

[0133] Next, the comprehensive evaluation scores are standardized. The comprehensive evaluation score is the original score of the textbook being evaluated, with the lowest and highest scores recorded in the sample, respectively. After standardization, the resulting standardized comprehensive scores are mapped to quality levels according to a pre-defined standard score range. This method compares the standardized scores with the corresponding quality levels, thus quickly and accurately determining the textbook's quality level. This process ensures the transparency and consistency of the evaluation results, providing data support and clear direction for textbook improvement and optimization.

[0134] As an optional implementation, this embodiment will also generate a report of the evaluation results, including a comprehensive evaluation score, a standardized score, a quality grade, and corresponding suggestions, which will be provided to textbook writers or users to facilitate targeted improvements.

[0135] Specifically, in the process of forming the evaluation report, the evaluation results are first collected, including the overall evaluation score, standardized scores, and the determined quality level. The evaluation results will be organized in the following format to ensure that the information is clear and easy to understand: First, list the name and version information of the evaluated textbook; then present the overall evaluation score and its calculation basis, the standardized score and its conversion method; finally, clearly indicate the quality level of the textbook (such as "Excellent", "Good", etc.). This section may also include the specific scoring range and standard scores to explain the evaluation basis to the textbook authors or users, enabling them to have a deeper understanding of the results.

[0136] Secondly, the report provides targeted improvement suggestions, pointing out the shortcomings of the textbook and key areas for improvement based on the characteristic data and quality level obtained from the evaluation. The suggestions will include specific optimization measures, such as improving semantic depth, reorganizing the structure to enhance logic, and improving readability. By describing these improvement measures in detail, textbook writers and users can clarify the direction and goals for subsequent revisions. In addition, the evaluation report can also include successful case studies or references for writers to study and learn from, thereby improving the overall quality and usability of the textbook.

[0137] Corresponding to the above methods, such as Figure 2 As shown, this embodiment also provides a higher education textbook quality evaluation system based on a particle swarm optimization large language model, including:

[0138] The data collection unit is used to collect textual data from higher education textbooks;

[0139] The data preprocessing unit is used to preprocess the text data using the particle swarm optimization algorithm to obtain preprocessed data;

[0140] The feature extraction unit is used to construct a large language model based on particle swarm optimization, and to extract features from the preprocessed data using the large language model to obtain feature data; the feature data includes semantic features, structural features, and quality features.

[0141] The comprehensive evaluation unit is used to comprehensively evaluate the feature data through an evaluation function to obtain a comprehensive evaluation score;

[0142] The quality determination unit is used to compare the comprehensive evaluation score with a preset standard score to determine the quality level of the teaching materials.

[0143] The beneficial effects of this invention are as follows:

[0144] (1) This invention, through comprehensive data analysis of higher education textbooks, can quantify the comprehensive evaluation score of textbooks from multiple dimensions such as semantics, structure, and quality characteristics, thereby effectively improving the accuracy of textbook quality evaluation.

[0145] (2) The present invention uses the particle swarm optimization algorithm for feature selection, which can automatically identify the most representative words and features, ensuring that the selected features can maximize the educational value of the teaching materials and improve the performance and applicability of the model.

[0146] (3) By setting preset standard scores and corresponding quality levels, the present invention compares the comprehensive evaluation score with industry standards, so that the evaluation results are not only objective, but also effectively reflect the actual quality level of the teaching materials, which makes it easier for educators to make targeted improvements.

[0147] (4) In addition to the evaluation score, the present invention can also provide a specific evaluation report, including quality level, scoring basis and improvement suggestions, to help textbook writers understand the advantages and disadvantages of the current textbooks, thereby promoting the continuous improvement of textbook quality.

[0148] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0149] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model, characterized in that, include: Collect textual data from higher education textbooks; The text data is preprocessed using the particle swarm optimization algorithm to obtain preprocessed data; A large language model based on particle swarm optimization is constructed, and feature extraction is performed on the preprocessed data through the large language model to obtain feature data; The feature data includes semantic features, structural features, and quality features; The feature data are comprehensively evaluated using an evaluation function to obtain a comprehensive evaluation score; The comprehensive evaluation score is compared with the preset standard score to determine the quality level of the teaching materials.

2. The method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model according to claim 1, characterized in that, The text data includes the chapters, content summaries, examples, and exercises of the higher education textbook.

3. The method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model according to claim 1, characterized in that, The text data is preprocessed using the particle swarm optimization algorithm to obtain preprocessed data, including: The text data is cleaned to remove irrelevant information, noise, and redundant data, resulting in cleaned data. The cleaned data is segmented into words or phrases to obtain segmented data. The particle swarm optimization algorithm is used to select features from the segmented data to identify the most representative words and features. The selected words and feature terms are vectorized to generate corresponding numerical representations, thus obtaining vectorized features. The vectorized features are standardized to ensure that each vectorized feature is within the same range, resulting in preprocessed data.

4. The method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model according to claim 3, characterized in that, The particle swarm optimization algorithm is used to perform feature selection on the segmented data to identify the most representative words and features, including: According to formula F i =w×f i +α×I i +β×G i Calculate the initial fitness value for each word; where F i Let I be the initial fitness value for segmented data i, w be the weighting coefficient controlling the influence of segmentation frequency on the fitness value, and α be the weighting coefficient controlling the influence of information gain on the fitness value. i To measure the amount of information contributed by segmented data i to the classification result, β is a weighting coefficient that controls the influence of mutual information on the fitness value, and G... i The value of the mutual information of the segmented data i; Initialize the particle swarm and assign each particle a random subset of features to represent the vocabulary to be selected; By iteratively updating the positions of particles, the feature subset is adjusted using the particle swarm optimization algorithm to maximize the overall fitness function; Evaluate the real-time fitness of the current particle position; The best-performing feature subset is selected based on the real-time fitness, and this best-performing feature subset is used as the most representative vocabulary and feature terms to be selected in the end.

5. The method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model according to claim 1, characterized in that, A large language model based on particle swarm optimization is constructed, and features are extracted from the preprocessed data using the large language model to obtain feature data, including: The structure and parameters of the large language model are defined; wherein, the number of layers of the large language model is determined to be 4, namely one input layer, two hidden layers and one output layer; the number of neurons in the input layer is 500; the number of neurons in the first hidden layer is 256; the number of neurons in the second hidden layer is 128; the activation function of the hidden layer is ReLU; and the activation function of the output layer is Softmax. Design the training dataset and corresponding labels based on the preprocessed data; The parameters of the large language model are initialized using a particle swarm optimization algorithm to set the number of particles and the initialization position. Based on the training dataset and the corresponding labels, the weights of the large language model are updated through an iterative optimization algorithm, and the model performance is evaluated using a cross-validation method to obtain a well-evaluated feature extraction model. The preprocessed data is used to extract features using the feature extraction model to obtain the semantic features, the structural features, and the quality features. The semantic features are the meaning and contextual information extracted from the text. The structural features are used to describe the attributes of the text structure. The structural features include chapter division, paragraph length, and sentence complexity. The quality features are indicators used to measure the quality of the teaching materials. The quality features include readability index and content consistency score.

6. The method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model according to claim 1, characterized in that, The calculation formula for the evaluation function is as follows: S=γ·S semantic +δ·S structural +θ·S quality Where S is the comprehensive evaluation score, representing the overall assessment of the textbook quality. semantic The scoring of semantic features reflects the depth and relevance of the text content. structural The scoring of structural features reflects the organization and logic of the teaching materials. quality The quality features are scored to assess readability, error rate, or consistency metrics. γ is the weighting coefficient of semantic features, which determines the importance of semantic features in the overall score. δ is the weighting coefficient of structural features, which determines the importance of structural features in the overall score. θ is the weighting coefficient of quality features, which determines the importance of quality features in the overall score. The formula for calculating the score of the semantic feature is as follows: Where N is the number of features, TFIDFF i It is the TF-IDF value of the segmented data i; The formula for calculating the score of the structural feature is as follows: Where M is the number of structural features, L j It is the length of chapter j, L max It is the maximum chapter length, P j It refers to the number of paragraphs; The formula for calculating the score of the quality characteristic is as follows: Where R is the readability score, D is the document length, and C is the content consistency score.

7. The method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model according to claim 1, characterized in that, The comprehensive evaluation score is compared with a preset standard score to determine the quality level of the teaching materials, including: A preset standard score is determined, which includes a score range for different quality levels; The comprehensive evaluation score is standardized to obtain a standardized comprehensive score. Based on the range of the standard scores, the standardized composite score is mapped to a specific quality level.

8. The method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model according to claim 7, characterized in that, The standardized score S normalized The formula is: Where S is the comprehensive evaluation score, S min S is the lowest score in the sample. max The highest score in the sample.

9. The method for evaluating the quality of higher education textbooks based on a particle swarm optimization large language model according to claim 8, characterized in that, Based on the range of the standard scores, the standardized composite score is mapped to a specific quality level, including: If S normalized If the score is ≥0.9, the standardized composite score will be marked as "excellent". If 0.75≤S normalized If the score is less than 0.9, the standardized composite score will be marked as "good". If 0.5≤S normalized If the score is less than 0.75, the standardized composite score will be marked as "moderate". If 0.3≤S normalized If the score is less than 0.5, the standardized composite score will be marked as "passing". If S normalized If the score is less than 0.3, the standardized composite score will be marked as "failing".

10. A quality evaluation system for higher education textbooks based on a particle swarm optimization large language model, characterized in that, include: The data collection unit is used to collect textual data from higher education textbooks; The data preprocessing unit is used to preprocess the text data using the particle swarm optimization algorithm to obtain preprocessed data; The feature extraction unit is used to construct a large language model based on particle swarm optimization, and to extract features from the preprocessed data using the large language model to obtain feature data. The feature data includes semantic features, structural features, and quality features; The comprehensive evaluation unit is used to comprehensively evaluate the feature data through an evaluation function to obtain a comprehensive evaluation score; The quality determination unit is used to compare the comprehensive evaluation score with a preset standard score to determine the quality level of the teaching materials.

Images