Intelligent reading recommendation system and method

Abstract

The invention discloses an intelligent reading recommendation system and method, belongs to the technical field of novel recommendation, and solves the problem of novel recommendation in advance under the condition that a user does not read a novel. Comprising a data collection module, a user behavior collection module and a recommendation module, the data collection module is used for collecting foreground and background activity duration of software, the user behavior collection module is used for collecting a user browsing software operation interface picture and a time axis, and the recommendation module updates and pushes a user novel according to feedback information. According to the method, the existing novel pushing method can be simplified, the data volume required to be acquired by novel software is reduced, on one hand, the information acquisition efficiency can be improved, and on the other hand, the operation difficulty and the burden of a processor can be reduced in the actual recommendation operation process; therefore, novel recommendation can be carried out in advance under the condition that the user does not carry out novel reading.

Description

Technical Field

[0001] The present invention relates to the technical field of novel recommendation, and in particular to an intelligent reading recommendation system and method. Background Art

[0002] Fiction reading is a beloved literary experience that allows readers to explore different worlds and lives through stories and characters carefully woven by the author. Fiction often attracts readers with rich plots, fascinating characters and profound themes, stimulating their imagination and emotional resonance. From classic literature to modern bestsellers, novels cover a variety of genres such as science fiction, fantasy, suspense, romance, history and realism, etc., to meet the tastes and interests of different readers.

[0003] Reading novels is not only a form of entertainment, but also a means of education and self-improvement. It can improve language skills, expand vocabulary, and enhance empathy and cultural understanding. In addition, novels can provide deep insights into complex social issues and trigger readers' thinking about human nature, morality and philosophical issues. Whether through paper books, e-books or audiobooks, novel reading is a convenient and flexible cultural activity that can be carried out at any time and any place, making reading a kind of enjoyment and spiritual journey.

[0004] At present, it is a very common behavior for people to read novels through mobile phone software. In order to better meet the needs of users, mobile phone software has also launched various algorithms to meet the preferences of users. For example, the application number is CN202410587151.5, which is a novel recommendation method and system based on the daily browsing habits of users. Its technical points are as follows: S1, data collection: first collect all browsing data of users on the novel platform, which includes the number of clicks, reading time, collection type, member purchase, etc. of various novels; S2, user behavior analysis: analyze the user behavior data collected by S1, including analyzing the reading habits, preferences, interests, etc. of users. Information; S3, user data analysis: analyze the user behavior analysis of S2 and then analyze the user data. The user data includes information about the user's age, gender, reading habits, etc. This information can help the recommendation system understand the user's needs and interests more accurately; S4, design recommendation algorithm: design a suitable recommendation algorithm based on S2 and S3, and the recommendation algorithm can be implemented using a content-based recommendation algorithm; S5, implement recommendation: after the recommendation algorithm is designed in S4, novels can be recommended to users. The recommendation system can recommend suitable novels to users based on user behavior analysis and data analysis; S6, feedback and optimization: establish a user feedback mechanism so that users can provide feedback on the recommendation results.

[0005] The above scheme collects information such as the number of clicks and reading time of users on novels on the novel platform, and analyzes user behavior, user information and novel content, so as to recommend novels to users based on user preferences, interests, retrieved topics, styles, etc., effectively improving the speed of novel recommendations.

[0006] It is now necessary to simplify the above method so as to be able to recommend novels in advance when the user has not read the novels.

[0007] Therefore, an intelligent reading recommendation system and method are proposed to solve or alleviate the above problems. Summary of the invention

[0008] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an intelligent reading recommendation system and method.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] An intelligent reading recommendation system includes a data collection module, a user behavior collection module, and a recommendation module. The data collection module is used to collect the activity duration of the foreground and background of the software, the user behavior collection module is used to collect the user's browsing of the software operation interface screen and timeline, and the recommendation module updates and pushes user novels according to the feedback information.

[0011] The present invention also provides an intelligent reading recommendation method, based on the above-mentioned intelligent reading recommendation system, comprising the following steps:

[0012] Intercept the software foreground activity duration and obtain the foreground operation timeline;

[0013] Intercept the operation timeline of the operation interface according to the foreground operation timeline;

[0014] Extract the frame images on the operation time axis of the operation interface through a sliding window algorithm to form a number of operation image frames;

[0015] Preprocessing several operation picture frames;

[0016] Perform text area detection and text recognition on the pre-processed operation screen frame, and finally store the recognition results;

[0017] The Levenshtein distance similarity of the texts in the adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface is calculated to obtain the text similarity;

[0018] The adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface are compared using a feature matching algorithm to obtain image similarity;

[0019] Arrange the text similarity and the image similarity on the operation timeline of the operation interface to obtain a similarity timeline;

[0020] The similar text similarity and image similarity are intercepted on the similarity time axis and classified into several groups of change similarity time period axes;

[0021] Extract the change similarity time period axis with a duration greater than the time threshold, and retain the frame picture in the middle of the group to form several observation pictures;

[0022] Extract keywords from several observation images through a character recognition algorithm and sort them according to the number of times the keywords appear;

[0023] The selection model is imported into the top five keywords and the top 50 novel contents are output in descending order of relevance.

[0024] Preferably, the preprocessing of the plurality of operation picture frames comprises the following steps:

[0025] Convert several operation screen frames into grayscale images;

[0026] Apply Gaussian blur to reduce image noise;

[0027] Use an edge detection algorithm to highlight areas that may contain the text.

[0028] Preferably, the performing text area detection and text recognition on the preprocessed operation picture frame comprises the following steps:

[0029] Use contour detection algorithm to identify contours in the operation picture frame;

[0030] The OCR algorithm is used to identify the text area of ​​the contour in the extraction recognition operation screen frame.

[0031] Preferably, the Levenshtein distance similarity calculation of texts in adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface includes:

[0032] The Levenshtein distance formula is used to calculate the distance between texts in adjacent operation screen frames. The Levenshtein distance formula includes

[0033] , where D(i,j) is the Levenshtein distance between the first i characters of string S and the first j characters of string T, S i is the i-th character of string S, T j is the jth character of string T, cost(S i ,T j) is the replacement cost, if S i Equal to T j If yes, it is 0, otherwise it is 1;

[0034] The similarity is calculated by Levenshtein distance. Where D(m,n) is the Levenshtein distance between the texts in the two operation screen frames, and m and n are the lengths of the texts in the two operation screen frames, respectively.

[0035] Preferably, the step of comparing images of adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface using a feature matching algorithm to obtain image similarity comprises the following steps:

[0036] Detect feature points in each operation screen frame;

[0037] Generate descriptors for detected feature points through feature point detection algorithm;

[0038] Use Euclidean distance to calculate the distance between descriptors and match the feature points between consecutive operation screen frames;

[0039] Calculate the similarity between adjacent operation screen frames, Where k is the number of matching point pairs, d(a i ,b i ) is the distance between the i-th pair of matching points.

[0040] Preferably, the picture extracts keywords through a character recognition algorithm and sorts them according to the number of times the keywords appear, including the following steps:

[0041] Recognize and extract all text from the picture through OCR algorithm;

[0042] Clean and standardize data from extracted text, remove stop words, segment words, and reduce vocabulary to its base form;

[0043] Extract keywords from the processed text and calculate the number of times each word appears in the text using TF-IDF;

[0044] Sort keywords according to TF-IDF value.

[0045] Preferably, the method for establishing the selection model comprises the following steps:

[0046] Collect the reviews of each novel and perform pre-processing such as word segmentation and stop word removal to form a novel review database;

[0047] Collect the category keywords of each novel;

[0048] Convert category keywords and comments into word vectors;

[0049] Import word vectors into the CNN model to determine the correlation between comments and keywords and output the selection model.

[0050] The present invention has the following beneficial effects:

[0051] The present invention significantly reduces the amount of data that needs to be collected when the novel software performs personalized recommendations by simplifying the existing push mechanism. This improvement not only improves the efficiency of information collection, making the data collection process faster and more convenient, but also greatly reduces the complexity of operations when the recommendation algorithm is actually executed, reduces the workload of the processor, and enables the system to run more smoothly. Therefore, before the user starts reading the novel, accurate novel recommendations can be made in advance through the user's browsing and collection in the operation interface. This advance recommendation function enables the user to access novels that may be of interest to the user more quickly, thereby improving the user experience, and also brings higher user satisfaction and retention rate to the operator of the novel software. In addition, reducing the amount of data collection also helps to protect the privacy of the user, because the less data collected, the lower the potential risk of privacy leakage. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0053] Figure 1 This is a flowchart of the intelligent reading recommendation method in the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0055] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0057] In the description of the present invention, it should be understood that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, or are the orientations or positional relationships in which the product of the invention is conventionally placed when in use, or are the orientations or positional relationships conventionally understood by those skilled in the art. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0058] Furthermore, the terms “first”, “second”, “third”, etc. are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0059] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0060] An intelligent reading recommendation system includes a data collection module, a user behavior collection module, and a recommendation module. The data collection module is used to collect the activity duration of the foreground and background of the software. The user behavior collection module is used to collect the user's browsing of the software operation interface screen and timeline. The recommendation module updates and pushes user novels based on the feedback information.

[0061] The intelligent reading recommendation system of the present invention significantly improves the relevance and timeliness of novel recommendations through partial data collection and processing, and the system not only speeds up the efficiency of information collection by reducing the amount of data required to be collected, but also reduces the complexity of recommendation operations and reduces the burden on the processor. This enables the system to predict the reading preferences of users based on their behavior before they start reading, and push related novels in advance, thereby enhancing the user experience, improving user satisfaction and the user stickiness of the software.

[0062] An intelligent reading recommendation method, such as Figure 1 As shown, the following steps are included:

[0063] Intercept the software foreground activity duration and obtain the foreground operation timeline;

[0064] Intercept the operation timeline of the operation interface according to the foreground operation timeline;

[0065] Extract the frame images on the operation time axis of the operation interface through a sliding window algorithm to form a number of operation image frames;

[0066] Preprocessing several operation picture frames;

[0067] Perform text area detection and text recognition on the pre-processed operation screen frame, and finally store the recognition results;

[0068] The Levenshtein distance similarity of the texts in the adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface is calculated to obtain the text similarity;

[0069] The adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface are compared using a feature matching algorithm to obtain image similarity;

[0070] Arrange the text similarity and the image similarity on the operation timeline of the operation interface to obtain a similarity timeline;

[0071] The similar text similarity and image similarity are intercepted on the similarity time axis and classified into several groups of change similarity time period axes;

[0072] Extract the change similarity time period axis with a duration greater than the time threshold, and retain the frame picture in the middle of the group to form several observation pictures;

[0073] Extract keywords from several observation images through a character recognition algorithm and sort them according to the number of times the keywords appear;

[0074] The selection model is imported into the top five keywords and the top 50 novel contents are output in descending order of relevance.

[0075] This method uses time series analysis to capture the temporal pattern of user behavior, while image processing techniques (such as Gaussian blur and edge detection) are used to improve image quality and highlight text areas. Text recognition techniques (such as OCR) are used to extract text information from images. Finally, by calculating the similarity between text and image, combined with keyword extraction and sorting, the system can understand the user's interests and recommend related novels. By capturing the foreground activity duration and the operation timeline of the operation interface, the system can capture the user's detailed behavior, and use the sliding window algorithm to extract the operation screen frame. Through image processing and text recognition technology, the system can identify and analyze the text information on the user interface. By calculating text similarity and image similarity, the system can identify the user's interest points, and then recommend related novels based on the correlation of keywords, which greatly improves the performance of the recommendation system and user satisfaction.

[0076] Preferably, preprocessing a plurality of operation picture frames comprises the following steps:

[0077] Convert several operation screen frames into grayscale images;

[0078] Apply Gaussian blur to reduce image noise;

[0079] Use an edge detection algorithm to highlight areas that may contain the text.

[0080] By preprocessing the operating screen frames, the accuracy of text area detection and recognition is significantly improved, thereby providing higher quality input data for the subsequent recommendation algorithm. Converting to grayscale, applying Gaussian blur and using edge detection algorithms work together to enhance the features of text in the image, reduce noise interference, and make the text area more prominent, which is convenient for subsequent text recognition and analysis.

[0081] The formula for converting several operation screen frames into grayscale images includes I gray (x,y)=0.299×I red (x,y)+0.587×I green (x,y)+0.114×I blue (x,y), where I gray (x,y) is the grayscale value at position (x,y), I red (x,y),I green (x,y),I blue (x, y) are the values ​​of the red, green, and blue channels at that location, respectively.

[0082] By converting the operation screen frame into a grayscale image, the present invention simplifies the image data processing process and improves the efficiency and accuracy of image processing. This step is a key link in image preprocessing, providing high-quality image data for subsequent image analysis and text recognition, thereby ensuring the efficient operation and accurate recommendation of the entire recommendation system.

[0083] Preferably, the formula for applying Gaussian blur to reduce image noise includes Among them, σ is the standard deviation, which controls the degree of blur, and I(xi,yj) is the pixel value of the original image at position (xi,yj).

[0084] By applying Gaussian blur technology, the present invention effectively reduces image noise and improves image quality, which is crucial for subsequent text area detection and text recognition. By smoothing the image, Gaussian blur reduces random noise and details in the image, making the main features in the image more prominent, thereby improving the accuracy of text recognition and the effectiveness of the recommendation system.

[0085] Preferably, a formula using an edge detection algorithm to highlight areas that may contain the text includes I x =G x *I smooth , I y =G y *I smooth ,

[0086] By using edge detection algorithms to highlight areas that may contain text, the present invention significantly improves the accuracy of text area detection and provides precise positioning for text recognition. Edge detection is a key step in image processing. By identifying edge information in an image, it is possible to effectively distinguish text areas from non-text areas, thereby improving the efficiency and accuracy of subsequent text recognition.

[0087] Preferably, performing text area detection and text recognition on the preprocessed operation picture frame comprises the following steps:

[0088] Use contour detection algorithm to identify contours in the operation picture frame;

[0089] The OCR algorithm is used to identify the text area of ​​the contour in the extraction recognition operation screen frame.

[0090] By performing text area detection and text recognition on the pre-processed operation screen frames, the present invention achieves an in-depth understanding of the user interface operation content. This step is the key to achieving accurate recommendations. By accurately identifying the text information on the user interface, the system can analyze the user's behavior patterns and preferences, and then provide personalized novel recommendations.

[0091] Preferably, the Levenshtein distance similarity calculation of the texts in adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface includes:

[0092] The Levenshtein distance formula is used to calculate the distance between texts in adjacent operation screen frames. The Levenshtein distance formula includes , where D(i,j) is the Levenshtein distance between the first i characters of string S and the first j characters of string T, S i is the i-th character of string S, T j is the jth character of string T, cost(S i ,T j ) is the replacement cost, if S i Equal to T j If yes, it is 0, otherwise it is 1;

[0093] The similarity is calculated by Levenshtein distance. Where D(m,n) is the Levenshtein distance between the texts in the two operation screen frames, and m and n are the lengths of the texts in the two operation screen frames, respectively.

[0094] By calculating the Levenshtein distance similarity of texts in adjacent operation screen frames on the operation timeline of the operation interface, the present invention can accurately evaluate the similarity of text content. The application of this technology enables the system to identify the user's operation behavior and reading preferences at different time points, thereby providing more accurate and personalized novel recommendations.

[0095] Preferably, using a feature matching algorithm to compare images of adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface to obtain image similarity comprises the following steps:

[0096] Detect feature points in each operation screen frame;

[0097] Generate descriptors for detected feature points through feature point detection algorithm;

[0098] Use Euclidean distance to calculate the distance between descriptors and match the feature points between consecutive operation screen frames;

[0099] Calculate the similarity between adjacent operation screen frames, Where k is the number of matching point pairs, d(a i ,b i ) is the distance between the i-th pair of matching points.

[0100] By using a feature matching algorithm to compare images of adjacent operation screen frames on the operation timeline of the operation interface, the present invention achieves accurate capture and analysis of user behavior. The application of this technology enables the system to identify subtle changes in user interface operations, thereby providing more accurate and personalized novel recommendations.

[0101] Preferably, the picture extracts keywords through a character recognition algorithm and sorts them according to the number of times the keywords appear, including the following steps:

[0102] Recognize and extract all text from the picture through OCR algorithm;

[0103] Clean and standardize data from extracted text, remove stop words, segment words, and reduce vocabulary to its base form;

[0104] Extract keywords from the processed text and calculate the number of times each word appears in the text using TF-IDF;

[0105] Sort keywords according to TF-IDF value.

[0106] By extracting keywords through character recognition algorithms and sorting them according to the number of times the keywords appear, the present invention achieves an in-depth understanding and analysis of user reading preferences. The application of this technology enables the system to identify and sort the keywords that users are most interested in, thereby providing more accurate and personalized novel recommendations.

[0107] Preferably, the method for establishing the selection model comprises the following steps:

[0108] Collect the reviews of each novel and perform pre-processing such as word segmentation and stop word removal to form a novel review database;

[0109] Collect the category keywords of each novel;

[0110] Convert category keywords and comments into word vectors;

[0111] Import word vectors into the CNN model to determine the correlation between comments and keywords and output the selection model.

[0112] The CNN model includes an input layer that receives the preprocessed novel review word vectors, an embedding layer that encodes the input discrete words into continuous word vectors of fixed dimension, a convolutional layer that extracts local features, a pooling layer that reduces the feature dimension and extracts the most important features, a fully connected layer that maps the features to the output space, and an output layer that outputs the correlation score with the keyword.

[0113] By establishing a selection model, the present invention realizes an accurate evaluation of the correlation between novel reviews and category keywords. The application of this technology enables the system to recommend related novels based on the review content, thereby improving the relevance of the recommendation and user satisfaction.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent reading recommendation system, characterized in that: It includes a data collection module, a user behavior collection module, and a recommendation module. The data collection module is used to collect the duration of foreground and background activities of the software. The user behavior collection module is used to collect the user's browsing of the software operation interface screen and timeline. The recommendation module updates and pushes user novels based on the feedback information.

2. An intelligent reading recommendation method, based on the intelligent reading recommendation system according to claim 1, characterized in that: The steps include: Intercept the software foreground activity duration and obtain the foreground operation timeline; Intercept the operation timeline of the operation interface according to the foreground operation timeline; Extract the frame images on the operation time axis of the operation interface through a sliding window algorithm to form a number of operation image frames; Preprocessing several operation picture frames; Perform text area detection and text recognition on the pre-processed operation screen frame, and finally store the recognition results; The Levenshtein distance similarity of the texts in the adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface is calculated to obtain the text similarity; The adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface are compared using a feature matching algorithm to obtain image similarity; Arrange the text similarity and the image similarity on the operation timeline of the operation interface to obtain a similarity timeline; The similar text similarity and image similarity are intercepted on the similarity time axis and classified into several groups of change similarity time period axes; Extract the change similarity time period axis with a duration greater than the time threshold, and retain the frame picture in the middle of the group to form several observation pictures; Extract keywords from several observation images through a character recognition algorithm and sort them according to the number of times the keywords appear; The selection model is imported into the top five keywords and the top 50 novel contents are output in descending order of relevance.

3. The intelligent reading recommendation method according to claim 2, characterized in that: The preprocessing of the plurality of operation picture frames comprises the following steps: Convert several operation screen frames into grayscale images; Apply Gaussian blur to reduce image noise; Use an edge detection algorithm to highlight areas that may contain the text.

4. The intelligent reading recommendation method according to claim 2, characterized in that: The text area detection and text recognition on the pre-processed operation picture frame includes the following steps: Use contour detection algorithm to identify contours in the operation picture frame; The OCR algorithm is used to identify the text area of ​​the contour in the extraction recognition operation screen frame.

5. The intelligent reading recommendation method according to claim 2, characterized in that: The Levenshtein distance similarity calculation of texts in adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface includes: The Levenshtein distance formula is used to calculate the distance between texts in adjacent operation screen frames. The Levenshtein distance formula includes Where D(i,j) is the Levenshtein distance between the first i characters of string S and the first j characters of string T, S i is the i-th character of string S, T j is the jth character of string T, cost(S i ,T j ) is the replacement cost, if S i Equal to T j If yes, it is 0, otherwise it is 1; The similarity is calculated by Levenshtein distance. Where D(m,n) is the Levenshtein distance between the texts in the two operation screen frames, and m and n are the lengths of the texts in the two operation screen frames, respectively.

6. The intelligent reading recommendation method according to claim 2, characterized in that: The method of comparing images of adjacent operation screen frames arranged in chronological order on the operation time axis of the operation interface using a feature matching algorithm to obtain image similarity comprises the following steps: Detect feature points in each operation screen frame; Generate descriptors for detected feature points through feature point detection algorithm; Use Euclidean distance to calculate the distance between descriptors and match the feature points between consecutive operation screen frames; Calculate the similarity between adjacent operation screen frames, Where k is the number of matching point pairs, d(a i ,b i ) is the distance between the i-th pair of matching points.

7. The intelligent reading recommendation method according to claim 2, characterized in that: The pictures extract keywords through a character recognition algorithm and sort them according to the number of times the keywords appear, including the following steps: Recognize and extract all text from the picture through OCR algorithm; Clean and standardize data from extracted text, remove stop words, segment words, and reduce vocabulary to its base form; Extract keywords from the processed text and calculate the number of times each word appears in the text using TF-IDF; Sort keywords according to TF-IDF value.

8. The intelligent reading recommendation method according to claim 2, characterized in that: The method for establishing the selection model comprises the following steps: Collect the reviews of each novel and perform pre-processing such as word segmentation and stop word removal to form a novel review database; Collect the category keywords of each novel; Convert category keywords and comments into word vectors; Import word vectors into the CNN model to determine the correlation between comments and keywords and output the selection model.

Images