A short video recommendation method based on big data statistics

Abstract

The application relates to the technical field of big data statistics, and specifically discloses a short video recommendation method based on big data statistics, which comprises the following steps: S1: obtaining the total number of short videos watched by a user in a sampling period, and dividing the same short videos according to short video labels; S2: obtaining the number of times of logging in a platform and the single online duration in the sampling period, and calculating the mean value and the dependence coefficient of the single online duration; S3: dividing the sampling period into a front interval and a rear interval, and sequentially calculating the trend coefficients of the same short videos in the front interval and the rear interval; S4: correcting the trend coefficients to obtain corrected coefficients; and S5: calculating the trend difference values of the same short videos, and analyzing the watching trend of the same short videos by the user based on the trend difference values. The application can analyze the preferences of the user and predict future trends, establish a flexible and real-time response mechanism for the change of the preferences of the user, realize the long-term development of the platform, and maintain the market competitiveness.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data statistics, and particularly relates to a short video recommendation method based on big data statistics. BACKGROUND

[0002] In today's digital era, short video platforms have become a highly influential and popular industry.

[0003] Under such a background, how to tailor and push short videos that users are really interested in according to their preferences has become the core and top priority in the operation process of short video platforms. Each user is a unique individual with specific interests, browsing habits and consumption psychology. However, the user's preferences are not a fixed pattern, and the user's interests will change over time and with changes in the environment. If the short video platform cannot perceive the dynamic changes in user preferences and still push according to the original push plan, it will put the platform in a difficult position. On the one hand, the content received by the user will be more and more inconsistent with their actual needs at the moment, resulting in a sharp decline in their satisfaction with the platform. On the other hand, when users cannot obtain satisfactory content on the platform, they are likely to turn to other more attractive platforms. SUMMARY

[0004] The purpose of the present application is to provide a short video recommendation method based on big data statistics to solve the above technical problems.

[0005] The purpose of the present application can be achieved by the following technical solutions:

[0006] A short video recommendation method based on big data statistics, comprising the following steps:

[0007] S1: obtaining the total number N of short videos watched by a user within a preset sampling period T, and dividing short videos with the same short video label into similar short videos, wherein the short video label refers to a classification identifier automatically generated by the platform according to the video content and user interest;

[0008] S2: obtaining the number M of times the user logs in to the platform within the sampling period T and the single online duration t, and calculating the mean value t of the single online duration ave ;

[0009] calculating the dependence coefficient of the user , wherein t s refers to the mean value of the single online duration of all users in the platform, M ave refers to the mean value of the number of times all users log in to the platform;

[0010] S3: dividing the sampling period T into a front interval [T now -T, Tnow -T / 2] and the later interval [T now -T / 2, T now ], wherein T now denotes the current time node;

[0011] Based on the former interval [T now -T, T now ] and the later interval [T now -T / 2, T now ], the trend coefficient q1 i and q2 i of the i-th short video of the same type are calculated.

[0012] S4: The trend coefficients q1 i and q2 i are corrected to obtain the corrected coefficients Q1 i = λ1 × Z1 × q1 i and Q2 i = λ2 × Z2 × q2 i , wherein λ1 and λ2 respectively represent the preset first weight coefficient and the second weight coefficient, 0 < λ1 < λ2, and Z1 and Z2 respectively represent the dependence coefficients of the former interval and the later interval.

[0013] S5: The trend difference value QC i = Q2 i - Q1 i of the i-th short video of the same type is calculated, and the change of the viewing trend of the i-th short video of the same type by the user is analyzed based on the trend difference value QC i .

[0014] As a further scheme of the present application, in the step S5, the method for analyzing the change of the viewing trend of the i-th short video of the same type by the user based on the trend difference value QC i includes:

[0015] If the trend difference value QC i ≥ QC sta , it represents that the viewing trend of the i-th short video of the same type by the user is getting higher, and the platform pushing amount of the i-th short video of the same type is increased by QC i / QC sta times;

[0016] If the trend difference value 0 ≤ QC i < QC sta , it represents that the change of the viewing trend of the i-th short video of the same type by the user is small, and the platform pushes according to the original scheme;

[0017] If the trend difference value QC i < 0, it represents that the viewing trend of the i-th short video of the same type by the user is getting lower, and the i-th short video of the same type is stopped pushing at this time.

[0018] wherein, QC sta represents preset standard trend difference and QC sta > 0.

[0019] As a further scheme of the present application: in the step S1, a preset viewing quantity threshold N min is set, and if the total viewing quantity N of the short video of the user within a preset sampling period T is less than N min , subsequent operations and analysis of the viewing trend are stopped.

[0020] As a further scheme of the present application: in the step S1, if the short video contains multiple short video tags, the short video is divided into the same type of short video corresponding to the main short video tag when the same type of short video is divided.

[0021] As a further scheme of the present application: in the step S2, the time interval of the user logging into the platform within the sampling period T is obtained, and if the time interval is less than a preset minimum time threshold t min , it is recorded as a single user login behavior;

[0022] If the single online duration t is less than the minimum time threshold t min , the current login behavior is not counted into the number of times M of logging into the platform.

[0023] As a further scheme of the present application: in the step S3, based on the front interval [T now -T, T now -T / 2] and the rear interval [T now -T / 2, T now ], the method for calculating the trend coefficients q1 i and q2 i of the i-th same type of short video of the user includes:

[0024] Obtaining the playback feedback of the user on the i-th same type of short video within the front interval [T now -T, T now -T / 2], the playback feedback including the number of likes A i and the number of collections S i of the user;

[0025] Calculating the trend coefficients , wherein N f represents the viewing quantity of the short video of the user within the front interval [T now -T, T now -T / 2], γ1, γ2, γ3 respectively represent preset first feedback coefficient, second feedback coefficient, and third feedback coefficient, and γ1< γ2< γ3;

[0026] According to the above steps, the trend coefficient q2 of the corresponding interval [T now -T / 2, T now ] is calculated. i .

[0027] As a further scheme of the application: in the step S4, the short video complained by the user is marked, and the complaint rate TS=I / I of the same type of short video is calculated. all Wherein, I represents the number of the same type of short video marked, and I all represents the total number of short videos in the same type of short video, if the complaint rate TS≥10%, the same type of short video is excluded and does not participate in subsequent calculation.

[0028] As a further scheme of the application: the same short video only participates in the calculation process once, and if it is repeatedly pushed, it will be excluded.

[0029] The beneficial effects of the application are: first, in order to predict the trend change of the user's short-term short video preference behavior, the recent short video viewing record is needed, so that the change of the user's viewing behavior in a short time can be analyzed, thereby providing data support for subsequent prediction behavior, improving the prediction accuracy, if the selected data is relatively old, due to the large difference in user's love interest in different periods, the prediction accuracy is low, and even misjudgment is caused.

[0030] It should be noted here that in the application, the remarks of the short video tags are not required, and the platform is mainly divided.

[0031] Then, the dependence coefficient of the user on the platform is judged according to the number of times the user logs in the platform and the online duration of each single login, according to the formula, the data of the entire platform user is taken as a reference, if the user logs in the platform more times, the dependence coefficient is larger, and when the single online duration is also higher than the average value, the dependence coefficient value is further increased, in summary, the dependence coefficient reflects the user's love degree for the platform.

[0032] Then, the sampling period is divided into a front interval and a rear interval, the purpose of division is to facilitate the analysis of the trend of the user's browsing short video, and the purpose of dividing the front interval and the rear interval here is to avoid the influence of irrelevant factors on the experimental accuracy, and the purpose of taking the current time node as one of the endpoints is to obtain the latest data and improve the prediction accuracy.

[0033] Then the trend coefficients of different short videos in the two intervals are calculated. According to the formula, since the number of views is usually higher than the number of likes and the number of collections, in order to balance the data and reduce the error, the square root of the number of views is taken to reduce the gap, while the number of likes and the number of collections can reflect the user's favorite degree of the same type of short video. The higher the value is, the more the user likes this type of short video, so the corresponding weight needs to be appropriately increased.

[0034] Then the trend coefficient is corrected. The purpose of correction is to comprehensively consider the influence of the platform and the short video on the user. The dependence coefficient represents the user's dependence on the platform, and the trend coefficient represents the attraction of the short video to the user. Therefore, the two need to be analyzed comprehensively. Because the second interval is closer to the current time node, the second weight coefficient is greater than the first weight coefficient, so as to focus on the importance of the current data. Finally, the trend difference is analyzed according to the obtained trend difference to analyze the change of the user's viewing trend of different short videos of the same type. In summary, the present application establishes a flexible and real-time response to user preference changes by analyzing user preferences and predicting future trends. The long-term development of the platform is realized, and the market competitiveness is maintained. BRIEF DESCRIPTION OF DRAWINGS

[0035] The present application will be further described below in conjunction with the accompanying drawings.

[0036] Figure 1 It is a flowchart of a short video recommendation method based on big data statistics. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0038] Please refer to Figure 1 The present application is a short video recommendation method based on big data statistics, which comprises the following steps:

[0039] S1: Obtain the total number of short video views N of the user in a preset sampling period T, and divide the short videos with the same short video label into the same short video. The short video label refers to the classification mark automatically generated by the platform according to the video content and user interest;

[0040] S2: Obtain the number of times M that the user logs in the platform in the sampling period T and the single online duration t, and calculate the mean value t of the single online duration ave ;

[0041] Calculating the dependence coefficient of the user , wherein t s denotes the mean value of the online duration of all users in the platform, M ave denotes the mean value of the login times of all users in the platform;

[0042] S3: Divide the sampling period T into the front interval [T now -T, T now -T / 2] and the rear interval [T now -T / 2, T now ], wherein T now denotes the current time node;

[0043] Based on the front interval [T now -T, T now -T / 2] and the rear interval [T now -T / 2, T now ], the trend coefficients q1 i and q2 i of the user for the i-th short video of the same type are calculated;

[0044] S4: The trend coefficients q1 i and q2 i are corrected to obtain the corrected coefficients Q1 i =λ1×Z1×q1 i and Q2 i =λ2×Z2×q2 i , wherein λ1 and λ2 respectively represent the first preset weight coefficient and the second weight coefficient, 0<λ1<λ2, Z1 and Z2 respectively represent the dependence coefficients of the front interval and the rear interval;

[0045] S5: The trend difference value QC i =Q2 i -Q1 i of the i-th short video of the same type is calculated, and the change in the viewing trend of the user for the i-th short video of the same type is analyzed based on the trend difference value QC i .

[0046] It should be noted that, in order to accurately predict the trend change of the user's short video preference behavior in the short term, the user's short video viewing record in the recent period of time is first obtained. The reason why "recent" is emphasized is that the user's interest and hobby are not immutable, especially in the field of short videos with rich content and rapid updates, the user's preference often presents the characteristics of dynamic and rapid evolution.

[0047] Specifically, through detailed analysis of user viewing behavior in a short period of time, a lot of valuable information can be captured. The prediction method based on recent data can significantly improve the accuracy of prediction and be closer to the real needs and preferences of users. If data with a long time selection is used for analysis and prediction, the risk will be greatly increased. Because over time, the difference in user's favorite interest in different periods can be very large. This change of interest is universal and difficult to predict, if still rely on old data for judgment, it is easy to lead to a significant reduction in prediction accuracy, and even serious misjudgment.

[0048] It needs to be particularly pointed out that in the related design and implementation process of the present application, no specific requirements are proposed for the remarks of short video tags. Mainly based on the classification and division standard of the platform itself for short video processing. Taking the platform division as the main basis, it can better adapt to the rules and ecology of different platforms, and also avoids the subjectivity and inconsistency that may be introduced by artificially setting tag remarks, thereby further improving the reliability and stability of the entire system.

[0049] In the process of in-depth exploration of user behavior patterns, the "dependence coefficient" index is further introduced to measure the user's stickiness and loyalty to the platform. Specifically, the calculation of this coefficient combines two core behavior characteristics of users: one is the total number of logins to the platform, and the other is the single online duration after each login. The data of all users on the platform is taken as the reference system to standardize and quantify the individual differences. Generally, the more times a user logs in to the platform, the higher the participation degree, and accordingly, the dependence coefficient will also increase. More precisely, when the single online duration of this user exceeds the average level of all users, this additional enthusiasm will be specially considered, thereby further increasing the value of the dependence coefficient. This double-dimensional consideration makes the dependence coefficient be able to comprehensively and sensitively reflect the real favorite degree and use stickiness of users to the platform.

[0050] To analyze the user's behavior trend in detail, the entire sampling period is divided into two equal parts, the front interval and the back interval. There are two reasons for this equalization strategy. On the one hand, it effectively isolates irrelevant variables that may interfere with the experimental results, ensuring that the analysis process is not affected by seasonal fluctuations, accidents or other external factors, thus maintaining the purity and accuracy of the experiment. On the other hand, by setting the current time node as the endpoint of one of the intervals, the latest data points can be directly captured. These real-time information is crucial for capturing market dynamics and predicting future trends. The inclusion of the latest data greatly improves the timeliness and accuracy of the prediction results, enabling a more acute insight into the changing trajectory of user interest and timely adjustment of strategies to adapt to the rapid changes in the market. By monitoring and analyzing user behavior data in these two intervals, we can clearly see how user interest evolves over time, which content types or themes are more popular in a particular time period, and thus accurately target the target audience to achieve efficient content distribution and maximize value.

[0051] After completing the division of the sampling period, the next key step is to calculate the trend coefficients of various short videos in the two intervals. From the design logic of the formula, since the number of views is often significantly higher than the number of likes and collections in real-world scenarios, this natural difference in data size can easily lead to biased analysis results and affect the accuracy of decision-making if not addressed. Therefore, to effectively balance the weights between the three key indicators and minimize the errors caused by different data sizes, the number of views is processed by taking the square root, reducing the numerical gap between the number of views and the other two indicators, making them comparable on the same scale.

[0052] At the same time, the number of likes and collections, as a direct reflection of users' active expression of preference, contains users' deep recognition of the short video content. The higher the values of these two indicators, the stronger the interest and affection of the user group towards this specific type of short video. Given this, when constructing the trend coefficient, a higher weight is given to the number of likes and collections to more accurately capture user preference. However, relying solely on the original trend coefficient is not enough to fully reflect the complex and changing user behavior patterns. Therefore, a correction mechanism is introduced to comprehensively consider the dual influence of platform characteristics and short video content on users. Among them, the dependence coefficient serves as a measure of user's overall stickiness to the platform, revealing the depth of user interaction with the platform; while the trend coefficient focuses on the attractiveness of specific short video content to users. Together, they form a three-dimensional framework for evaluating the dynamics of user interest.

[0053] In the correction process, particular attention is paid to the time dimension. Considering that the data of the later interval is closer to the current time node and can better reflect the latest interest change trend of the user, the second weight coefficient is greater than the first weight coefficient, ensuring that the analysis result can focus more on the recent user behavior data, thereby improving the timeliness and accuracy of the prediction.

[0054] Finally, by differentiating the corrected trend coefficient, the dynamic evolution of the user's viewing trend of different categories of short videos can be understood. Not only can it identify which types of videos are rising to become new hotspots, but also can timely perceive which once popular themes may be gradually losing their appeal. In summary, the present application realizes accurate insight into user preferences and effective prediction of future trends by constructing a refined analysis system based on user behavior data.

[0055] In another preferred embodiment of the present application, the trend difference QC i The method for analyzing the change of the user's viewing trend of the i-th type of short video includes:

[0056] If the trend difference QC i ≥ QC sta , it represents that the user's viewing trend of the i-th type of short video is increasing, and the platform pushing amount of the i-th type of short video is increased by QC i / QC sta times;

[0057] If the trend difference 0 ≤ QC i < QC sta , it represents that the user's viewing trend of the i-th type of short video changes less, and the platform pushes according to the original plan;

[0058] If the trend difference QC i < 0, it represents that the user's viewing trend of the i-th type of short video is decreasing, and the i-th type of short video is stopped pushing at this time;

[0059] Wherein, QC sta represents a preset standard trend difference and QC sta > 0.

[0060] In another preferred embodiment of the present application, a viewing quantity threshold N min is preset, and if the total number of short video views of the user within the preset sampling period T is less than N min , the subsequent operation and analysis of the viewing trend are stopped.

[0061] It can be understood that when the total number of short video views within the preset sampling period T is less than the set threshold N min, which means that the user's viewing behavior is less active during this period, and cannot provide sufficient effective and representative sample data. In this case, in order to ensure the reliability and effectiveness of the analysis results, the system will automatically stop the subsequent operation process to avoid false conclusions and invalid decisions caused by insufficient data or poor quality.

[0062] In another preferred embodiment of the application, if the short video contains multiple short video tags, the short video is divided into the same short video corresponding to the short video main tag.

[0063] It should be noted that this approach aims to ensure that each short video can be reasonably classified according to its most prominent and core theme attributes, thereby ensuring relatively high homogeneity and relevance within the same short video group, facilitating subsequent unified analysis, recommendation and other related business processing work based on category characteristics, making the entire short video management system more orderly, efficient and consistent with the content logic architecture.

[0064] In another preferred embodiment of the application, the time interval of the user logging into the platform within the sampling period T is obtained, and if the time interval is less than a preset minimum time threshold t min , it is recorded as a user login behavior;

[0065] If the single online duration t is less than the minimum time threshold t min , this login behavior is not counted in the number of times M of logging into the platform.

[0066] It should be noted that the specific time interval information of the user logging into the platform within a given sampling period T is obtained. When the time interval between the adjacent two login operations is less than the pre-set minimum time threshold t min , the system will recognize this as an effective user login behavior; at the same time, for each independent login session, if it is detected that the single online duration t does not meet the standard requirement of the minimum time threshold t min , it is determined that this login does not meet the conditions of effective login, and accordingly, it is not included in the total number of times M of logging into the platform within the statistical range. To provide a reliable basis for subsequent in-depth analysis and decision-making based on real user behavior, effectively exclude invalid login interference caused by short-term false touch or non-subjective will, thereby improving the quality of the overall data and the accuracy of the analysis results.

[0067] In another preferred embodiment of the application, based on the front interval [T now -T, T now -T / 2] and the rear interval [T now -T / 2, T now ], the trend coefficient q1i and q2 i The method comprises the following steps:

[0068] Obtaining the playing feedback of the user on the i-th short video of the same type in the previous interval [T now -T, T now -T / 2], wherein the playing feedback comprises the number of likes A i and the number of collections S i of the user.

[0069] Calculating the trend coefficient q1 , wherein N f represents the number of views of the short video of the user in the previous interval [T now -T, T now -T / 2], and γ1, γ2, γ3 respectively represent the first preset feedback coefficient, the second feedback coefficient and the third feedback coefficient, and γ1< γ2< γ3.

[0070] According to the above steps, the trend coefficient q2 i corresponding to the subsequent interval [T now -T / 2, T now ] is calculated.

[0071] In another preferred embodiment of the present application, the short video complained by the user is marked, and the complaint rate TS=I / I all of the short video of the same type is calculated, wherein I represents the number of the marked short video of the same type, and I all represents the total number of the short video of the same type, and if the complaint rate TS≥10%, the short video of the same type is excluded and does not participate in the subsequent calculation.

[0072] Notably, by marking the short video complained by the user and calculating the complaint rate of the short video of the same type, the system can automatically identify the type of video that is not popular with the user. When the complaint rate of a certain type of video exceeds 10%, it will be directly excluded from the subsequent recommendation pool, so as to avoid pushing the controversial content to the user.

[0073] In another preferred embodiment of the present application, the same short video only participates in the calculation process once, and if it is repeatedly pushed, it will be excluded.

[0074] Notably, the principle of participating in a complete calculation only once for each short video is followed. Specifically, when encountering a situation where the same short video is pushed multiple times, the system will automatically start the intelligent recognition mechanism, and the subsequent repeated short video will be excluded from the calculation queue. This measure aims to ensure that each short video is only considered once in the calculation process, avoiding data redundancy and interference caused by repeated pushing, thereby ensuring the accuracy and objectivity of the calculation results, enabling the entire analysis model to operate based on unique and effective short video samples, and thus providing more accurate and reliable content recommendation and trend prediction services for users, effectively maintaining the rigor and scientific nature of system data processing.

[0075] The above describes one embodiment of the present application in detail, but the content described is only a preferred embodiment of the present application and cannot be considered to limit the scope of the implementation of the present application. Any equivalent changes and improvements made in accordance with the scope of the present application should still be within the scope of the present application.

Claims

1.A short video recommendation method based on big data statistics, characterized in that, Comprising the following steps: S1: obtaining the total number N of short videos watched by a user within a preset sampling period T, and dividing short videos with the same short video label into the same type of short videos, wherein the short video label refers to a classification identifier automatically generated by a platform according to video content and user interest; S2: Obtain the number of logins M of the user on the platform within a sampling period T and the online duration t each time, and calculate the average online duration t each time ave ; Calculating a dependency coefficient of a user where t s M refers to the mean value of the online duration of all users in the platform ave M refers to the mean value of the login times of all users in the platform S3: divide the sampling period T into a front interval [T now -T, T now ] and a rear interval [T now -T / 2, T now ], wherein T now denotes the current time node; Based on the front interval [T now -T, T now ] and the rear interval [T now -T / 2, T now ], the trend coefficients q1 i and q2 i of the user for the i-th short video of the same type are calculated. S4: correcting the trend coefficients q1 i and q2 i to obtain correction coefficients Q1 i = λ1 × Z1 × q1 i and Q2 i = λ2 × Z2 × q2 i , wherein λ1 and λ2 represent preset first and second weight coefficients respectively, 0 < λ1 < λ2, and Z1 and Z2 represent dependency coefficients corresponding to the preceding interval and the following interval respectively. S5: Calculate the trend difference value QC of the i-th short video of the same type i = Q2 i - Q1 i , based on the trend difference value QC i Analyze the change of the user's viewing trend of the i-th short video of the same type. 2.The short video recommendation method based on big data statistics of claim 1, wherein, In the step S5, the trend difference QC i The method for analyzing the change of the user's viewing trend of the i-th same type short video comprises: If the trend difference QC i ≥ QC sta , it represents that the trend of the user watching the i-th short video of the same type is higher, and the platform pushing amount of the i-th short video of the same type is increased by QC i / QC sta times; If the trend difference 0≤QC i <QC sta , it represents that the trend change of the user watching the i-th same short video is small, and the platform pushes according to the original scheme; If the trend difference QC i <0, representing that the user's viewing trend of the i-th short video of the same type decreases, and at this time, the pushing of the i-th short video of the same type is stopped; wherein QC sta represents a preset standard trend difference value and QC sta > 0. 3.The short video recommendation method based on big data statistics of claim 1, characterized in that, In the step S1, a viewing quantity threshold N is preset min If the total viewing quantity N of the short video of the user in the preset sampling period T is less than N min , the subsequent operation and the analysis of the viewing trend are stopped. 4.The short video recommendation method based on big data statistics of claim 1, characterized in that, In the step S1, if a short video contains multiple short video labels, when dividing the same type of short videos, the short video is divided into the same type of short videos corresponding to the main label of the short video. 5.The short video recommendation method based on big data statistics of claim 1, wherein, In the step S2, the time interval of the user logging into the platform within a sampling period T is obtained, and if the time interval is less than a preset minimum time threshold t min , it is recorded as a user login behavior. If the single online duration t is less than the minimum time threshold t min The current login behavior is not counted in the number of login platforms M. 6.The short video recommendation method based on big data statistics of claim 1, wherein, In step S3, based on the previous interval [T] now -T, T now -T / 2] and the subsequent interval [T now -T / 2,T now Calculate the trend coefficient q1 of the user's preference for the i-th type of short video. i and q2 i The methods include: Obtaining the playing feedback of the i-th same short video in the previous interval [T now -T, T now -T / 2] by the user, wherein the playing feedback comprises the number of likes A i and the number of collections S i of the user. Computing a trend coefficient where N f represents the number of short video views of the user in the preceding interval [T now -T, T now -T / 2], and γ1, γ2, γ3 represent preset first, second, and third feedback coefficients respectively, and γ1<γ2<γ3. According to the above steps, the trend coefficient q2 now -T / 2, T now corresponding to the interval [T i . 7.The short video recommendation method based on big data statistics of claim 1, wherein, In the step S4, the short video complained by the user is marked, and the complaint rate TS=I / I of the same short video is calculated all Wherein, I represents the number of the same short video marked, I all represents the total number of short videos in the same short video, if the complaint rate TS≥10%, the same short video is excluded and does not participate in subsequent calculation. 8.The short video recommendation method based on big data statistics of claim 1, wherein, The same short video only participates in one calculation process, and if it is repeatedly pushed, it will be excluded.

Images