A video recommendation method and system based on big data
Through feedback adjustment and habit analysis of user history viewing information, feature labels are calculated and user groups of similar interests are divided, the problem that video quality does not match user interests in the existing video recommendation methods is solved, and the accuracy and user experience of video recommendation are improved.
Patent Information
- Application Number
- CN202411161477.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-08-23
AI Technical Summary
The existing video recommendation methods only use the video recommendation model for pushing, resulting in the video quality or video tags that do not meet the interests of most users, resulting in a decrease in user viewing interest and a decrease in experience.
By obtaining multiple historical viewing information of users, performing feedback adjustment processing and habit analysis, calculating user feature tags, dividing user groups with similar interests, and pushing priority videos based on user groups to improve the accuracy of recommendations and user experience.
It improves the intelligence and accuracy of video recommendations, ensures that the videos received by users meet their interests, and improves the user's viewing experience and recommendation quality.
Smart Images

Figure CN118690047B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of video recommendation, and in particular to a video recommendation method and system based on big data. Background Art
[0002] Personalized recommendations are based on the user's historical behavior to infer the user's interests and recommend a list of videos that the user is most likely to be interested in. Personalized recommendations comprehensively use the user's historical behavior, including rating, viewing, subscription, search, tagging, sharing and commenting, rather than just the current single browsing or viewing behavior, so it more accurately reflects the user's interests.
[0003] Existing improvements for video recommendation usually use a training set to train a video model, and use the video model to recommend videos to all users. For example, in a Chinese patent with invention publication number CN116486195A, a video recommendation model training method, a video recommendation method and a device are disclosed. This solution is to train the video recommendation model to be trained by using the feature sequence of the training sample to obtain a trained video recommendation model. However, due to the large number of new videos, the quality of the videos or video tags pushed by the video recommendation model often do not meet the interests of most users, making it difficult for most users to obtain high-quality works created by others in addition to their own interests. In view of this, it is necessary to improve the existing video recommendation method. Summary of the invention
[0004] The present invention aims to solve one of the technical problems in the prior art to at least a certain extent. By improving the video recommendation method, it is used to solve the problem that in the prior art, only a video recommendation model is used to push videos, resulting in the pushed video quality or video tags not meeting the interests of most users, thereby reducing users' viewing interest and deteriorating viewing experience.
[0005] To achieve the above objectives, in a first aspect, the present invention provides a video recommendation method based on big data, comprising:
[0006] Acquire multiple historical viewing information, perform feedback adjustment processing on the historical viewing information; calculate the historical viewing information after the feedback adjustment processing, and output the viewing progress; perform habit analysis processing on the viewing progress, and process the historical viewing information based on the habit analysis processing;
[0007] Calculate the remaining historical viewing information and output the number of tags, number of categories and total viewings; calculate the total viewings and number of tags and output the viewing ratio; analyze the viewing ratio and output the characteristic tags of each user based on the analysis results;
[0008] Calculate the interest similarity values between different users based on the feature tags, and divide the users into the same user group based on the interest similarity values;
[0009] Push videos to users based on the start time of users in the user group, monitor the push times and click times of the pushed videos, calculate the push times and click times, and output the click ratio; mark the pushed videos based on the click ratio, and output the priority videos;
[0010] Push to other users in the user group based on the priority video.
[0011] Further, obtaining a plurality of historical viewing information and performing feedback adjustment processing on the historical viewing information includes:
[0012] Acquire multiple historical viewing information of multiple users, wherein the historical viewing information includes video tags, video duration, viewing duration, and feedback information; the feedback information includes like feedback, dislike feedback, and empty feedback;
[0013] Performing feedback adjustment processing on the historical viewing information, the feedback adjustment processing comprising:
[0014] If the historical viewing information includes dislike feedback, the historical viewing information will be deleted; if the historical viewing information includes empty feedback, no processing will be performed;
[0015] If the historical viewing information includes like feedback, the historical viewing information is labeled with a high weight.
[0016] Further, the historical viewing information after the feedback adjustment processing is calculated and the viewing progress is output; the viewing progress is subjected to habit analysis processing, and the historical viewing information is processed based on the habit analysis processing, including:
[0017] Calculate the ratio of viewing time to video time in each remaining historical viewing information, and mark it as viewing progress;
[0018] Perform habit analysis on the viewing progress, and delete or do not process the historical viewing information based on the habit analysis result.
[0019] Furthermore, the habit analysis process includes:
[0020] Using the first value as the dividing interval, the viewing progress is divided from 0 to 1 into progress intervals of the second value;
[0021] For each user, calculate the number of viewing progress in each progress interval, marked as the interval number;
[0022] Create a progress histogram with the number of intervals as the vertical axis and the viewing progress as the horizontal axis, and mark the highest progress interval in the progress histogram as the habit interval;
[0023] comparing the left endpoint of the habit interval with the third value, and when the left endpoint of the habit interval is less than or equal to the third value, marking the left endpoint of the habit interval as the viewing threshold;
[0024] When the left endpoint of the habit interval is greater than the third value, a difference between the left endpoint of the habit interval and twice the first value is calculated and marked as a viewing threshold;
[0025] For each user, the historical viewing information whose viewing progress is less than the viewing threshold corresponding to the user is deleted; when the viewing progress is greater than or equal to the viewing threshold, no processing is performed.
[0026] Furthermore, the remaining historical viewing information is calculated, and the output tag count, category count, and total viewing count include:
[0027] For each user, calculate the number of occurrences of each video tag in the user's historical viewing information, marked as the tag count;
[0028] Calculate the number of different video tags in the user's historical viewing information, marked as the number of categories;
[0029] If any historical viewing information has a high-weight label, the first number will be counted additionally when calculating the number of labels;
[0030] Get the total number of historical viewing information remaining for the user, marked as the total number of views.
[0031] Furthermore, the total number of views and the number of tags are calculated, and the viewing ratio is output; the viewing ratio is analyzed, and the characteristic tags of each user are output based on the analysis results, including:
[0032] Calculate the ratio of each tag's number of views to the total number of views, and mark it as the viewing ratio;
[0033] Sort the viewing proportions in ascending order to obtain a proportion sequence;
[0034] The video label of the fourth value on the rightmost side of the ratio sequence is taken as the user's feature label, and the feature label is output.
[0035] Further, calculating interest similarity values between different users based on the feature tags, and grouping the users into the same user group based on the interest similarity values includes:
[0036] Two users a and b are randomly selected. The first set of user a's feature tags is defined as N(a), and the second set of user b's feature tags is defined as N(b).
[0037] The first set and the second set are calculated using a similarity calculation formula to output an interest similarity value; the similarity calculation formula is configured as: ; Where SX is the interest similarity value;
[0038] Users whose interest similarity values are greater than or equal to the similarity threshold are divided into the same group and marked as user group.
[0039] Furthermore, based on the start time of the user in the user group, the video is pushed to the user, the number of pushes and clicks of the pushed video is monitored, the number of pushes and clicks are calculated, and the click ratio is output; the pushed video is marked based on the click ratio, and the priority video is output, including:
[0040] The time when users in the same user group start the video software is monitored, and the users who start the video software in the first proportion are marked as trial users; the feature labels of the trial users are analyzed by using a deep learning model and the pushed video is output;
[0041] A push video is randomly selected. When the push video is pushed to the test user, the push count of the push video is increased by one. When the test user clicks on the push video, the viewing progress of the test user on the push video is monitored. When the viewing progress is less than the test user's habit threshold, the click is marked as an invalid click. When the viewing progress is greater than or equal to the test user's habit threshold, the click count of the test video is increased by one.
[0042] Get the push count and click count for each pushed video;
[0043] Calculate the ratio of the number of clicks to the number of pushes for each pushed video, marked as click ratio;
[0044] The recommended video corresponding to the fourth numerical click ratio is marked as a priority video.
[0045] Further, pushing the video to other users in the user group based on the priority video includes:
[0046] Mark all users except the trial users in the user group as push users;
[0047] Push priority videos to push users.
[0048] In a second aspect, the present invention further provides a video recommendation system based on big data, including a feature filtering module, a feature extraction module, an interest grouping module and a video push module; the feature filtering module is used to obtain a plurality of historical viewing information, perform feedback adjustment processing on the historical viewing information; calculate the historical viewing information after the feedback adjustment processing, and output the viewing progress; perform habit analysis processing on the viewing progress, and process the historical viewing information based on the habit analysis processing;
[0049] The feature extraction module calculates the remaining historical viewing information, outputs the number of tags, the number of categories and the total number of views; calculates the total number of views and the number of tags, and outputs the viewing ratio; analyzes the viewing ratio, and outputs the feature tags of each user based on the analysis results;
[0050] The interest grouping module is used to calculate interest similarity values between different users based on feature tags, and divide users into the same user group based on the interest similarity values;
[0051] The video push module is used to push videos to users based on the start-up time of users in the user group, monitor the number of pushes and clicks of the pushed videos, calculate the number of pushes and clicks, and output the click ratio; mark the pushed videos based on the click ratio and output the priority videos; and push the priority videos to other users in the user group.
[0052] Beneficial effects of the present invention: The present invention obtains multiple historical viewing information of the user, and successively performs feedback adjustment processing, calculation of viewing progress and habit analysis processing on the historical viewing information, thereby being able to delete irrelevant historical viewing information and leave historical viewing information that meets the viewing interests of the user. At the same time, by setting high-weighted labels for the video tags that the user likes in the feedback adjustment processing, the weight value of the favorite video tags in the feature tag calculation is increased. By analyzing from multiple aspects, the feature tags that meet the viewing interests of the user can be finally obtained, thereby improving the intelligence and accuracy in the feature analysis.
[0053] The present invention also calculates the similarity values of interests between users through feature tags, and divides the users into the same user group according to the similarity values. The users in each group have similar viewing interests. On this basis, the video recommendation model is used to recommend new videos to a small number of users in the group. By analyzing the interactive information of the recommended videos, the rankings of all pushed videos can be obtained. Finally, the highly ranked videos are pushed to other users in the user group. The clustering of people based on interests is achieved through grouping, and high-quality videos are pushed to each other in the crowd clustering. This can reduce the computational complexity of the video push model and improve the usage experience and viewing experience of most users. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] Figure 1 It is a functional block diagram of the system of the present invention.
[0055] Figure 2 The figure is a flow chart of the steps of the method of the present invention.
[0056] Figure 3 The figure is a schematic diagram of a progress histogram of a possible situation of the present invention.
[0057] Figure 4 The figure is a schematic diagram of a progress histogram of another possible situation of the present invention. DETAILED DESCRIPTION
[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0059] Example 1, please refer to Figure 1 As shown, the present invention provides a video recommendation system based on big data, including a feature filtering module, a feature extraction module, an interest grouping module and a video pushing module;
[0060] The feature filtering module is used to obtain multiple historical viewing information, perform feedback adjustment processing on the historical viewing information; calculate the historical viewing information after feedback adjustment processing, and output the viewing progress; perform habit analysis processing on the viewing progress, and process the historical viewing information based on the habit analysis processing;
[0061] The feature filtering module is configured with a feedback adjustment strategy, which includes:
[0062] Obtain multiple historical viewing information of multiple users, where the historical viewing information includes video tags, video duration, viewing duration, and feedback information; the feedback information includes like feedback, dislike feedback, and empty feedback;
[0063] Feedback adjustment processing is performed on the historical viewing information, and the feedback adjustment processing includes:
[0064] If the historical viewing information includes dislike feedback, the historical viewing information will be deleted; if the historical viewing information includes empty feedback, no processing will be performed;
[0065] If the historical viewing information includes like feedback, the historical viewing information is labeled with a high weight;
[0066] It should be noted that the purpose of feedback adjustment processing is to adjust the weight value of video tags when analyzing and calculating user interests according to user feedback information; for example, when a user watches a video tagged as sports, but submits dislike feedback after watching the video, then in the subsequent analysis of the user's viewing interests, the historical viewing information with dislike feedback will not be included in the analysis scope; similarly, if the user submits like feedback for a video tagged as sports, the historical viewing information will be marked with a high-weight label, which means that in the subsequent analysis of the user's viewing interests, videos tagged as sports have a higher weight; empty feedback means that the user has not submitted any feedback, and normal analysis steps are taken for videos with empty feedback.
[0067] The feature filtering module is also configured with a progress analysis strategy, which includes:
[0068] Calculate the ratio of viewing time to video time in each remaining historical viewing information, and mark it as viewing progress;
[0069] Perform habit analysis on viewing progress, and delete or not process historical viewing information based on the habit analysis result;
[0070] Habit analysis processing includes:
[0071] Using the first value as the dividing interval, the viewing progress is divided from 0 to 1 into progress intervals of the second value;
[0072] In a specific implementation, the first value is set to 0.1, so that the second value is 10;
[0073] It should be noted that the viewing progress from 0 to 1 represents the viewing progress from 0% to 100%, and the first value is set to 0.1, which means that each interval of 10% is divided into a progress interval;
[0074] For each user, calculate the number of viewing progress in each progress interval, marked as the interval number;
[0075] Create a progress histogram with the number of intervals as the vertical axis and the viewing progress as the horizontal axis, and mark the highest progress interval in the progress histogram as the habit interval;
[0076] comparing the left endpoint of the habit interval with the third value, and when the left endpoint of the habit interval is less than or equal to the third value, marking the left endpoint of the habit interval as the viewing threshold;
[0077] In a specific implementation, the third value is set to 0.5;
[0078] When the left endpoint of the habit interval is greater than the third value, a difference between the left endpoint of the habit interval and twice the first value is calculated and marked as a viewing threshold;
[0079] See also Figure 3 As shown, Figure 3 Take the remaining 500 historical viewing information of each user after the feedback adjustment strategy as an example. Figure 3 It can be seen that the user's habit interval is [0.8, 0.9], which means that for any video, the user will usually watch at least 80% of the video when he is interested. At this time, the left endpoint of the habit interval 0.8 is greater than 0.5, and the user's viewing threshold is calculated to be 0.8-2×0.1=0.6; this means that when the user's viewing progress of any video is less than 60%, it means that the user is not interested in the video, so it is deleted.
[0080] Due to different viewing habits of different users, Figure 3 Unlike other users, such as Figure 4 As shown, Figure 4 The corresponding user habit interval is [0.4, 0.5]. At this time, the left endpoint 0.4 of the habit interval is less than 0.5. The user's viewing threshold is directly set to 0.4, which means that when the user watches 40% or more of the video content, it is enough to show that the user is interested in the video.
[0081] For each user, the historical viewing information whose viewing progress is less than the viewing threshold corresponding to the user is deleted; when the viewing progress is greater than or equal to the viewing threshold, no processing is performed.
[0082] The feature extraction module calculates the remaining historical viewing information and outputs the number of tags, the number of categories, and the total number of views; calculates the total number of views and the number of tags and outputs the viewing ratio; analyzes the viewing ratio and outputs the feature tags of each user based on the analysis results;
[0083] The feature extraction module is configured with a feature calculation strategy, which includes:
[0084] For each user, calculate the number of occurrences of each video tag in the user's historical viewing information, marked as the tag count;
[0085] Calculate the number of different video tags in the user's historical viewing information, marked as the number of categories;
[0086] In particular, if any historical viewing information has a high-weight label, the first number is additionally counted when calculating the number of labels;
[0087] In the specific implementation, the first number is set to 4, and the additional first number means: if the number of labels is 50, when the historical viewing information with high-weight annotations is counted, the number of labels is updated to 50+5=55, and the purpose of feedback adjustment processing is achieved by additional calculation;
[0088] Get the total number of historical viewing information remaining for the user, marked as the total number of views.
[0089] The feature extraction module is also configured with a label analysis strategy, which includes:
[0090] Calculate the ratio of each tag's number of views to the total number of views, and mark it as the viewing ratio;
[0091] Sort the viewing proportions in ascending order to obtain a proportion sequence;
[0092] Take the video label of the fourth value on the rightmost side of the ratio sequence as the user's feature label, and output the feature label;
[0093] In a specific implementation, the fourth value is set to 0.2, and the setting of the fourth value is negatively correlated with the number of types. The larger the number of types, the smaller the fourth value is set, so that the number of feature tags of the user is stable within a certain range; taking the remaining 500 historical viewing information as an example, if there are 50 types of times in the 500 historical viewing information, there are 50 different video tags; taking the video tag with the number of types of the fourth value on the far right of the proportional sequence means taking 20% of the video tags on the far right of the proportional sequence, and 10 feature tags for each user can be obtained.
[0094] The interest grouping module is used to calculate the interest similarity values between different users based on feature tags, and divide the users into the same user group based on the interest similarity values;
[0095] The interest grouping module is configured with interest grouping strategies, which include:
[0096] Two users a and b are randomly selected. The first set of user a's feature tags is defined as N(a), and the second set of user b's feature tags is defined as N(b).
[0097] The first set and the second set are calculated using a similarity calculation formula to output interest similarity values; the similarity calculation formula is configured as: ; Where SX is the interest similarity value;
[0098] It should be noted that the numerator represents the number of intersections of the set N(a) and the set N(b), and the denominator represents the number of unions of the set N(a) and the set N(b).
[0099] Users whose interest similarity values are greater than a similarity threshold are divided into the same group and marked as a user group;
[0100] In specific implementation, the similarity threshold is set according to the number of users obtained. Ideally, when the number of users obtained is large enough, the similarity threshold can be set to 100%. This means that when the feature labels of two users are exactly the same, the two users are classified into the same user group. In other cases, the similarity threshold can be set to 90% to ensure that there is a sufficiently high similarity between the two users.
[0101] The video push module is used to push videos to users based on the start time of users in the user group, monitor the push times and click times of the pushed videos, calculate the push times and click times, and output the click ratio; mark the pushed videos based on the click ratio, and output the priority videos; push the priority videos to other users in the user group;
[0102] The video push module is configured with a test push strategy, which includes:
[0103] The time when users in the same user group start the video software is monitored, and the users who start the video software in the first proportion are marked as trial users; the feature labels of the trial users are analyzed by using a deep learning model and the pushed video is output;
[0104] In specific implementation, the first ratio is set to 40%. A possible implementation method of "marking users who start the video software in the first ratio as trial users" is: taking 12 o'clock and 0 o'clock every day as the dividing points, for example, there are 2000 users, and when the time reaches 12 o'clock and 0 o'clock on the same day, start to detect the first 800 users who start the video software in each user group, and mark these users as trial users, the purpose is to directly use the video recommendation model to push newly uploaded videos to them, judge the quality of the videos by collecting feedback information of the pushed videos, and sort the videos, among which the top ranked videos may indicate high-quality video content; push the top ranked videos to other users in other user groups in accordance with the viewing interests of the users, and the high video quality can provide users with a better viewing experience;
[0105] In addition, the time of the split point is changed at intervals to prevent users who are accustomed to watching video software at a specific time from being marked as trial users.
[0106] A push video is randomly selected. When the push video is pushed to the test user, the push count of the push video is increased by one. When the test user clicks on the push video, the viewing progress of the test user on the push video is monitored. When the viewing progress is less than the test user's habit threshold, the click is marked as an invalid click. When the viewing progress is greater than or equal to the test user's habit threshold, the click count of the test video is increased by one.
[0107] Get the push count and click count for each pushed video;
[0108] Calculate the ratio of the number of clicks to the number of pushes for each pushed video, marked as click ratio;
[0109] The recommended video corresponding to the fourth numerical click ratio is marked as a priority video.
[0110] The video push module is also configured with priority push strategies, which include:
[0111] Mark all users except the trial users in the user group as push users;
[0112] Push priority videos to push users.
[0113] Example 2, please refer to Figure 2 As shown, the present invention also provides a video recommendation method based on big data, including: step S1, obtaining multiple historical viewing information, and performing feedback adjustment processing on the historical viewing information; calculating the historical viewing information after the feedback adjustment processing, and outputting the viewing progress; performing habit analysis processing on the viewing progress, and processing the historical viewing information based on the habit analysis processing; step S1 also includes the following sub-steps:
[0114] Step S101, obtaining multiple historical viewing information of multiple users, where the historical viewing information includes video tags, video duration, viewing duration, and feedback information; the feedback information includes like feedback, dislike feedback, and empty feedback;
[0115] Step S102, performing feedback adjustment processing on the historical viewing information, step S102 also includes:
[0116] Step S1021, if the historical viewing information includes dislike feedback, the historical viewing information is deleted; if the historical viewing information includes empty feedback, no processing is performed;
[0117] Step S1022: if the historical viewing information includes like feedback, then a high-weight label is given to the historical viewing information;
[0118] Step S103, calculating the ratio of the viewing time to the video time in each remaining historical viewing information, and marking it as viewing progress;
[0119] Step S104, performing habit analysis on the viewing progress, and deleting or not processing the historical viewing information based on the habit analysis result;
[0120] Step S104 also includes:
[0121] Step S1041, dividing the viewing progress from 0 to 1 into progress intervals of second values with the first value as the dividing interval;
[0122] Step S1042, for each user, the number of viewing progress in each progress interval is calculated, and marked as the interval number;
[0123] Step S1043, creating a progress histogram with the interval number as the vertical axis and the viewing progress as the horizontal axis, marking the highest progress interval in the progress histogram as the habitual interval;
[0124] Step S1044, comparing the left endpoint of the habit interval with the third value, when the left endpoint of the habit interval is less than or equal to the third value, marking the left endpoint of the habit interval as the viewing threshold; when the left endpoint of the habit interval is greater than the third value, calculating the difference between the left endpoint of the habit interval and twice the first value, marking it as the viewing threshold;
[0125] Step S1045, for each user, the historical viewing information whose viewing progress is less than the viewing threshold corresponding to the user is deleted; when the viewing progress is greater than or equal to the viewing threshold, no processing is performed.
[0126] Step S2, calculate the remaining historical viewing information, output the number of tags, the number of categories and the total number of views; calculate the total number of views and the number of tags, and output the viewing ratio; analyze the viewing ratio, and output the characteristic tags of each user based on the analysis results; Step S2 also includes the following sub-steps:
[0127] Step S201, for each user, calculate the number of occurrences of each video tag in the user's historical viewing information, and mark it as the tag number;
[0128] Step S202, calculating the number of all different video tags in the user's historical viewing information, and marking them as the number of categories; in particular, if any of the historical viewing information has a high-weighted label, the first number is additionally counted when calculating the number of labels;
[0129] Step S203, obtaining the total number of the user's remaining historical viewing information, marked as the total viewing number;
[0130] Step S204, calculating the ratio of each tag's number of views to the total number of views, and marking it as the viewing ratio;
[0131] Step S205, sorting the viewing ratios in ascending order to obtain a ratio sequence;
[0132] Step S206, taking the video label of the fourth value of the rightmost value of the ratio sequence as the user's feature label, and outputting the feature label.
[0133] Step S3, calculating interest similarity values between different users based on the feature tags, and classifying the users into the same user group based on the interest similarity values; Step S3 also includes the following sub-steps:
[0134] Step S301: randomly select two users a and b, specify that the first set of feature tags of user a is N(a), and the second set of feature tags of user b is N(b);
[0135] Step S302, using a similarity calculation formula to calculate the first set and the second set, and outputting an interest similarity value; the similarity calculation formula is configured as: ; Where SX is the interest similarity value;
[0136] Step S303: users whose interest similarity values are greater than or equal to a similarity threshold are grouped into the same group and marked as a user group.
[0137] Step S4, based on the start time of the user in the user group, push the video to the user, monitor the push times and click times of the pushed video, calculate the push times and click times, and output the click ratio; mark the pushed video based on the click ratio, output the priority video; push the priority video to other users in the user group; Step S4 also includes the following sub-steps:
[0138] Step S4011, monitoring the time of starting the video software of users in the same user group, marking the users who start the video software in the first proportion as trial users; analyzing the feature labels of the trial users by using a deep learning model and outputting the pushed video;
[0139] Step S4012, randomly select a push video, and when the push video is pushed to the test user, the push count of the push video is increased by one; when the test user clicks on the push video, monitor the test user's viewing progress of the push video, and when the viewing progress is less than the test user's habit threshold, mark this click as an invalid click; when the viewing progress is greater than or equal to the test user's habit threshold, add one to the click count of the test video;
[0140] Step S4013, obtaining the number of pushes and clicks of each pushed video;
[0141] Step S4014, calculating the ratio of the number of clicks on each pushed video to the number of pushes, marked as click ratio;
[0142] Step S4015: Mark the recommended video corresponding to the fourth numerical click ratio as a priority video.
[0143] Step S402: Mark all users except the trial user in the user group as push users; and push priority videos to the push users.
[0144] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.
[0145] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0146] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.
Claims
1. A video recommendation method based on big data, characterized in that: include: Acquire multiple historical viewing information, and perform feedback adjustment processing on the historical viewing information; Calculate the historical viewing information after feedback adjustment and output the viewing progress; Perform habit analysis on viewing progress, and process historical viewing information based on the habit analysis; Calculate the remaining historical viewing information, output the number of tags, the number of categories, and the total number of views; calculate the total number of views and the number of tags, and output the viewing ratio; Analyze the viewing ratio and output the feature label of each user based on the analysis results; Calculate the interest similarity values between different users based on the feature tags, and divide the users into the same user group based on the interest similarity values; Push videos to users based on the start time of users in the user group, monitor the push times and click times of the pushed videos, calculate the push times and click times, and output the click ratio; Mark the pushed videos based on click ratio and output the priority videos; Pushing videos to other users in the user group based on priority videos; Acquiring multiple historical viewing information and performing feedback adjustment processing on the historical viewing information includes: Acquire multiple historical viewing information of multiple users, wherein the historical viewing information includes video tags, video duration, viewing duration, and feedback information; the feedback information includes like feedback, dislike feedback, and empty feedback; Performing feedback adjustment processing on the historical viewing information, the feedback adjustment processing comprising: If the historical viewing information includes dislike feedback, the historical viewing information will be deleted; if the historical viewing information includes empty feedback, no processing will be performed; If the historical viewing information includes like feedback, the historical viewing information is labeled with a high weight; The historical viewing information after feedback adjustment processing is calculated and the viewing progress is output; the viewing progress is subjected to habit analysis processing, and the historical viewing information is processed based on the habit analysis processing, including: Calculate the ratio of viewing time to video time in each remaining historical viewing information, and mark it as viewing progress; Perform habit analysis on viewing progress, and delete or not process historical viewing information based on the habit analysis result; The habit analysis process includes: Using the first value as the dividing interval, the viewing progress is divided from 0 to 1 into progress intervals of the second value; For each user, calculate the number of viewing progress in each progress interval, marked as the interval number; Create a progress histogram with the number of intervals as the vertical axis and the viewing progress as the horizontal axis, and mark the highest progress interval in the progress histogram as the habit interval; comparing the left endpoint of the habit interval with the third value, and when the left endpoint of the habit interval is less than or equal to the third value, marking the left endpoint of the habit interval as the viewing threshold; When the left endpoint of the habit interval is greater than the third value, a difference between the left endpoint of the habit interval and twice the first value is calculated and marked as a viewing threshold; For each user, the historical viewing information whose viewing progress is less than the viewing threshold corresponding to the user is deleted; when the viewing progress is greater than or equal to the viewing threshold, no processing is performed.
2. The video recommendation method based on big data according to claim 1, characterized in that: Calculate the remaining historical viewing information, and output the number of tags, types, and total views, including: For each user, calculate the number of occurrences of each video tag in the user's historical viewing information, marked as the tag count; Calculate the number of different video tags in the user's historical viewing information, marked as the number of categories; If any historical viewing information has a high-weight label, the first number will be counted additionally when calculating the number of labels; Get the total number of historical viewing information remaining for the user, marked as the total number of views.
3. The video recommendation method based on big data according to claim 2, characterized in that: Calculate the total number of views and the number of tags, and output the viewing ratio; The viewing ratio is analyzed, and the characteristic labels of each user are output based on the analysis results, including: Calculate the ratio of each tag's number of views to the total number of views, and mark it as the viewing ratio; Sort the viewing proportions in ascending order to obtain a proportion sequence; The video label of the fourth value on the rightmost side of the ratio sequence is taken as the user's feature label, and the feature label is output.
4. The video recommendation method based on big data according to claim 3, characterized in that: Calculating interest similarity values between different users based on feature tags and grouping users into the same user group based on interest similarity values includes: Two users a and b are randomly selected. The first set of user a's feature tags is defined as N(a), and the second set of user b's feature tags is defined as N(b). The first set and the second set are calculated using a similarity calculation formula to output an interest similarity value; the similarity calculation formula is configured as: ; Where SX is the interest similarity value; Users whose interest similarity values are greater than or equal to the similarity threshold are divided into the same group and marked as user group.
5. The video recommendation method based on big data according to claim 4, characterized in that: Push videos to users based on the start time of users in the user group, monitor the push times and click times of the pushed videos, calculate the push times and click times, and output the click ratio; Mark the pushed videos based on the click ratio, and output priority videos include: The time when users in the same user group start the video software is monitored, and the users who start the video software in the first proportion are marked as trial users; the feature labels of the trial users are analyzed by using a deep learning model and the pushed video is output; A push video is randomly selected. When the push video is pushed to the test user, the push count of the push video is increased by one. When the test user clicks on the push video, the viewing progress of the test user on the push video is monitored. When the viewing progress is less than the test user's habit threshold, the click is marked as an invalid click. When the viewing progress is greater than or equal to the test user's habit threshold, the click count of the test video is increased by one. Get the push count and click count for each pushed video; Calculate the ratio of the number of clicks to the number of pushes for each pushed video, marked as click ratio; The recommended video corresponding to the fourth numerical click ratio is marked as a priority video.
6. The video recommendation method based on big data according to claim 5, characterized in that: Pushing videos to other users in the user group based on priority videos includes: Mark all users except the trial users in the user group as push users; Push priority videos to push users.
7. A video recommendation system based on big data, used to implement a video recommendation method based on big data according to any one of claims 1 to 6, characterized in that: It includes feature filtering module, feature extraction module, interest grouping module and video push module; The feature filtering module is used to obtain multiple historical viewing information, perform feedback adjustment processing on the historical viewing information; calculate the historical viewing information after feedback adjustment processing and output the viewing progress; perform habit analysis processing on the viewing progress, and process the historical viewing information based on the habit analysis processing; The feature extraction module calculates the remaining historical viewing information, outputs the number of tags, the number of categories and the total number of views; calculates the total number of views and the number of tags, and outputs the viewing ratio; analyzes the viewing ratio, and outputs the feature tags of each user based on the analysis results; The interest grouping module is used to calculate interest similarity values between different users based on feature tags, and divide users into the same user group based on the interest similarity values; The video push module is used to push videos to users based on the start time of users in the user group, monitor the push times and click times of the pushed videos, calculate the push times and click times, and output the click ratio; Mark the pushed videos based on click ratio and output the priority videos; Push to other users in the user group based on the priority video.
