Short video personalized recommendation method and system based on big data

Abstract

This invention relates to the field of personalized recommendation and discloses a method and system for personalized short video recommendation based on big data. It monitors user behavior by analyzing browsing and emotional behaviors, identifies similarities between users by analyzing user viewing activity and social interactions with other users on short video platforms, monitors and identifies dynamic changes in user preferences by analyzing user dwell time, repeated viewing frequency, sharing behavior, and collection behavior data on a particular short video interface, and sorts the playback order of short videos by comprehensively analyzing user behavior monitoring coefficients, activity distribution coefficients, user similarity, and user preference identification coefficients. Based on the monitored personalized recommendation effect, the playback order of short videos is adjusted, improving the accuracy, dynamic adaptability, and diversity of recommended content.

Description

Technical Field

[0001] This invention relates to the field of personalized recommendation technology, and more specifically to a method and system for personalized recommendation of short videos based on big data. Background Technology

[0002] Traditional short video recommendation methods often rely on simple popular ranking or keyword matching, failing to meet the diverse needs of users. Popular ranking simply shows users videos with high view counts and likes, but these videos may not necessarily match every user's interests. Furthermore, keyword matching is prone to inaccurate matching, resulting in a discrepancy between the recommended videos and the user's actual needs.

[0003] With the continuous development of big data analytics technologies, such as data mining, machine learning, and deep learning, valuable information and patterns can be extracted from massive amounts of data. These can be used to discover potential patterns and association rules in user behavior. Machine learning algorithms can build user interest models to predict and classify user behavior. By utilizing neural network models in deep learning, video content and user features can be deeply encoded to calculate the similarity between users and videos, thereby achieving more accurate recommendations. However, the above process still has the following drawbacks: First, existing recommendation methods cannot capture rapid changes in user interests and preferences in a timely manner. When user interests shift, the recommended content remains based on old preferences and cannot provide users with the latest content that best meets their current needs. Secondly, existing recommendation methods do not fully consider user behaviors other than viewing, such as social interaction, collection, and sharing, resulting in an incomplete understanding of user interests and a lack of diversity in recommended content.

[0004] Third, existing recommendation methods struggle to accurately grasp users' complex interests and needs in different scenarios, leading to discrepancies between recommended content and users' actual preferences. Users end up seeing a large number of videos they are not interested in, reducing the click-through rate and viewing time of recommended content. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a short video personalized recommendation method and system based on big data, so as to solve the problems existing in the background art.

[0006] This invention provides the following technical solution: a personalized recommendation method for short videos based on big data, comprising: S1: Real-time monitoring and collection of user behavior data, viewed content data, and social interaction data; S2: Real-time monitoring of user behavior status: Based on the collected user behavior data, analyze user browsing behavior and emotional behavior to obtain user behavior monitoring coefficients; S3: Based on the collected viewing content data, analyze the user's viewing activity at different time periods to obtain the activity distribution coefficient; S4: Identify similarity between users: Based on the collected social interaction data, analyze the social interactions between users and other users on the short video platform to obtain user similarity; S5: Monitor and identify dynamic changes in user preferences: Based on the analysis of user dwell time, number of times they watch, sharing behavior data, and collection behavior data on a short video interface, a user preference identification coefficient is obtained; S6: Based on a comprehensive analysis of user behavior monitoring coefficients, activity distribution coefficients, user similarity, and user preference identification coefficients, a personalized recommendation index is calculated; based on the personalized recommendation index, the playback order of short videos is sorted, and the playback order of short videos is recommended to users according to the sorting results. S7: Monitor the personalized recommendation effect in real time, obtain the recommendation effect evaluation coefficient, and adjust the playback order of short videos in real time based on the recommendation effect evaluation coefficient.

[0007] Preferably, the specific method of S1 is as follows: by embedding a log recording function on the short video platform, user behavior data is captured in real time; by extracting the video metadata during the video uploading or transcoding stage, and then associating the user's viewing behavior with the video content, the user's viewing content data is extracted; by setting up tracking points on social interaction elements, the user's social interaction behavior is monitored in real time, and the collected data is periodically uploaded in batches to the data processing center, where the data processing center preprocesses and stores the collected data according to the collection time.

[0008] Preferably, the specific method of S2 is as follows: Step S201: Record the actual viewing time of each short video, the view rate of each short video, the number of videos viewed by the user on the video list page, and the number of times each short video is watched repeatedly; based on the recorded data, analyze the user's browsing behavior and calculate the user browsing behavior monitoring index: ; Where I represents the user browsing behavior monitoring index, t i t represents the actual viewing time of the i-th video. max t represents the longest actual viewing time among all videos. min r represents the shortest actual viewing time among all videos. i The view rate of the i-th video, rmax r represents the maximum number of views among all videos. min Let m represent the minimum view count among all videos. i m represents the number of times the i-th video has been watched repeatedly. max m represents the maximum number of times a video has been viewed repeatedly. min The minimum number of times a video is viewed repeatedly is represented by , where n represents the number of short videos watched by the user, and N represents the number of videos viewed by the user. max This indicates the preset threshold for the maximum number of views for the video list; Step S202: Using natural language processing technology, perform sentiment analysis on user comments to calculate the user sentiment behavior monitoring index: ; Where P represents the user's emotional behavior monitoring index, S j Let represent the sentiment score of the j-th comment, and M represent the total number of comments within the time window; Step S203: Based on a comprehensive analysis of the user browsing behavior monitoring index and the user emotional behavior monitoring index, the user behavior monitoring coefficient is calculated as follows: ; Where R represents the user behavior monitoring coefficient, and w1 and w2 represent the weighting coefficients, respectively.

[0009] Preferably, the specific method of S3 is as follows: Step S301: By dividing a day into H equal-length time periods h, the viewing duration of each time period is calculated as follows: ; Where TVD(u) represents the viewing time of user u in each time period, and D k This represents the duration of the k-th viewing record for user u within each time period; The total viewing time for users was calculated as follows: ; Where T(u) represents the total viewing time of user u; Step S302: Based on the viewing duration of each time period and the total viewing duration of users, calculate the viewing percentage of each time period as follows: ; Among them, Y h (u) represents the percentage of time user u is viewed in each time period h; Step S303: Based on the analysis of the viewing percentage for each time period, the activity distribution coefficient is calculated as follows: ; Where A represents the activity distribution coefficient.

[0010] Preferably, the specific method of S4 is as follows: A user-content matrix is ​​created using collected social interaction data between users on short video platforms, including likes, comments, shares, favorites, and follows. Each element represents the intensity of a user's interaction with specific content. Based on the created matrix, user similarity is calculated using the following formula: ; Where, x u, d x represents the intensity of user u's interaction with content d. v, d This represents the interaction intensity of user v with content d, where u and v represent two different users.

[0011] Preferably, the specific formula for calculating the user preference recognition coefficient in S5 is as follows: ; Where E represents the user preference recognition coefficient, t i Let n represent the actual viewing time of the i-th video, n represent the number of short videos watched by the user, and b represent the number of videos watched repeatedly. 总 α represents the total number of videos viewed, q represents the number of times a user interacts with the videos they watch, p represents the total number of times a user watches videos within a specific time period, c represents the number of videos saved, z represents the number of videos watched, l represents the number of videos shared, and α1, α2, α3, α4, and α5 represent the weighting coefficients.

[0012] Preferably, the specific method of S6 is as follows: by comprehensively analyzing the user behavior monitoring coefficient, activity distribution coefficient, user similarity, and user preference identification coefficient, the personalized recommendation index is calculated as follows: ; Among them, Q i R represents the personalized recommendation index, i=1,2,…,n; i Let θ represent the user behavior monitoring coefficient for the i-th video. R A represents the preset user behavior monitoring threshold. i Let θ represent the activity distribution coefficient of the i-th video. A This represents the preset activity distribution threshold, cos(u, v). i Let θ represent the user similarity of the i-th video. cos(u, v) E represents the preset user similarity threshold. i Let θ represent the user preference recognition coefficient for the i-th video. E This indicates a preset user preference threshold; Based on the personalized recommendation index Qi Sort the playback order of the short videos and generate a sorted list as G = sort(Q1, Q2, ..., Q...). n The sorted list of short videos will be sent to the user's device.

[0013] Preferably, the specific method of S7 is as follows: by real-time monitoring of the personalized recommendation effect indicators of each video, including click-through rate, viewing duration, completion rate, interaction rate, and bounce rate, and by comprehensively calculating based on the personalized recommendation effect indicators of each video, a recommendation effect evaluation coefficient is obtained. The specific calculation formula is as follows: ; Where, η i β represents the recommendation effectiveness evaluation coefficient for the i-th video. i, v C represents the weight coefficient of the v-th recommendation performance metric for the i-th video. i, v This represents the v-th recommendation performance metric for the i-th video; By using the recommendation effectiveness evaluation coefficient η i Compared with the preset recommended effect threshold γ, when η i If the value is less than γ, the recommendation effect is considered poor, the playback order of the video needs to be adjusted, and the short videos should be sorted according to the real-time calculated recommendation effect evaluation coefficient value, with the video with the higher recommendation effect evaluation coefficient value placed first, and the current video replaced by an alternative video with an even higher recommendation effect evaluation coefficient value.

[0014] This invention also provides a short video personalized recommendation system based on big data, used to implement the aforementioned short video personalized recommendation method based on big data, comprising: Data acquisition module: used to monitor and collect user behavior data, viewed content data, and social interaction data in real time; User behavior analysis module: Used to analyze user browsing behavior and emotional behavior based on the collected user behavior data, and obtain user behavior monitoring coefficients; User activity analysis module: Based on the collected viewing content data, it analyzes the user's viewing activity at different time periods and obtains the activity distribution coefficient; User social interaction analysis module: Based on the collected social interaction data, it analyzes the social interactions between users and other users on the short video platform to obtain user similarity. User preference analysis module: This module analyzes user dwell time, number of times they watch a video repeatedly, sharing behavior data, and collection behavior data on a short video interface to obtain user preference identification coefficients. Personalized recommendation module: This module calculates a personalized recommendation index based on a comprehensive analysis of user behavior monitoring coefficients, activity distribution coefficients, user similarity, and user preference identification coefficients. It then sorts the playback order of short videos based on the personalized recommendation index and recommends the playback order of short videos to users according to the sorting results. Recommendation effect evaluation module: Used to monitor the effect of personalized recommendations in real time, obtain the recommendation effect evaluation coefficient, and adjust the playback order of short videos in real time based on the recommendation effect evaluation coefficient.

[0015] The technical effects and advantages of this invention are as follows: By analyzing user browsing and emotional behaviors, user viewing activity at different times, and user dwell time, repetition count, sharing and collection data on a particular short video page, we can promptly capture rapid changes in user interests and preferences and provide users with the latest and most relevant content.

[0016] By analyzing users from different perspectives, such as user behavior, viewing activity, social interaction, and user preferences, we can gain a deeper understanding of users' interests and needs, making the understanding of user interests more comprehensive and enabling the diversity of recommended content.

[0017] Based on a comprehensive analysis of user behavior monitoring coefficients, viewing activity analysis results, user similarity, and user preference identification coefficients, a personalized recommendation index is calculated. This index comprehensively considers various user characteristics and needs, improves the accuracy of recommendations, and further increases the click-through rate and viewing time of recommended content. Attached Figure Description

[0018] Figure 1 This is a diagram illustrating the method steps of the present invention.

[0019] Figure 2 This is a system structure block diagram of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The short video personalized recommendation method and system based on big data involved in the present invention are not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, this embodiment provides a short video personalized recommendation method based on big data, including: S1: Real-time monitoring and collection of user behavior data, viewed content data, and social interaction data; S2: Real-time monitoring of user behavior status: Based on the collected user behavior data, analyze user browsing behavior and emotional behavior to obtain user behavior monitoring coefficients; S3: Based on the collected viewing content data, analyze the user's viewing activity at different time periods to obtain the activity distribution coefficient; S4: Identify similarity between users: Based on the collected social interaction data, analyze the social interactions between users and other users on the short video platform to obtain user similarity; S5: Monitor and identify dynamic changes in user preferences: Based on the analysis of user dwell time, number of times they watch, sharing behavior data, and collection behavior data on a short video interface, a user preference identification coefficient is obtained; S6: Based on a comprehensive analysis of user behavior monitoring coefficients, activity distribution coefficients, user similarity, and user preference identification coefficients, a personalized recommendation index is calculated; based on the personalized recommendation index, the playback order of short videos is sorted, and the playback order of short videos is recommended to users according to the sorting results. S7: Monitor the personalized recommendation effect in real time, obtain the recommendation effect evaluation coefficient, and adjust the playback order of short videos in real time based on the recommendation effect evaluation coefficient.

[0022] In this embodiment, the specific method of S1 is as follows: by embedding a log recording function on the short video platform, user behavior data is captured in real time; by extracting the video metadata during the video uploading or transcoding stage, and then associating the user's viewing behavior with the video content, the user's viewing content data is extracted; by setting up tracking points on social interaction elements, the user's social interaction behavior is monitored in real time, and the collected data is periodically uploaded in batches to the data processing center, where the data processing center preprocesses and stores the collected data according to the collection time.

[0023] In this embodiment, the specific method of S2 is as follows: Step S201: Record the actual viewing time of each short video, the view rate of each short video, the number of videos viewed by the user on the video list page, and the number of times each short video is watched repeatedly; based on the recorded data, analyze the user's browsing behavior and calculate the user browsing behavior monitoring index: ; Where I represents the user browsing behavior monitoring index, t i t represents the actual viewing time of the i-th video. max t represents the longest actual viewing time among all videos. minr represents the shortest actual viewing time among all videos. i The view rate of the i-th video, r max r represents the maximum number of views among all videos. min Let m represent the minimum view count among all videos. i m represents the number of times the i-th video has been watched repeatedly. max m represents the maximum number of times a video has been viewed repeatedly. min The minimum number of times a video is viewed repeatedly is represented by , where n represents the number of short videos watched by the user, and N represents the number of videos viewed by the user. max This indicates the preset threshold for the maximum number of views for the video list; Step S202: Using natural language processing technology, perform sentiment analysis on user comments to calculate the user sentiment behavior monitoring index: ; Where P represents the user's emotional behavior monitoring index, S j Let represent the sentiment score of the j-th comment, and M represent the total number of comments within the time window; Step S203: Based on a comprehensive analysis of the user browsing behavior monitoring index and the user emotional behavior monitoring index, the user behavior monitoring coefficient is calculated as follows: ; Where R represents the user behavior monitoring coefficient, and w1 and w2 represent the weighting coefficients, respectively.

[0024] In this embodiment, the specific method of S3 is as follows: Step S301: By dividing a day into H equal-length time periods h, the viewing duration of each time period is calculated as follows: ; Where TVD(u) represents the viewing time of user u in each time period, and D k This represents the duration of the k-th viewing record for user u within each time period; The total viewing time for users was calculated as follows: ; Where T(u) represents the total viewing time of user u; Step S302: Based on the viewing duration of each time period and the total viewing duration of users, calculate the viewing percentage of each time period as follows: ; Among them, Y h (u) represents the percentage of time user u is viewed in each time period h; Step S303: Based on the analysis of the viewing percentage for each time period, the activity distribution coefficient is calculated as follows: ; Where A represents the activity distribution coefficient.

[0025] In this embodiment, the specific method of S4 is as follows: A user-content matrix is ​​created using collected social interaction data between users on the short video platform, including likes, comments, shares, favorites, and follows. Each element represents the intensity of a user's interaction with specific content. Based on the created matrix, user similarity is calculated using the following formula: ; Where, x u, d x represents the intensity of user u's interaction with content d. v, d This represents the interaction intensity of user v with content d, where u and v represent two different users.

[0026] In this embodiment, the specific calculation formula for the user preference recognition coefficient in S5 is as follows: ; Where E represents the user preference recognition coefficient, t i Let n represent the actual viewing time of the i-th video, n represent the number of short videos watched by the user, and b represent the number of videos watched repeatedly. 总 α represents the total number of videos viewed, q represents the number of times a user interacts with the videos they watch, p represents the total number of times a user watches videos within a specific time period, c represents the number of videos saved, z represents the number of videos watched, l represents the number of videos shared, and α1, α2, α3, α4, and α5 represent the weighting coefficients.

[0027] In this embodiment, the specific method of S6 is as follows: by comprehensively analyzing the user behavior monitoring coefficient, activity distribution coefficient, user similarity, and user preference identification coefficient, the personalized recommendation index is calculated as follows: ; Among them, Q i R represents the personalized recommendation index, i=1,2,…,n; i Let θ represent the user behavior monitoring coefficient for the i-th video. R A represents the preset user behavior monitoring threshold. i Let θ represent the activity distribution coefficient of the i-th video. A This represents the preset activity distribution threshold, cos(u, v). i Let θ represent the user similarity of the i-th video. cos(u, v) E represents the preset user similarity threshold.i Let θ represent the user preference recognition coefficient for the i-th video. E This indicates a preset user preference threshold; Based on the personalized recommendation index Q i Sort the playback order of the short videos and generate a sorted list as G = sort(Q1, Q2, ..., Q...). n The sorted list of short videos will be sent to the user's device.

[0028] In this embodiment, the specific method of S7 is as follows: By real-time monitoring of the personalized recommendation effect indicators of each video, including click-through rate, viewing duration, completion rate, interaction rate, and bounce rate, and by comprehensively calculating based on the personalized recommendation effect indicators of each video, a recommendation effect evaluation coefficient is obtained. The specific calculation formula is as follows: ; Where, η i β represents the recommendation effectiveness evaluation coefficient for the i-th video. i, v C represents the weight coefficient of the v-th recommendation performance metric for the i-th video. i, v This represents the v-th recommendation performance metric for the i-th video; By using the recommendation effectiveness evaluation coefficient η i Compared with the preset recommended effect threshold γ, when η i If the value is less than γ, the recommendation effect is considered poor, the playback order of the video needs to be adjusted, and the short videos should be sorted according to the real-time calculated recommendation effect evaluation coefficient value, with the video with the higher recommendation effect evaluation coefficient value placed first, and the current video replaced by an alternative video with an even higher recommendation effect evaluation coefficient value.

[0029] It should be noted that by continuously monitoring user interaction with recommended videos, the recommendation effect evaluation coefficient of each video is calculated in real time. Based on the preset strategy and the real-time calculated E value, the playback order of short videos is dynamically adjusted. For example, if the E value of a video is consistently below the threshold, it can be considered to move it to the back or replace it; if the E value of a video is consistently above the threshold, it can be considered to move it to the front.

[0030] This embodiment also provides a short video personalized recommendation system based on big data, used to implement the aforementioned short video personalized recommendation method based on big data. The system includes a data acquisition module, a user behavior analysis module, a user activity analysis module, a user social interaction analysis module, a user preference analysis module, a personalized recommendation module, and a recommendation effect evaluation module. The data acquisition module is connected to the user behavior analysis module, the user activity analysis module, the user social interaction analysis module, and the user preference analysis module. All of these modules are connected to the personalized recommendation module, which is also connected to the recommendation effect evaluation module.

[0031] Data acquisition module: used to monitor and collect user behavior data, viewed content data, and social interaction data in real time; User behavior analysis module: Used to analyze user browsing behavior and emotional behavior based on the collected user behavior data, and obtain user behavior monitoring coefficients; User activity analysis module: Based on the collected viewing content data, it analyzes the user's viewing activity at different time periods and obtains the activity distribution coefficient; User social interaction analysis module: Based on the collected social interaction data, it analyzes the social interactions between users and other users on the short video platform to obtain user similarity. User preference analysis module: This module analyzes user dwell time, number of times they watch a video repeatedly, sharing behavior data, and collection behavior data on a short video interface to obtain user preference identification coefficients. Personalized recommendation module: This module calculates a personalized recommendation index based on a comprehensive analysis of user behavior monitoring coefficients, activity distribution coefficients, user similarity, and user preference identification coefficients. It then sorts the playback order of short videos based on the personalized recommendation index and recommends the playback order of short videos to users according to the sorting results. Recommendation effect evaluation module: Used to monitor the effect of personalized recommendations in real time, obtain the recommendation effect evaluation coefficient, and adjust the playback order of short videos in real time based on the recommendation effect evaluation coefficient.

[0032] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0033] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A personalized recommendation method for short videos based on big data, characterized in that, include: S1: Real-time monitoring and collection of user behavior data, viewed content data, and social interaction data; S2: Real-time monitoring of user behavior status: Based on the collected user behavior data, analyze user browsing behavior and emotional behavior to obtain user behavior monitoring coefficients; S3: Based on the collected viewing content data, analyze the user's viewing activity at different time periods to obtain the activity distribution coefficient; S4: Identify similarity between users: Based on the collected social interaction data, analyze the social interactions between users and other users on the short video platform to obtain user similarity; S5: Monitor and identify dynamic changes in user preferences: Based on the analysis of user dwell time, number of times they watch, sharing behavior data, and collection behavior data on a short video interface, a user preference identification coefficient is obtained; S6: Based on a comprehensive analysis of user behavior monitoring coefficients, activity distribution coefficients, user similarity, and user preference identification coefficients, a personalized recommendation index is calculated. Based on a personalized recommendation index, the playback order of short videos is sorted, and the playback order of short videos is recommended to users according to the sorting results. S7: Monitor the personalized recommendation effect in real time, obtain the recommendation effect evaluation coefficient, and adjust the playback order of short videos in real time based on the recommendation effect evaluation coefficient.

2. The short video personalized recommendation method based on big data according to claim 1, characterized in that, The specific method of S1 is as follows: by embedding a log recording function on the short video platform, user behavior data is captured in real time; by extracting the video metadata during the video uploading or transcoding stage, the user's viewing behavior is associated with the video content, thereby extracting the user's viewing content data; By setting up tracking points on social interaction elements, users' social interaction behavior can be monitored in real time. The collected data is then uploaded in batches to the data processing center on a regular basis. The data processing center preprocesses and stores the collected data according to the collection time.

3. The short video personalized recommendation method based on big data according to claim 1, characterized in that, The specific method of S2 is as follows: Step S201: Record the actual viewing time of each short video, the view rate of each short video, the number of videos viewed by the user on the video list page, and the number of times each short video is watched repeatedly; based on the recorded data, analyze the user's browsing behavior and calculate the user browsing behavior monitoring index: ； Where I represents the user browsing behavior monitoring index, t i t represents the actual viewing time of the i-th video. max t represents the longest actual viewing time among all videos. min r represents the shortest actual viewing time among all videos. i The view rate of the i-th video, r max r represents the maximum number of views among all videos. min Let m represent the minimum view count among all videos. i m represents the number of times the i-th video has been watched repeatedly. max m represents the maximum number of times a video has been viewed repeatedly. min The minimum number of times a video is viewed repeatedly is represented by , where n represents the number of short videos watched by the user, and N represents the number of videos viewed by the user. max This indicates the preset threshold for the maximum number of views for the video list; Step S202: Using natural language processing technology, perform sentiment analysis on user comments to calculate the user sentiment behavior monitoring index: ； Where P represents the user's emotional behavior monitoring index, S j Let represent the sentiment score of the j-th comment, and M represent the total number of comments within the time window; Step S203: Based on a comprehensive analysis of the user browsing behavior monitoring index and the user emotional behavior monitoring index, the user behavior monitoring coefficient is calculated as follows: ； Where R represents the user behavior monitoring coefficient, and w1 and w2 represent the weighting coefficients, respectively.

4. The short video personalized recommendation method based on big data according to claim 1, characterized in that, The specific method of S3 is as follows: Step S301: By dividing a day into H equal-length time periods h, the viewing duration of each time period is calculated as follows: ； Where TVD(u) represents the viewing time of user u in each time period, and D k This represents the duration of the k-th viewing record for user u within each time period; The total viewing time for users was calculated as follows: ； Where T(u) represents the total viewing time of user u; Step S302: Based on the viewing duration of each time period and the total viewing duration of users, calculate the viewing percentage of each time period as follows: ； Among them, Y h (u) represents the percentage of time user u is viewed in each time period h; Step S303: Based on the analysis of the viewing percentage for each time period, the activity distribution coefficient is calculated as follows: ； Where A represents the activity distribution coefficient.

5. The short video personalized recommendation method based on big data according to claim 1, characterized in that, The specific method of S4 is as follows: A user-content matrix is ​​created using collected social interaction data between users on short video platforms, including likes, comments, shares, favorites, and follows. Each element represents the intensity of a user's interaction with specific content. Based on the created matrix, user similarity is calculated using the following formula: ； Where, x u, d x represents the intensity of user u's interaction with content d. v, d This represents the interaction intensity of user v with content d, where u and v represent two different users.

6. The method for personalized recommendation of short videos based on big data according to claim 1, characterized in that, The specific formula for calculating the user preference identification coefficient in S5 is as follows: ； Where E represents the user preference recognition coefficient, t i Let n represent the actual viewing time of the i-th video, n represent the number of short videos watched by the user, and b represent the number of videos watched repeatedly. 总 α represents the total number of videos viewed, q represents the number of times the user interacted with the viewed videos, p represents the total number of times the user watched videos within a specific time period, c represents the number of videos saved, z represents the number of videos viewed, l represents the number of videos shared, and α1, α2, α3, α4, and α5 represent the weighting coefficients.

7. The short video personalized recommendation method based on big data according to claim 1, characterized in that, The specific method of S6 is as follows: by comprehensively analyzing the user behavior monitoring coefficient, activity distribution coefficient, user similarity, and user preference identification coefficient, the personalized recommendation index is calculated as follows: ； Among them, Q i R represents the personalized recommendation index, i=1,2,…,n; i Let θ represent the user behavior monitoring coefficient for the i-th video. R A represents the preset user behavior monitoring threshold. i Let θ represent the activity distribution coefficient of the i-th video. A This represents the preset activity distribution threshold, cos(u, v). i Let θ represent the user similarity of the i-th video. cos(u, v) E represents the preset user similarity threshold. i Let θ represent the user preference recognition coefficient for the i-th video. E This indicates a preset user preference threshold; Based on the personalized recommendation index Q i Sort the short videos according to their playback order and generate a sorted list as G = sort(Q1,Q2,…,Q n The sorted list of short videos will be sent to the user's device.

8. The method for personalized recommendation of short videos based on big data according to claim 1, characterized in that, The specific method of S7 is as follows: By real-time monitoring of the personalized recommendation effect indicators of each video, including click-through rate, viewing duration, completion rate, interaction rate, and bounce rate, and by comprehensively calculating based on the personalized recommendation effect indicators of each video, a recommendation effect evaluation coefficient is obtained. The specific calculation formula is as follows: ； Where, η i β represents the recommendation effectiveness evaluation coefficient for the i-th video. i, v C represents the weight coefficient of the v-th recommendation performance metric for the i-th video. i, v This represents the v-th recommendation performance metric for the i-th video; By using the recommendation effectiveness evaluation coefficient η i Compared with the preset recommended effect threshold γ, when η i If the value is less than γ, the recommendation effect is considered poor, the playback order of the video needs to be adjusted, and the short videos should be sorted according to the real-time calculated recommendation effect evaluation coefficient value, with the video with the higher recommendation effect evaluation coefficient value placed first, and the current video replaced by an alternative video with an even higher recommendation effect evaluation coefficient value.

9. A short video personalized recommendation system based on big data, used to implement the short video personalized recommendation method based on big data as described in any one of claims 1-8, characterized in that, include: Data acquisition module: used to monitor and collect user behavior data, viewed content data, and social interaction data in real time; User behavior analysis module: Used to analyze user browsing behavior and emotional behavior based on the collected user behavior data, and obtain user behavior monitoring coefficients; User activity analysis module: Based on the collected viewing content data, it analyzes the user's viewing activity at different time periods and obtains the activity distribution coefficient; User social interaction analysis module: Based on the collected social interaction data, it analyzes the social interactions between users and other users on the short video platform to obtain user similarity. User preference analysis module: This module analyzes user dwell time, number of times they watch a video repeatedly, sharing behavior data, and collection behavior data on a short video interface to obtain user preference identification coefficients. Personalized recommendation module: Used to calculate personalized recommendation index based on comprehensive analysis of user behavior monitoring coefficient, activity distribution coefficient, user similarity and user preference identification coefficient; The playback order of short videos is sorted based on a personalized recommendation index, and the playback order of short videos is recommended to users based on the sorting results. Recommendation effect evaluation module: Used to monitor the effect of personalized recommendations in real time, obtain the recommendation effect evaluation coefficient, and adjust the playback order of short videos in real time based on the recommendation effect evaluation coefficient.

Images