Short video recommendation method based on artificial intelligence
By analyzing audience switching behavior using artificial intelligence and clustering algorithms, and calculating the degree of aversion to short videos, this technology solves the problem of lacking audience likes and dislikes measurement in existing technologies, and achieves more accurate video recommendation and filtering.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 湖北雅派文化传播有限公司
- Filing Date
- 2023-12-27
- Publication Date
- 2026-05-01
AI Technical Summary
Existing short video recommendation methods lack the ability to measure audience preferences, making it difficult to accurately select popular videos from those that are currently trending.
Using an artificial intelligence-based approach, a clustering algorithm is employed to analyze viewers' switching behavior, extract perceived aversion points in videos, calculate the degree of aversion, and use the degree of aversion to perform reverse ranking and video recommendation.
It achieves accurate data processing based on popularity, segments out areas that users may dislike, improves the accuracy of video recommendations and the measurement of audience interest, and enhances the effectiveness of video selection and recommendation.
Smart Images
Figure CN121958604A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, specifically a short video recommendation method based on artificial intelligence. Background Technology
[0002] Currently, with the rapid development of the internet, the popularity of short videos has injected new vitality into the online world. Generally, short video recommendation methods rely on metrics such as tags and completion rates to form recommendation mechanisms. The popularity of short videos is already well-calculated, for example, through common metrics like likes and daily views. However, research on measuring audience engagement with short videos is still lacking. Therefore, a solution is needed to evaluate whether short videos are popular with users, selectively filtering out popular works from the high-popularity videos. Under this guidance, short video content needs to consider not only popularity but also audience engagement. Summary of the Invention
[0003] To address the aforementioned technical problems, the present invention aims to provide an artificial intelligence-based short video recommendation method. Based on tagged works and existing popularity, this method designs a way to find points of viewer dislike, introduces a clustering algorithm to better extract target information features, and ultimately finds a recommendation strategy.
[0004] This invention provides an artificial intelligence-based short video recommendation method, comprising the following steps:
[0005] S1. Select a preset number of short videos from the database according to the tags, and group short videos with the same tags into one category;
[0006] S2. Calculate the popularity of short videos with the same tag;
[0007] S3. Initialize all short videos to be processed, and re-normalize and recalibrate the videos by percentage.
[0008] S4. Select the browsing history of 100 short videos with the same tag;
[0009] S5. For a single short video, query the time when you exit the short video or switch to a new short video each time, and record the switching ratio p.
[0010] S6. Use the k-means algorithm to classify the switching ratio p and obtain the locations of perceived aversion.
[0011] S7. Calculate the overall level of viewer aversion to the video (y) by obtaining the perceived aversion points;
[0012] S8. Recommend short videos to viewers by reversing the order of aversion level y.
[0013] Preferably, step S2 specifically includes:
[0014] 2a counts the number of short videos with the same tag, denoted as m, the number of likes for a single short video, denoted as n, and the number of times it has been played, denoted as j.
[0015] 2b
[0016] Among them, the number of short videos is m, the number of likes is n, and the number of times it has been played is j;
[0017] 2c uses the same method to iterate through all videos under the same tag, obtaining all popularity values r, and denoted as r1, r2, r3...r m .
[0018] Preferably, in step S3, the initialization process is: to normalize the video duration by a percentage and delete the last 10% of the video.
[0019] Preferably, step S4 specifically includes:
[0020] 4a When the number is greater than 100, simply select 100 short videos with the same tag from the browsing history.
[0021] 4b When the quantity is less than 100, a default number is randomly selected for compensation.
[0022] Preferably, step S5 specifically includes:
[0023] 5a Find the time point when the user cuts out of the short video and mark it as a percentage. If the user cuts out when the video is n% played, set the parameter p as the switching ratio and assign it the value 0.n.
[0024] Similarly, 5b completes the query of the switching time points of 100 browsing records, denoted as p1, p2...p 100 .
[0025] Preferably, step S6 specifically includes:
[0026] 6a uses the k-means algorithm, sets the k value to 2, classifies 100 switching ratios p into two groups of data, performs an arithmetic mean on each group of data, and obtains two perceived aversion points, denoted as a1 and a2.
[0027] 6b continues to set the k value to 3, classifies the 100 switching ratios p into 3 groups of data, discards the group with the fewest number of data, keeps two groups and performs an arithmetic average to obtain two perceived aversion points, denoted as b1 and b2.
[0028] 6c Then set the k value to 4, classify the 100 switching ratios p into 4 groups of data. Similarly, in order to keep only two groups of data, sort and discard the two groups with the fewest switching ratios p, and perform an arithmetic average to obtain two perceived aversion points, denoted as c1 and c2.
[0029] The perceived aversion points obtained from 6D are a1, a2, b1, b2, c1, and c2. These perceived aversion points have values in the range of [0, 1], and represent the time points when viewers like to switch short videos.
[0030] Preferably, in step S7, the formula for calculating the degree of aversion y is as follows:
[0031]
[0032] Where a1, a2, b1, b2, c1, and c2 are points of perceived aversion, and r is the current popularity of the video.
[0033] Preferably, the method further includes step S9:
[0034] By iterating through different short videos under the same tag, we obtain the level of aversion y for each short video, denoted as y1, y2...y. 100 After filtering these parameters, a weighted average is taken to obtain the overall label aversion level y. all ; Determine the degree of aversion to the label y all The reciprocal of the value is used as the interest index for that tag, and short videos are recommended to viewers based on the interest index Q.
[0035] Preferably, the interest index Q is calculated using the following formula:
[0036]
[0037] Compared with the prior art, the beneficial effects of the present invention are:
[0038] 1. This invention, based on tagging, analyzes viewers' interests in short videos. Its clustering method pinpoints specific switching points, ensuring data processing accuracy and effectively segmenting areas that might cause user aversion. Building upon the clustering algorithm, different grouping methods and selection mechanisms preserve data authenticity. Furthermore, it derives the aversion level 'y' for a given short video, which can then be applied to video filtering and recommendation.
[0039] 2. By iterating through all videos and using normalization methods such as filtering and weighted averaging, the aversion level y can be obtained. all Taking the reciprocal of , we get the final interest index Q, which can be used to measure the value of a tag. Attached Figure Description
[0040] Figure 1 This is a flowchart illustrating the principle of the present invention. Detailed Implementation
[0041] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0042] like Figure 1 As shown, this invention provides a short video recommendation method based on artificial intelligence, including the following steps:
[0043] Step 1: Extraction and initialization of short video information.
[0044] 1. Select a preset number of short videos from the database according to tags, and group short videos with the same tag into one category.
[0045] 2. Calculate the popularity of short videos with the same tag separately.
[0046] There are many methods for calculating heat; one such method is given below.
[0047] 2a counts the number of short videos with the same tag, denoted as m, the number of likes for a single short video, denoted as n, and the number of times it has been played, denoted as j.
[0048] 2b
[0049] Among them, the number of short videos is m, the number of likes is n, and the number of times it has been played is j.
[0050] 2c This allows us to easily calculate the current video's popularity r. Using the same method, we iterate through all videos with the same tag, obtaining a large number of popularity values r, which are then denoted as r1, r2, r3...r m .
[0051] There is more than one way to calculate popularity (r); this is just one simple method. Popularity (r) is one of the basic parameters for measuring the influence of a short video. Calculating the popularity in advance allows subsequent evaluation to be based on the existing popularity (r) and further calculate the audience's interest in the video.
[0052] Step 2: Collect feature information from short videos and perform clustering processing to calculate aversion values.
[0053] 3. Initialize all short videos to be processed (percentage normalize the video duration and remove the last 10%, because the aversion index is calculated, and when a video finishes playing 100%, it will inevitably switch to the next video, making it impossible to determine whether the switch is a user-initiated action. Therefore, when using this method to evaluate user aversion points, the end of the video is cut off, and the last 10% is removed here). Then, normalize again and recalibrate the percentage of the video.
[0054] 4. To ensure processing efficiency, we selected the browsing history of 100 short videos with the same tag.
[0055] 4a When the quantity is greater than 100, random selection is sufficient.
[0056] 4b When the quantity is less than 100, a default number is randomly selected for compensation.
[0057] Ultimately, 100 samples were selected.
[0058] 5. For a single short video, query the time when you exit the short video or switch to a new short video each time, and record the switching ratio p.
[0059] 5a This statistical method will identify a large number of times when users switch out of short videos, and mark them as percentages. If a user switches out when the video is n% played, the parameter p is set as the switching ratio and assigned a value of 0.n.
[0060] Similarly, 5b, the switching points at which 100 browsing records are queried are denoted as p1, p2...p 100 .
[0061] Switching ratio p is an important parameter for measuring when a user exits a video. It means that at this point in the video, the user feels aversion to the earlier part of the video, leading to boredom and the user switching to another video. The short video point with a switching ratio p is also the point where the user feels aversion.
[0062] 6. Furthermore, the k-means algorithm is used to classify the switching ratio p.
[0063] In order to identify as many user-perceived aversion points as possible, and to ensure the effectiveness of these aversion points, the following clustering method was designed.
[0064] 6a uses the k-means algorithm, sets the value of k to 2, classifies the 100 switching ratios p into two groups of data, and then performs an arithmetic mean on each group of data to obtain two perceived aversion points, denoted as a1 and a2.
[0065] 6b continues to set the k value to 3, classifies the 100 switching ratios p into 3 groups of data, discards the group with the fewest data, keeps two groups and performs an arithmetic average, also obtaining two perceptual aversion points, denoted as b1 and b2.
[0066] 6c At this point, the k value is set to 4, and the 100 switching ratios p are classified into 4 groups of data. Similarly, in order to keep only two groups of data, the two groups with the fewest switching ratios p need to be sorted and discarded, and the arithmetic average is performed to obtain two perceived aversion points, denoted as c1 and c2.
[0067] The perceived aversion points obtained from 6D are a1, a2, b1, b2, c1, and c2. These perceived aversion points have values in the range of [0, 1], and represent the time points when viewers like to switch short videos.
[0068] 7. Therefore, the overall level of aversion (y) of the audience to the video can be calculated by obtaining the points of perceived aversion.
[0069] The formula for calculating the degree of aversion y is as follows:
[0070]
[0071] Where a1, a2, b1, b2, c1, and c2 are points of perceived aversion, and r is the current popularity of the video.
[0072] The aversion level y is calculated in this way because the sample size decreases with each of the three clustering groups. a1 and a2 are derived from clustering all 100 points; b1 and b2 use a smaller sample size because some points are discarded; c1 and c2 may have discarded even more. Therefore, it can be seen that the aversion level y is weighted in a stepwise manner for the clustering analysis of the samples.
[0073] 8. The level of aversion (y) can help us directly see how interested the audience is in the short video. The higher the value of the level of aversion (y), the lower the likelihood that the audience will be interested in the short video. By reversing the order of aversion (y) and then recommending it to the audience, we can further improve the promotional ability of short videos under this tag.
[0074] 9. Furthermore, by extensively traversing different short videos under the same tag, the level of aversion y for each short video can be obtained, denoted as y1, y2…y. 100 After filtering these parameters and taking a weighted average, the overall label aversion level y can be obtained. all .
[0075] Finally, to obtain the interest index of the tag, we can directly take the tag's aversion level y. all The reciprocal of the value is used as the interest index for that tag, i.e.
[0076] In summary, this AI-based short video recommendation method analyzes viewers' interests in short videos based on tagging. Its clustering method to locate specific switching points not only ensures the accuracy of data processing but also effectively segments out areas that users may dislike.
[0077] Based on clustering algorithms, different grouping methods and selection mechanisms are used to preserve the authenticity of the data. Furthermore, the aversion level y for a short video is derived. The aversion level y can be applied to video screening.
[0078] Finally, by iterating through all videos and using normalization methods such as filtering and weighted averaging, the aversion level y can be obtained. all Taking the reciprocal of , we get the final interest index Q, which can be used to measure the value of a tag.
[0079] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. A short video recommendation method based on artificial intelligence, characterized in that, Includes the following steps: S1. Select a preset number of short videos from the database according to the tags, and group short videos with the same tags into one category; S2. Calculate the popularity of short videos with the same tag; S3. Initialize all short videos to be processed, and re-normalize and recalibrate the videos by percentage. S4. Select the browsing history of 100 short videos with the same tag; S5. For a single short video, query the time when you exit the short video or switch to a new short video each time, and record the switching ratio p. S6. Use the k-means algorithm to classify the switching ratio p and obtain the locations of perceived aversion. S7. Calculate the overall level of viewer aversion to the video (y) by obtaining the perceived aversion points; S8. Recommend short videos to viewers by reversing the order of aversion level y.
2. The short video recommendation method based on artificial intelligence according to claim 1, characterized in that, Step S2 is as follows: 2a counts the number of short videos with the same tag, denoted as m, the number of likes for a single short video, denoted as n, and the number of times it has been played, denoted as j. 2b Calculation of Heat Among them, the number of short videos is m, the number of likes is n, and the number of times it has been played is j; 2c uses the same method to iterate through all videos under the same tag, obtaining all popularity values r, and denoting them as r1, r2, r3...r m .
3. The short video recommendation method based on artificial intelligence according to claim 2, characterized in that, In step S3, the initialization process is as follows: the video duration is normalized by percentage and the last 10% of the video is deleted.
4. The short video recommendation method based on artificial intelligence according to claim 3, characterized in that, Step S4 is as follows: 4a When the number is greater than 100, simply select 100 short videos with the same tag from the browsing history. 4b When the quantity is less than 100, a default number is randomly selected for compensation.
5. The short video recommendation method based on artificial intelligence according to claim 4, characterized in that, Step S5 is as follows: 5a Find the time point when the user cuts out of the short video and mark it as a percentage. If the user cuts out when the video is n% played, set the parameter p as the switching ratio and assign it the value 0.n. Similarly, 5b completes the query of the switching time points of 100 browsing records, denoted as p1, p2...p 100 .
6. The short video recommendation method based on artificial intelligence according to claim 5, characterized in that, Step S6 is as follows: 6a uses the k-means algorithm, sets the k value to 2, classifies 100 switching ratios p into two groups of data, performs an arithmetic mean on each group of data, and obtains two perceived aversion points, denoted as a1 and a2. 6b continues to set the k value to 3, classifies the 100 switching ratios p into 3 groups of data, discards the group with the fewest number of data, keeps two groups and performs an arithmetic average to obtain two perceived aversion points, denoted as b1 and b2. 6c Then set the k value to 4, classify the 100 switching ratios p into 4 groups of data. Similarly, in order to keep only two groups of data, sort and discard the two groups with the fewest switching ratios p, and perform an arithmetic average to obtain two perceived aversion points, denoted as c1 and c2. The perceived aversion points obtained from 6D are a1, a2, b1, b2, c1, and c2. These perceived aversion points have values in the range of [0, 1], and represent the time points when viewers like to switch short videos.
7. The short video recommendation method based on artificial intelligence according to claim 6, characterized in that, In step S7, the formula for calculating the degree of aversion y is as follows: Where a1, a2, b1, b2, c1, and c2 are points of perceived aversion, and r is the current popularity of the video.
8. The short video recommendation method based on artificial intelligence according to claim 7, characterized in that, It also includes step S9: By iterating through different short videos under the same tag, we obtain the level of aversion y for each short video, denoted as y1, y2...y. 100 After filtering these parameters, a weighted average is taken to obtain the overall label aversion level y. all ; Determine the degree of aversion to the label y all The reciprocal of the value is used as the interest index for that tag, and short videos are recommended to viewers based on the interest index Q.
9. The short video recommendation method based on artificial intelligence according to claim 8, characterized in that, The interest index Q is calculated using the following formula: