A computer-based intelligent network data filtering system and method

Abstract

This invention discloses a computer-based intelligent network data filtering system and method, belonging to the field of network data filtering technology. This invention improves the accuracy and precision of text classification by fully exploring semantic similarity to establish connections between text nodes and making full use of the correspondence and frequency between semantic and tag information. It extracts video information from other users who mutually follow the main user, enriching the main user's personal consumption interests through their social groups, thus increasing the richness of the filtered main user's consumption interests D. The main user's consumption interests D are embedded one by one into layers, and after filtering, tags of interest to users with conversion value are obtained. Dependencies in tags are captured using the same layer as the horizontal direction and different layers as the vertical direction, ensuring system stability even when tag dependencies are incomplete or missing, further improving the accuracy of filtering tags of interest to users with conversion value.

Description

Technical Field

[0001] This invention relates to the field of network data filtering technology, specifically a computer-based intelligent network data filtering system and method. Background Technology

[0002] In the field of digital marketing, precise target customer positioning and accurate filtering of online data allow businesses to better understand consumer needs and preferences, and develop more effective marketing strategies. Data mining plays a crucial role in digital marketing, permeating multiple processes such as data collection and preprocessing, customer segmentation and positioning, market trend analysis and forecasting, marketing strategy customization and optimization, personalized recommendations, and marketing and marketing effectiveness evaluation.

[0003] The system collects and analyzes user-generated interactive behavior data to identify the videos and content that users are most interested in from multiple videos they browse. Based on this content, it selects products that customers are interested in, and uses videos with compelling copy and emotional appeal as the medium to push videos that customers are interested in to users, thereby encouraging users to purchase the products in the advertisements and improving the information recall rate of the advertisements.

[0004] In users' lives, items of interest that focus more on emotional experience and consumer interests that focus more on basic living needs often coexist. Products aligned with consumer interests have a higher probability of purchase, but lower profit margins. Conversely, products aligned with user interests have a lower probability of purchase, but higher profit margins. Users don't frequently buy products aligned with their interests, but they also don't always run out of stock. Therefore, setting the push frequency for products aligned with user interests is crucial. Pushing too many or too few of a particular type of product is detrimental to improving ad recall. Current technology cannot effectively filter out user-interested tags with conversion value, resulting in low ad recall rates. Summary of the Invention

[0005] The purpose of this invention is to provide a computer-based intelligent network data filtering system and method to solve the problems raised in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a computer-based intelligent network data filtering method, the method comprising:

[0007] Step S1: Obtain the interaction behavior data generated by the main user;

[0008] Step S2: Analyze the interactive behavior data generated by the main user and filter out videos that the user is interested in;

[0009] Step S3: Perform content analysis on the video obtained in Step S2 and filter out tags that are of interest to the main user and have conversion value.

[0010] Step S2 involves analyzing the interactive behavior data generated by the main user and filtering out videos that the main user is interested in, including:

[0011] Step S21: Assign values ​​to the video's viewing progress, number of views, likes, favorites, and comments: Viewing progress ranges from 0 to 1; 1-2 views are recorded as 0.2, 3-5 views as 0.5, and more than 5 views as 1; no likes are recorded as 0, and likes as 0.1; no favorites are recorded as 0, and favorites as 0.2; no comments are recorded as 0, comments with 1-3 characters are recorded as 0.1, comments with 4-10 characters are recorded as 0.3, and comments with more than 10 characters are recorded as 0.5.

[0012] Step S22: Add up the values ​​corresponding to the browsing progress, number of views, likes, favorites, and comments to obtain the main user's interest value. If the main user's interest value exceeds the interest threshold, the video is determined to be a video that the main user is interested in.

[0013] Step S3 describes performing content analysis on the video obtained in step S2 to obtain tags that the main user is interested in, including:

[0014] Step S31: Analyze the interactive behavior data generated by users browsing other people's videos, and filter the tags M={tag M1, tag M2, ..., tag Mn} that the main user is interested in in the video;

[0015] Step S32: Extract information from videos posted by the main user and extract information from videos posted by other users who are mutually followed by the main user, and filter the main user's consumption interests D;

[0016] Step S33: Simultaneously consider the tags that the main user is interested in and the main user's consumption interests D, and filter the tags that users are interested in that have conversion value.

[0017] Step S31 includes:

[0018] Step S311: Extract the text contained in the video content through the sentences and voice in the video to form a text dataset X={text1, text2, ..., textn}, which represents the text with the label number n. Input the text dataset into the trained text classification network to obtain the classification result predicted by the text classification network.

[0019] Step S312: Treat text nodes and word nodes as two types of nodes, and word-word, word-text, and text-text edges as three types of edges; the text contains a set of words C = {word c1, word c2, ..., word c...} m The cosine similarity formula is used to calculate the similarity between two text nodes x. i x j The cosine similarity between two text nodes is calculated. If the cosine similarity between two text nodes is greater than k1, a straight line is added between the two text nodes, and the cosine similarity value is used as the weight between the text nodes. If the cosine similarity between two text nodes is less than or equal to k1, no straight line is added between the two text nodes.

[0020] A word c i The probability of appearing on the edge is related to another word c. j The product of the probabilities of appearing on the edge is P1, and in the corpus, word c i And the word c j The joint probability of occurrence is P2. Calculate the logarithm of P2 to the base P1. When the logarithm of P2 to the base P1 is greater than 0, it indicates that the correlation between the two word nodes is high, and this value is used as the weight between the two word nodes. A straight line is added between the two words. When the logarithm of P2 to the base P1 is less than or equal to 0, it indicates that the correlation between the two word nodes is low, and no straight line is added between the two words.

[0021] The frequency of a word node appearing in a text node is denoted as P3. If P3 is greater than k2, the word is considered highly important to the text, and P3 serves as the weight between the word node and the text node, with a straight line added between the text and the word. If P3 is less than or equal to k2, the word is considered less important to the text, and no straight line is added between the text and the word. The weights of the edges formed between nodes are transformed into an adjacency matrix graph. By fully exploring semantic similarity to establish connections between text nodes, the connections between text and words, between texts, and between words are improved. By fully utilizing the structural information between each node, and fully utilizing the correspondence and frequency between semantic information and label information, the oversmoothing problem is solved, improving the performance of multi-level text classification and increasing the accuracy and precision of text classification.

[0022] Step S313: Extract adjacency matrix graph information using a multi-layer graph convolutional network. After multiple iterations, obtain nodes as key nodes. Represent the text containing the key nodes using the text of the dataset. Label the text in the dataset with the corresponding labels to obtain the labels that the user is interested in, M={label M1, label M2, ..., label Mn}.

[0023] Step S32 includes:

[0024] Step S321: Information extraction includes extracting the shooting environment, clothing, language, and background music from the video, and extracting the main user's purchase order information. By analyzing the shooting environment, clothing, language, and background music in the video, the main user's consumption interests A = {consumption interest A1, consumption interest A2, ..., consumption interest An} and other users' consumption interests B = {consumption interest B1, consumption interest B2, ..., consumption interest Bn} are obtained. The subsets within consumption interest B are sorted according to the intensity of consumption interest. The intersection of consumption interest A and consumption interest B is consumption interest C. The subsets of consumption interest B corresponding to consumption interest C are marked in consumption interest B and denoted as the central subset. The subsets of consumption interest B adjacent to the central subset are marked and denoted as the marginal subsets. The union of the marginal subsets, the central subset, and consumption interest A is the main user's consumption interest D. Extract information from videos posted by other users who mutually follow the main user, expand and filter the main user's consumption interests using the consumption interests of these other users, enrich the main user's personal consumption interests through their social groups, and increase the richness of the filtered main user's consumption interests D.

[0025] Step S33 includes:

[0026] Step S331: Based on the type of tags, stratify and assign weights to the tags that users are interested in, embed the main user's consumption interest D into each layer, and assign weights to the main user's consumption interest D according to the intensity of the consumption interest; capture the dependencies in the tags by using the same layer as the horizontal direction and different layers as the vertical direction, to ensure the system stability when the tag dependency relationship is incomplete or missing, and further improve the accuracy of filtering user-interested tags with conversion value.

[0027] Step S332: In the same layer, add the weight of the tag that the user is interested in to the weight of the corresponding consumption interest D of the main user to obtain a comparison value. If the comparison value exceeds the preset threshold, it is a tag that the user is interested in and has conversion value.

[0028] A computer-based intelligent network data filtering system, comprising a data acquisition module, a data filtering module one, and a data filtering module two connected to each other;

[0029] The data acquisition module is used to acquire user-generated interactive behavior data through the device used by the user. The user-generated interactive behavior data includes videos posted by other users that the main user browses, videos posted by the main user himself, and interactive behavior logs between the main user and others.

[0030] The data filtering module is used to analyze the interactive behavior data generated by the main user and filter out videos that the user is interested in.

[0031] The second data filtering module is used to perform content analysis on the video obtained in step S2 and filter out tags that are of interest to the main user and have conversion value.

[0032] The data filtering module one includes:

[0033] The main user's interest value calculation module is used to assign values ​​to the video's viewing progress, number of views, likes, favorites, and comments. The values ​​corresponding to the viewing progress, number of views, likes, favorites, and comments are accumulated to obtain the main user's interest value.

[0034] The main user's interest value filtering module is used to filter the main user's interest value. If the main user's interest value exceeds the interest threshold, the video is determined to be a video that the main user is interested in.

[0035] The second data filtering module includes:

[0036] The interest tag extraction module is used to analyze the interactive behavior data generated by users browsing other people's videos and filter the tags that the main user is interested in in the video M={tag M1, tag M2, ..., tag Mn};

[0037] The consumer interest extraction module is used to extract information from videos posted by the main user and videos posted by other users who follow the main user, and to filter the main user's consumer interests (D).

[0038] The value tag extraction module is used to simultaneously consider the tags M that the main user is interested in and the main user's consumption interests D, and to filter the tags M that users are interested in and that have conversion value.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] This invention establishes connections between text nodes by fully exploiting semantic similarity, thus improving the connections between text and words, between texts, and between words. It fully utilizes the structural information between nodes, the correspondence and frequency between semantic and tag information, and solves the oversmoothing problem, improving the performance, accuracy, and precision of multi-level text classification. It extracts video information from other users who mutually follow the main user, expands and filters the main user's consumption interests using the consumption interests of these users, and enriches the main user's personal consumption interests through their social groups, increasing the richness of the filtered consumption interests D. The main user's consumption interests D are embedded layer by layer, and after filtering, tags of interest to users with conversion value are obtained. Dependencies in tags are captured using the same layer as the horizontal direction and different layers as the vertical direction, ensuring system stability even when tag dependencies are incomplete or missing, further improving the accuracy of filtering tags of interest to users with conversion value. Attached Figure Description

[0041] Figure 1 This is a flowchart of a computer-based intelligent network data filtering method according to the present invention;

[0042] Figure 2 This is a diagram illustrating the composition of a computer-based intelligent network data filtering system according to the present invention. Detailed Implementation

[0043] 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.

[0044] Example: Figure 1 As shown, this invention provides a computer-based intelligent network data filtering method, which includes:

[0045] Step S1: Obtain the interaction behavior data generated by the main user;

[0046] Step S2: Analyze the interactive behavior data generated by the main user and filter out videos that the user is interested in;

[0047] Step S3: Perform content analysis on the video obtained in Step S2 and filter out tags that are of interest to the main user and have conversion value.

[0048] Step S2 involves analyzing the interactive behavior data generated by the main user and filtering out videos that the main user is interested in, including:

[0049] Step S21: Assign values ​​to the video's viewing progress, number of views, likes, favorites, and comments: Viewing progress ranges from 0 to 1; 1-2 views are recorded as 0.2, 3-5 views as 0.5, and more than 5 views as 1; no likes are recorded as 0, and likes as 0.1; no favorites are recorded as 0, and favorites as 0.2; no comments are recorded as 0, comments with 1-3 characters are recorded as 0.1, comments with 4-10 characters are recorded as 0.3, and comments with more than 10 characters are recorded as 0.5.

[0050] Step S22: Add up the values ​​corresponding to the browsing progress, number of views, likes, favorites, and comments to obtain the main user's interest value. If the main user's interest value exceeds the interest threshold, the video is determined to be a video that the main user is interested in.

[0051] Step S3 describes performing content analysis on the video obtained in step S2 to obtain tags that the main user is interested in, including:

[0052] Step S31: Analyze the interactive behavior data generated by users browsing other people's videos, and filter the tags M={tag M1, tag M2, ..., tag Mn} that the main user is interested in in the video;

[0053] Step S32: Extract information from videos posted by the main user and extract information from videos posted by other users who are mutually followed by the main user, and filter the main user's consumption interests D;

[0054] Step S33: Simultaneously consider the tags that the main user is interested in and the main user's consumption interests D, and filter the tags that users are interested in that have conversion value.

[0055] Step S31 includes:

[0056] Step S311: Extract the text contained in the video content through the sentences and voice in the video to form a text dataset X={text1, text2, ..., textn}, which represents the text with the label number n. Input the text dataset into the trained text classification network to obtain the classification result predicted by the text classification network.

[0057] Step S312: Treat text nodes and word nodes as two types of nodes, and word-word, word-text, and text-text edges as three types of edges; the text contains a set of words C = {word c1, word c2, ..., word c...} m The cosine similarity formula is used to calculate the similarity between two text nodes x. i x jThe cosine similarity between two text nodes is calculated. If the cosine similarity between two text nodes is greater than k1, a straight line is added between the two text nodes, and the cosine similarity value is used as the weight between the text nodes. If the cosine similarity between two text nodes is less than or equal to k1, no straight line is added between the two text nodes.

[0058] A word c i The probability of appearing on the edge is related to another word c. j The product of the probabilities of appearing on the edge is P1, and in the corpus, word c i And the word c j The joint probability of occurrence is P2. Calculate the logarithm of P2 to the base P1. When the logarithm of P2 to the base P1 is greater than 0, it indicates that the correlation between the two word nodes is high, and this value is used as the weight between the two word nodes. A straight line is added between the two words. When the logarithm of P2 to the base P1 is less than or equal to 0, it indicates that the correlation between the two word nodes is low, and no straight line is added between the two words.

[0059] The frequency of a word node appearing in a text node is denoted as P3. If P3 is greater than k2, the word is considered highly important to the text, and P3 serves as the weight between the word node and the text node, with a straight line added between the text and the word. If P3 is less than or equal to k2, the word is considered less important to the text, and no straight line is added between the text and the word. The weights of the edges formed between nodes are transformed into an adjacency matrix graph. By fully exploring semantic similarity to establish connections between text nodes, the connections between text and words, between texts, and between words are improved. By fully utilizing the structural information between each node, and fully utilizing the correspondence and frequency between semantic information and label information, the oversmoothing problem is solved, improving the performance of multi-level text classification and increasing the accuracy and precision of text classification.

[0060] Step S313: Extract adjacency matrix graph information using a multi-layer graph convolutional network. After multiple iterations, obtain nodes as key nodes. Represent the text containing the key nodes using the text of the dataset. Label the text in the dataset with the corresponding labels to obtain the labels that the user is interested in, M={label M1, label M2, ..., label Mn}.

[0061] Step S32 includes:

[0062] Step S321: Information extraction includes extracting the shooting environment, clothing, language, and background music from the video, and extracting the main user's purchase order information. By analyzing the shooting environment, clothing, language, and background music in the video, the main user's consumption interests A = {consumption interest A1, consumption interest A2, ..., consumption interest An} and other users' consumption interests B = {consumption interest B1, consumption interest B2, ..., consumption interest Bn} are obtained. The subsets within consumption interest B are sorted according to the intensity of consumption interest. The intersection of consumption interest A and consumption interest B is consumption interest C. The subsets of consumption interest B corresponding to consumption interest C are marked in consumption interest B and denoted as the central subset. The subsets of consumption interest B adjacent to the central subset are marked and denoted as the marginal subsets. The union of the marginal subsets, the central subset, and consumption interest A is the main user's consumption interest D. Extract information from videos posted by other users who mutually follow the main user, expand and filter the main user's consumption interests using the consumption interests of these other users, enrich the main user's personal consumption interests through their social groups, and increase the richness of the filtered main user's consumption interests D.

[0063] Step S33 includes:

[0064] Step S331: Based on the type of tags, stratify and assign weights to the tags that users are interested in, embed the main user's consumption interest D into each layer, and assign weights to the main user's consumption interest D according to the intensity of the consumption interest; capture the dependencies in the tags by using the same layer as the horizontal direction and different layers as the vertical direction, to ensure the system stability when the tag dependency relationship is incomplete or missing, and further improve the accuracy of filtering user-interested tags with conversion value.

[0065] Step S332: In the same layer, add the weight of the tag that the user is interested in to the weight of the corresponding consumption interest D of the main user to obtain a comparison value. If the comparison value exceeds the preset threshold, it is a tag that the user is interested in and has conversion value.

[0066] Figure 2 A computer-based intelligent network data filtering system, comprising a data acquisition module, a data filtering module one, and a data filtering module two connected to each other;

[0067] The data acquisition module is used to acquire user-generated interactive behavior data through the device used by the user. The user-generated interactive behavior data includes videos posted by other users that the main user browses, videos posted by the main user himself, and interactive behavior logs between the main user and others.

[0068] The data filtering module is used to analyze the interactive behavior data generated by the main user and filter out videos that the user is interested in.

[0069] The second data filtering module is used to perform content analysis on the video obtained in step S2 and filter out tags that are of interest to the main user and have conversion value.

[0070] The data filtering module one includes:

[0071] The main user's interest value calculation module is used to assign values ​​to the video's viewing progress, number of views, likes, favorites, and comments. The values ​​corresponding to the viewing progress, number of views, likes, favorites, and comments are accumulated to obtain the main user's interest value.

[0072] The main user's interest value filtering module is used to filter the main user's interest value. If the main user's interest value exceeds the interest threshold, the video is determined to be a video that the main user is interested in.

[0073] The second data filtering module includes:

[0074] The interest tag extraction module is used to analyze the interactive behavior data generated by users browsing other people's videos and filter the tags that the main user is interested in in the video M={tag M1, tag M2, ..., tag Mn};

[0075] The consumer interest extraction module is used to extract information from videos posted by the main user and videos posted by other users who follow the main user, and to filter the main user's consumer interests (D).

[0076] The value tag extraction module is used to simultaneously consider the tags that the main user is interested in and the main user's consumption interests (D), and to filter the tags that users are interested in and that have conversion value.

[0077] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A computer-based intelligent network data filtering method, characterized in that: The method includes: Step S1: Obtain the interaction behavior data generated by the main user; Step S2: Analyze the interactive behavior data generated by the main user and filter out videos that the user is interested in; Step S3: Perform content analysis on the video obtained in step S2 and filter out tags that are of interest to the main user and have conversion value; Step S3 describes performing content analysis on the video obtained in step S2 to obtain tags that the main user is interested in, including: Step S31: Analyze the interactive behavior data generated by users browsing other people's videos, and filter the tags M={tag M1, tag M2, ..., tag Mn} that the main user is interested in in the video; Step S32: Extract information from videos posted by the main user and extract information from videos posted by other users who are mutually followed by the main user, and filter the main user's consumption interests D; Step S33: Simultaneously consider the tags that the main user is interested in and the main user's consumption interests D, and filter the tags that users are interested in that have conversion value; Step S31 includes: Step S311: Extract the text contained in the video content through the sentences and voice in the video to form a text dataset X={text1, text2, ..., textn}, which represents the text with the label number n. Input the text dataset into the trained text classification network to obtain the classification result predicted by the text classification network. Step S312: Treat text nodes and word nodes as two types of nodes, and word-word, word-text, and text-text edges as three types of edges; the text contains a set of words C = {word c1, word c2, ..., word c...} m The cosine similarity formula is used to calculate the similarity between two text nodes x. i x j The cosine similarity between two text nodes is calculated. If the cosine similarity between two text nodes is greater than k1, a straight line is added between the two text nodes, and the cosine similarity value is used as the weight between the text nodes. If the cosine similarity between two text nodes is less than or equal to k1, no straight line is added between the two text nodes. A word c i The probability of appearing on the edge is related to another word c. j The product of the probabilities of appearing on the edge is P1, and in the corpus, word c i And the word c j The joint probability of occurrence is P2. Calculate the logarithm of P2 to the base P1. When the logarithm of P2 to the base P1 is greater than 0, it indicates that the correlation between the two word nodes is high, and this value is used as the weight between the two word nodes. A straight line is added between the two words. When the logarithm of P2 to the base P1 is less than or equal to 0, it indicates that the correlation between the two word nodes is low, and no straight line is added between the two words. The frequency of a word node appearing in a text node is denoted as P3. If P3 is greater than k2, the word is highly important to the text, and P3 is used as the weight between the word node and the text node. A straight line is added between the text and the word. If P3 is less than or equal to k2, the word is less important to the text, and no straight line is added between the text and the word. The weights of the edges between nodes are transformed into an adjacency matrix graph. Step S313: Extract adjacency matrix graph information using a multi-layer graph convolutional network. After multiple iterations, obtain nodes as key nodes. Represent the text containing the key nodes using the text of the dataset. Label the text in the dataset with the corresponding labels to obtain the labels that the user is interested in, M={label M1, label M2, ..., label Mn}. Step S32 includes: Step S321: Information extraction includes extracting the shooting environment, clothing, language, and background music from the video, extracting the main user's purchase order information, and analyzing the shooting environment, clothing, language, and background music in the video to obtain the main user's consumption interests A = {consumption interest A1, consumption interest A2, ..., consumption interest An} and other users' consumption interests B = {consumption interest B1, consumption interest B2, ..., consumption interest Bn}. The subsets within consumption interest B are sorted according to the intensity of consumption interest. The intersection of consumption interest A and consumption interest B is consumption interest C. In consumption interest B, the subsets of consumption interest B corresponding to consumption interest C are marked and denoted as the central subset. The subsets of consumption interest B adjacent to the central subset are then marked and denoted as the marginal subsets. The union of the marginal subsets, the central subset, and consumption interest A is the main user's consumption interest D. Step S33 includes: Step S331: Based on the type of tags, the tags M that users are interested in are layered and weighted, and the main user's consumption interests D are embedded into the layers one by one. Based on the intensity of consumption interests, the main user's consumption interests D are weighted one by one. Step S332: In the same layer, add the weight of the tag M that the user is interested in and the corresponding weight of the consumption interest D that the main user is interested in to obtain a comparison value. If the comparison value exceeds the preset threshold, then it is the tag M that the user is interested in and has conversion value.

2. The computer-based intelligent network data filtering method according to claim 1, characterized in that: Step S2 involves analyzing the interactive behavior data generated by the main user and filtering out videos that the main user is interested in, including: Step S21: Assign values ​​to the video's viewing progress, number of views, likes, favorites, and comments; Step S22: The values ​​corresponding to the browsing progress, number of views, likes, favorites, and comments are summed to obtain the main user's interest value. If the main user's interest value exceeds the interest threshold, the video is determined to be a video that the main user is interested in.

3. A computer-based intelligent network data filtering system, applied to the computer-based intelligent network data filtering method described in any one of claims 1-2, characterized in that: The system includes a data acquisition module, a data filtering module one, and a data filtering module two that are connected to each other; The data acquisition module is used to acquire user-generated interactive behavior data through the device used by the user. The user-generated interactive behavior data includes videos posted by other users that the main user browses, videos posted by the main user himself, and interactive behavior logs between the main user and others. The second data filtering module includes: The interest tag extraction module is used to analyze the interactive behavior data generated by users browsing other people's videos and filter the tags that the main user is interested in in the video M={tag M1, tag M2, ..., tag Mn}; The consumer interest extraction module is used to extract information from videos posted by the main user and videos posted by other users who are mutually followed by the main user, and to filter the main user's consumer interests (D). The value tag extraction module is used to simultaneously consider the tags that the main user is interested in and the main user's consumption interests (D), and to filter the tags that users are interested in and that have conversion value.

4. The computer-based intelligent network data filtering system according to claim 3, characterized in that: The data filtering module is used to analyze the interactive behavior data generated by the main user and filter out videos that the user is interested in. The second data filtering module is used to perform content analysis on the video obtained in step S2 and filter out tags that are of interest to the main user and have conversion value.

5. The computer-based intelligent network data filtering system according to claim 3, characterized in that: The data filtering module one includes: The main user's interest value calculation module is used to assign values ​​to the video's viewing progress, number of views, likes, favorites, and comments. The values ​​corresponding to the viewing progress, number of views, likes, favorites, and comments are accumulated to obtain the main user's interest value. The main user's interest value filtering module is used to filter the main user's interest value. If the main user's interest value exceeds the interest threshold, the video is determined to be a video that the main user is interested in.

Images