Classification retrieval method for online course learning system

Abstract

The invention discloses a classification retrieval method for an online course learning system, which comprises the following steps of: acquiring text data of all courses from an online course platform, and carrying out data preprocessing on the collected text data; extracting course keywords from the text data after data preprocessing, and distributing corresponding course tags according to the course keywords; converting the text data of all courses and the user query text data into vectors, and calculating the text similarity; screening out candidate courses according to the course labels and the text similarity; calculating a comprehensive score of each candidate course for the screened candidate courses in combination with user behavior data; and sorting the candidate courses in a descending order according to the comprehensive score, and displaying the sorted retrieval results to the user in sequence. According to the method, efficient and accurate course retrieval and personalized recommendation are realized, and the learning experience of the user is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of online education and learning and big data analysis, and in particular to a classification retrieval method for an online course learning system. Background Art

[0002] With the rapid development of Internet technology and big data analysis, online course learning systems have become an important part of modern education. Online education platforms have brought together a huge amount of course resources, covering multiple fields such as basic education, higher education, and vocational training. In order to meet the diverse and personalized learning needs of users, how to quickly and accurately retrieve courses that meet user needs from the huge course library has become a difficult problem that needs to be solved.

[0003] Existing online course retrieval methods are mainly based on keyword matching, that is, directly matching the keywords entered by the user with the vocabulary in the course title or description. This method is simple to implement and has high computational efficiency, but it has obvious limitations and cannot effectively handle natural language phenomena such as synonyms, near-synonyms and polysemous words, resulting in low relevance and accuracy of retrieval results. In addition, with the surge in the number of courses, simple keyword matching methods are difficult to meet users' higher requirements for retrieval accuracy and personalized recommendations.

[0004] In order to solve the above problems, natural language processing and machine learning technologies are introduced into the existing technology to improve the effect of online course retrieval. However, the existing technology still has shortcomings in practical applications. First, the key features and subject information of the course may be ignored in the text vectorization process, resulting in unsatisfactory similarity calculation results; secondly, the use of user behavior data is often not sufficient, and there is a lack of comprehensive analysis of multiple behavior data, which cannot fully reflect the user's interest preferences; in addition, the automatic allocation and management of course tags also face challenges. The existing methods need to be improved in accuracy and efficiency, and it is difficult to adapt to the processing needs of large-scale course data. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract of the specification and the title of the invention of this application to avoid blurring the purpose of this section, the abstract of the specification and the title of the invention, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] In view of the above existing problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by the present invention is that the existing online course retrieval technology has the problems of insufficient retrieval accuracy, low personalization and inability to efficiently process massive course data.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions: obtaining text data of all courses from an online course platform, and performing data preprocessing on the collected text data;

[0009] Extracting course keywords from the text data after data preprocessing, and assigning corresponding course tags according to the course keywords;

[0010] Convert the text data of all courses and user query text data into vectors, and calculate their text similarity;

[0011] Screening out candidate courses according to the course label and the text similarity;

[0012] For the selected candidate courses, combined with user behavior data, a comprehensive score of each candidate course is calculated;

[0013] The candidate courses are sorted in descending order according to the comprehensive scores, and the sorted search results are displayed to the user in order.

[0014] As a preferred solution of the classification retrieval method for an online course learning system described in the present invention, the text data of all courses at least includes course titles, course introductions, course chapter contents, teaching teacher introductions and course categories.

[0015] As a preferred solution of the classification retrieval method for an online course learning system described in the present invention, the data preprocessing includes:

[0016] Using a Chinese word segmentation tool to segment the text data of all the courses to obtain word sequences;

[0017] Establishing a stop word list, the stop word list at least includes common meaningless words, and then removing the stop words from the word sequence to obtain a filtered word sequence;

[0018] Use morphological restoration techniques to restore words to their basic forms;

[0019] Unify synonyms and inflected words to get a standardized word sequence;

[0020] The standardized word sequence text is then normalized.

[0021] As a preferred solution of the classification retrieval method for the online course learning system of the present invention, the text normalization includes:

[0022] Convert all words in the standardized word sequence to lower case;

[0023] removing or replacing punctuation and special characters in the standardized word sequence;

[0024] Unify the format of numbers to obtain a standardized word sequence.

[0025] As a preferred solution of the classification retrieval method for the online course learning system of the present invention, course keywords are extracted from the standardized word sequence, and the course keyword extraction adopts the TF-IDF algorithm, including:

[0026] Calculate the word frequency TF of each word;

[0027] Calculate the inverse document frequency (IDF) of each word;

[0028] Calculate the TF-IDF value of each word;

[0029] For each course, the words ranked top k in terms of TF-IDF value are selected as the course keyword set.

[0030] As a preferred solution of the classification retrieval method for the online course learning system of the present invention, the course keywords are matched with a predefined course label set, and a corresponding course label is assigned to each course, including:

[0031] Create a predefined set of course tags;

[0032] Matching the course keyword set with the predefined course tag set;

[0033] If the course keyword is the same as or a synonym of one of the tags, assign the tag to the course corresponding to the course keyword;

[0034] If the course keyword fails to match a corresponding tag, a naive Bayes classifier is used to assign the most appropriate tag to the course based on the text features of the course.

[0035] As a preferred solution of the classification retrieval method for the online course learning system of the present invention, the text data of all courses and the user query text data are converted into vectors, and their text similarity is calculated, including:

[0036] Combine the course title, introduction, and chapter content into a complete text;

[0037] Input the complete text into the pre-trained model ERNIE, extract the output vector corresponding to the CLS tag as the vector representation of the course;

[0038] Perform data preprocessing on user query text;

[0039] Input the preprocessed query text into the pre-trained model ERNIE to obtain a query vector;

[0040] The cosine similarity is used to calculate the text similarity between the course vector and the query vector. The mathematical expression formula is as follows:

[0041]

[0042] Where n is the vector dimension, Q is the query vector, and C d is the vector representation of course d, i.e., the course vector;

[0043] For each course, calculate the text similarity Sim(Q,C d ) to get the similarity list.

[0044] As a preferred solution of the classification retrieval method for the online course learning system of the present invention, screening candidate courses according to the course label and the text similarity includes:

[0045] Extracting query keywords from the user query text data;

[0046] Matching the query keyword with a predefined course tag set to obtain a candidate tag set;

[0047] Filter courses whose course labels intersect with the candidate label set to form an initial candidate set;

[0048] Set the similarity threshold θ;

[0049] Select from the initial candidate set the candidate that satisfies Sim(Q,C d )≥θ to form the final course candidate set.

[0050] As a preferred solution of the classification retrieval method for the online course learning system described in the present invention, the comprehensive score of each candidate course is calculated for the final selected course candidate set in combination with user behavior data. The mathematical expression formula of the comprehensive score is as follows:

[0051] Score d =α×Sim(Q,C d )+β×Click d +γ×Rate d +δ×Favorite d +∈×Share d

[0052] Among them, Score d is the comprehensive score of course d, Sim(Q,C d ) is the similarity between the query vector and the course vector. Click d is the course click rate, Rated Average rating for the course, Favorite d is the number of times the user has collected the data. d is the number of times the user shares, α, β, γ, δ are all weight coefficients, satisfying α+β+γ+δ+=1;

[0053] The user behavior data at least includes click-through rate, average rating, learning completion rate, number of user collections and number of shares.

[0054] As a preferred solution of the classification retrieval method for an online course learning system described in the present invention, the course candidate set is sorted in descending order according to the comprehensive score to obtain a sorted list R = {d1, d2, ..., d h}, where h is the course ranking number;

[0055] For each retrieved course, display: course title, course description, average rating, click-through rate and course tags.

[0056] Beneficial effects of the present invention: The present invention realizes rapid and accurate screening of candidate courses that meet user needs from massive courses through processing of online course text data, keyword extraction, label assignment, text vectorization and similarity calculation; and combines user behavior data for comprehensive sorting, further enhancing the personalization and accuracy of recommendations, solving the problems of low matching accuracy and insufficient personalized recommendations in existing online course retrieval systems, and improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0058] Figure 1 It is a flow chart of the classification retrieval method for an online course learning system shown in the present invention;

[0059] Figure 2 A schematic diagram of a comparison of the retrieval accuracy of the traditional method and the method of the present invention shown in the present invention;

[0060] Figure 3 A line diagram showing a comparison of user satisfaction between the traditional method shown in the present invention and the method of the present invention. DETAILED DESCRIPTION

[0061] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments.

[0062] Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0063] In the following description, many specific details are set forth to facilitate a thorough understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0064] According to an embodiment of the present invention, in combination with Figure 1 the flowchart shown, a classification and retrieval method for an online course learning system specifically includes the following steps:

[0065] S1. Obtain the text data of all courses from an online course platform and perform data preprocessing on the collected text data. Here, it should be noted that the data preprocessing includes:

[0066] Use a Chinese word segmentation tool to segment the text data of all courses to obtain a word sequence W = {w1, w2,..., w i}, where w i is the i-th word after word segmentation;

[0067] Establish a stop word list that at least contains common meaningless words (such as "de", "le", "he"), and then remove the stop words from the word sequence to obtain a filtered word sequence W ′ ;

[0068] Utilize word stemming technology to reduce words to their basic forms to improve the unity of synonyms and near-synonyms;

[0069] Unify synonyms and words with morphological variations, such as unifying "study", "have studied", "be studying" into "study" to obtain a standardized word sequence W ″ ;

[0070] Then normalize the text of the standardized word sequence.

[0071] Furthermore, text normalization includes:

[0072] Convert all words in the standardized word sequence to lowercase;

[0073] Remove or replace punctuation marks and special characters in the standardized word sequence;

[0074] Unify the format of numbers, for example, convert "one" and "1" to "1" to obtain the standardized word sequence W″′.

[0075] As an example, the text data of all courses contains at least the course title, course introduction, course chapter content, instructor introduction, and course category.

[0076] As an example, course title: such as "Introduction to Python Data Analysis"; course introduction: a brief description of the course content; course chapter content: the detailed text content of each chapter; course category: such as "Programming Language", "Data Analysis"; instructor introduction: the background information of the instructor.

[0077] It should be noted that data preprocessing uses a word segmentation tool (such as a tokenizer) to split the original text into independent words, while removing meaningless stop words and unifying the word form, making the text more structured and standardized. This process realizes the preliminary cleaning and optimization of course data, ensuring that subsequent processing steps are based on data input with good consistency.

[0078] Preferably, the processing of this step significantly reduces the noise data in the original text, such as irrelevant words, punctuation marks, and synonym ambiguities, thereby improving the accuracy of subsequent keyword extraction and similarity calculation.

[0079] S2. Extract course keywords from the text data after data preprocessing and assign corresponding course tags according to the course keywords. Among them, it should be noted in this step that:

[0080] Extract course keywords from the standardized word sequence. The course keyword extraction uses the TF-IDF algorithm, including:

[0081] Calculate the term frequency TF of each word, and its mathematical expression formula is as follows:

[0082]

[0083] Among them, f t,d is the number of times the word t appears in the document d, and f t′,d is the total number of times all words appear in the document d;

[0084] Calculate the inverse document frequency IDF of each word, and its mathematical expression formula is as follows:

[0085]

[0086] Among them, N is the total number of courses, and n t is the number of courses containing the word t;

[0087] Calculate the TF-IDF value of each word. The mathematical expression formula is as follows:

[0088] TF-IDF(t,d)=TF(t,d)×IDF(t)

[0089] For each course, select the top k words with the highest TF-IDF value as the course keyword set K d .

[0090] As an example, the value range of k is [5,20], which is determined according to the length of the course text and actual needs.

[0091] As an example, the TextRank algorithm can be combined with a word graph model to extract important keywords using iterative calculations of the graph.

[0092] Furthermore, the course keywords are matched with a predefined set of course tags, and corresponding course tags are assigned to each course, including:

[0093] Establish a predefined course label set L = {l1,l2,...,l m}, where l m is the mth label;

[0094] Match the course keyword set with the predefined course tag set;

[0095] If the course keyword is the same as one of the tags or is a synonym of it, then the tag is assigned to the course corresponding to the course keyword;

[0096] If the course keywords fail to match the corresponding tags, the naive Bayes classifier is used to assign the most appropriate tags based on the text features of the course.

[0097] It should be noted that the TF-IDF algorithm identifies representative words in the course text by calculating word frequency and inverse document frequency, and then constructs the subject tag of the course. It achieves semantic understanding of the course content through keyword extraction, and combines the allocation of course tags to give the course data a semantic information hierarchy.

[0098] Preferably, this step enables users to search for courses based on specific topics or fields through accurate label assignment, thereby improving the classification and recommendation efficiency of the system.

[0099] S3, converting all course text data and user query text data into vectors, and calculating their text similarity. Among them, what needs to be explained in this step is:

[0100] Convert all course text data and user query text data into vectors and calculate their text similarity, including:

[0101] Combine the course title, introduction, and chapter content into a complete text d ;

[0102] Input the complete text into the pre-trained model ERNIE (to improve the ability to capture fine-grained semantics), extract the output vector corresponding to the CLS tag, and use it as the vector representation of the course C d ;

[0103] Perform the same data preprocessing process as step S1 above on the user query text;

[0104] Input the preprocessed query text into the pre-trained model ERNIE to obtain the query vector Q;

[0105] The cosine similarity is used to calculate the text similarity between the course vector and the query vector. The mathematical expression formula is as follows:

[0106]

[0107] Where n is the vector dimension, Q is the query vector, and C d is the vector representation of course d, i.e., the course vector;

[0108] For each course, calculate the text similarity Sim(Q,C d ) to get the similarity list.

[0109] As an example, when Sim(Q,C d ) is closer to 1, the more similar the texts are; otherwise, the less similar they are. For example, if Sim(Q,C d )≥θ(θ is 0.5), then it is considered that the similarity between course d and query Q is high.

[0110] As an example, the similarity list S = {(d1,Sim1),(d2,Sim2),...,(d N ,Sim N )},in, Then, based on the degree of similarity, the courses can be sorted and filtered to improve retrieval accuracy.

[0111] It should be noted that this step uses a pre-trained model (such as ERNIE) to convert course text and user query text into vector representations, and uses cosine similarity to calculate the text similarity between courses and queries. This can effectively capture the deep semantic structure of the text and avoid the limitations of traditional keyword matching. In particular, by representing the text as a high-dimensional vector, it takes into account contextual information and semantic associations between words, ensuring more accurate similarity calculations.

[0112] Preferably, this step greatly improves the accuracy and relevance of the search results by introducing a vectorized semantic analysis model, and solves the problem of matching synonyms and polysemous words.

[0113] S4. Filter candidate courses based on course labels and text similarity. The following points should be noted in this step:

[0114] Extract query keywords K from user query text data q ;

[0115] Match the query keywords with the predefined course tag set L to obtain the candidate tag set L q ;

[0116] Filter courses whose course labels intersect with the candidate label set to form an initial candidate set S;

[0117] Set the similarity threshold θ (e.g. θ is 0.5);

[0118] Select from the initial candidate set S the candidate that satisfies Sim(Q,C d )≥θ to form the final course candidate set S′.

[0119] In an optional embodiment, when the number of candidate courses is insufficient, the similarity threshold θ may be gradually lowered or the scope of tag matching may be expanded, including considering related tags and synonymous tags.

[0120] It should be noted that this embodiment extracts query keywords based on user queries and matches them with a set of course tags to screen out a preliminary set of candidate courses. On this basis, courses are further filtered based on a similarity threshold to form a final set of candidate courses. This step not only relies on the calculation of text similarity, but also combines the screening mechanism of course tags to ensure that the candidate courses are highly consistent with user needs in terms of subject and semantics.

[0121] Preferably, through the processing of this step, the candidate range is effectively narrowed, the retrieval efficiency is improved, and too many irrelevant courses are avoided from interfering with the user experience. By combining the dual screening mechanism of tags and similarities, accurately matched course results are provided more efficiently.

[0122] S5. Calculate the comprehensive score of each candidate course based on the selected candidate courses and user behavior data.

[0123] For the final selected course candidate set, combined with user behavior data, the comprehensive score of each candidate course is calculated. The mathematical expression formula of the comprehensive score is as follows:

[0124] Score d =α×Sim(Q,Cd )+β×Click d +γ×Rate d +δ×Favorite d +∈×Share d

[0125] Among them, Score d is the comprehensive score of course d, Sim(Q,C d ) is the similarity between the query vector and the course vector. Click d is the course click rate, Rate d Average rating for the course, Favorite d is the number of times the user has collected the collection, share d is the number of times the user shares, α, β, γ, δ are all weight coefficients, satisfying α+β+γ+δ+=1;

[0126] As an example, user behavior data includes at least click-through rate, average rating, learning completion rate, number of user collections, and number of shares.

[0127] As an example, the mathematical expression of the course click-through rate is as follows:

[0128]

[0129] Among them, N click,d is the number of clicks on course d, N click,total is the total number of clicks for all courses;

[0130] As an example, the mathematical expression for the average course rating is as follows:

[0131]

[0132] Among them, r i is the user's rating of course d, n d is the number of scorers;

[0133] As an example, the mathematical expression of learning completion rate is as follows:

[0134]

[0135] It should be noted that this embodiment, by introducing multi-dimensional user behavior data, can more comprehensively evaluate the popularity and actual learning effect of each course, comprehensively consider similarity and user interaction behavior, and ensure that the recommended courses not only meet the semantic relevance, but also meet the user's real preferences and needs, thereby enhancing the personalized recommendation capabilities and improving the system's understanding of user interests.

[0136] S6. Sort the candidate courses in descending order according to the comprehensive scores, and display the sorted search results to the user in order.

[0137] As an example, the course candidate set is sorted in descending order according to the comprehensive score, and the sorted list R = {d1, d2, ..., d h}, where h is the course ranking number.

[0138] As an example, for each course retrieved, show:

[0139] Course title, course description, average rating, click-through rate, and course tags.

[0140] It should be noted that the ranking algorithm used in this embodiment is used to assign weights based on the semantic similarity of the courses and the comprehensive score of user behavior, ensuring that high-quality courses are displayed to users first. This process not only improves the user search experience, but also effectively promotes the exposure of high-quality content.

[0141] This sorting method ensures that the courses users see are not only highly relevant to their queries, but also reflect the choices and evaluations of other users to a certain extent. This helps users quickly find high-quality courses that meet their needs, improving the usability of search results and user satisfaction.

[0142] The aforementioned keyword matching using the naive Bayes classifier can be performed using methods and means in the prior art, which will not be described in detail in this example.

[0143] Furthermore, in order to verify the advantages of a classification retrieval method for an online course learning system provided by an embodiment of the present invention in terms of retrieval accuracy and user satisfaction, a comparative experiment was conducted between the method of the present invention and the prior art method. The experiment aims to evaluate the degree of improvement of the method of the present invention in retrieval effect and user experience by simulating user search behavior, and to prove that it has significant improvements over the prior art.

[0144] Existing technical methods (method A): using traditional keyword matching retrieval, directly matching user queries with keywords in course titles or descriptions, lacking in-depth semantic understanding and personalized recommendations;

[0145] The method of the present invention (method B) improves the retrieval accuracy and personalization through course label assignment, text vectorization representation, similarity calculation and comprehensive sorting combined with user behavior data.

[0146]

Hardware Configuration

[0147] Processor (CPU): Intel Core i7-9700K;

[0148] Memory (RAM): 16GB DDR4;

[0149] Storage: 512GB SSD;

[0150]

Software Configuration

[0151] Operating system: Windows 10;

[0152] Programming language: Python 3.8;

[0153]

Main libraries and frameworks

[0154] Natural language processing: NLTK, jieba;

[0155] Machine learning: scikit-learn;

[0156] Data processing and visualization: pandas, matplotlib;

[0157] Course Dataset: Collect information on 500 courses from online education platforms, covering the fields of computer science, language, and art. Each course includes text data of title, introduction, and category.

[0158] User query samples: simulate 10 common user queries, such as "Python programming basics" and "English speaking improvement".

[0159] [Operation framework of prior art method (method A)]

[0160] Keyword extraction: Segment user queries and course titles to extract keywords;

[0161] Retrieval process: Match the user's query keywords with the keywords in the course title, and the courses with high matching degree will be used as the retrieval results;

[0162] Result sorting: Sort by the number of matching keywords, with courses with more matching words at the front.

[0163] [Operation framework of the method of the present invention (method B)]

[0164] Data preprocessing: segment course texts and user queries, remove stop words, restore lemmas, and unify synonyms;

[0165] Keyword extraction and course label assignment: Use the TF-IDF algorithm to extract course keywords, match course keywords with a predefined set of course labels, and assign labels to courses;

[0166] Text vectorization and similarity calculation: Convert course text and user queries into vector representations and calculate cosine similarity;

[0167] Comprehensive ranking based on user behavior data: Introduce simulated user behavior data (click rate, rating) and calculate the comprehensive score;

[0168] Result sorting and display: Sort by comprehensive score and display to users.

[0169]

Evaluation indicators

[0170] Retrieval accuracy: the proportion of relevant courses in the retrieval results;

[0171] User satisfaction rating: simulates the user's satisfaction with the search results, with a rating range of 1 to 5.

[0172] In order to more intuitively display the experimental results, this embodiment provides comparative data of retrieval accuracy and user satisfaction score through MATLB simulation software, and presents it in the form of a line graph, such as Figure 2 and Figure 3 shown.

[0173] Preferably, from Figure 2 It can be seen that the accuracy of the method of the present invention (method B) is significantly higher than that of the prior art (method A) in all queries, with an average increase of about 25%; the curve of method B is above the curve of method A as a whole, with less fluctuation, showing a more stable high accuracy. Method B has achieved an accuracy of more than 80% in all queries, with the highest reaching 90%; while the accuracy of method A is generally between 50% and 65%, with large fluctuations.

[0174] Preferably, from Figure 3 It can be seen that the user satisfaction scores of the method of the present invention (method B) are all above 4.2 points, with an average of 4.5 points, while the user satisfaction score of the prior art (method A) is lower, with an average of about 2.9 points; therefore, the classification retrieval method for an online course learning system provided by an embodiment of the present invention does achieve efficient and accurate course retrieval and personalized recommendation, and improves the user's learning experience.

[0175] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A classification retrieval method for an online course learning system, characterized in that: include: Acquire text data of all courses from the online course platform, and perform data preprocessing on the collected text data; Extracting course keywords from the text data after data preprocessing, and assigning corresponding course tags according to the course keywords; Convert the text data of all courses and user query text data into vectors, and calculate their text similarity; Screening out candidate courses according to the course label and the text similarity; For the selected candidate courses, combined with user behavior data, a comprehensive score of each candidate course is calculated; The candidate courses are sorted in descending order according to the comprehensive scores, and the sorted search results are displayed to the user in order.

2. The classification retrieval method for an online course learning system according to claim 1, characterized in that: The text data of all courses at least includes course title, course introduction, course chapter content, teaching teacher introduction and course category.

3. The classification retrieval method for an online course learning system according to claim 1 or 2, characterized in that: The data preprocessing includes: Using a Chinese word segmentation tool to segment the text data of all the courses to obtain word sequences; Establishing a stop word list, the stop word list at least includes common meaningless words, and then removing the stop words from the word sequence to obtain a filtered word sequence; Use morphological restoration techniques to restore words to their basic forms; Unify synonyms and inflected words to get a standardized word sequence; The standardized word sequence text is then normalized.

4. The classification retrieval method for an online course learning system according to claim 3, characterized in that: The text normalization includes: Convert all words in the standardized word sequence to lower case; removing or replacing punctuation and special characters in the standardized word sequence; Unify the format of numbers to obtain a standardized word sequence.

5. The classification retrieval method for an online course learning system according to claim 4, characterized in that: Extracting course keywords from the standardized word sequence, wherein the course keyword extraction adopts the TF-IDF algorithm, includes: Calculate the word frequency TF of each word; Calculate the inverse document frequency (IDF) of each word; Calculate the TF-IDF value of each word; For each course, the words ranked top k in terms of TF-IDF value are selected as the course keyword set.

6. The classification retrieval method for an online course learning system according to claim 5, characterized in that: Match the course keywords with a predefined set of course tags, and assign corresponding course tags to each course, including: Create a predefined set of course tags; Matching the course keyword set with the predefined course tag set; If the course keyword is the same as or a synonym of one of the tags, assign the tag to the course corresponding to the course keyword; If the course keyword fails to match a corresponding tag, a naive Bayes classifier is used to assign the most appropriate tag to the course based on the text features of the course.

7. The classification retrieval method for an online course learning system according to claim 1, characterized in that: The text data of all courses and user query text data are converted into vectors, and their text similarity is calculated, including: Combine the course title, introduction, and chapter content into a complete text; Input the complete text into the pre-trained model ERNIE, extract the output vector corresponding to the CLS tag as the vector representation of the course; Perform data preprocessing on user query text; Input the preprocessed query text into the pre-trained model ERNIE to obtain a query vector; The cosine similarity is used to calculate the text similarity between the course vector and the query vector. The mathematical expression formula is as follows: Where n is the vector dimension, Q is the query vector, and C d is the vector representation of course d, i.e., the course vector; For each course, calculate the text similarity Sim(Q,C d ) to get the similarity list.

8. The classification retrieval method for an online course learning system according to claim 1, characterized in that: According to the course label and the text similarity, candidate courses are screened out, including: Extracting query keywords from the user query text data; Matching the query keyword with a predefined course tag set to obtain a candidate tag set; Filter courses whose course labels intersect with the candidate label set to form an initial candidate set; Set the similarity threshold θ; Select from the initial candidate set the candidate that satisfies Sim(Q,C d )≥θ to form the final course candidate set.

9. The classification retrieval method for an online course learning system according to claim 8, characterized in that: For the final selected course candidate set, combined with user behavior data, the comprehensive score of each candidate course is calculated. The mathematical expression formula of the comprehensive score is as follows: Score d =α×Sim(Q,C d )+β×Click d +γ×Rate d +δ×Favorite d +∈×Share d Among them, Score d is the comprehensive score of course d, Sim(Q,C d ) is the similarity between the query vector and the course vector. Click d is the course click rate, Rate d Average rating for the course, Favorite d is the number of times the user has collected the collection, share d is the number of times the user shares, α, β, γ, δ are all weight coefficients, satisfying α+β+γ+δ+=1; The user behavior data at least includes click-through rate, average rating, learning completion rate, number of collections and sharing times of the user.

10. The classification retrieval method for an online course learning system according to claim 9, characterized in that: According to the comprehensive scores, the course candidate set is sorted in descending order to obtain a sorted list R = {d1, d2, ..., d h }, where h is the course ranking number; For each retrieved course, display: course title, course description, average rating, click-through rate and course tags.

Images