Video recommendation method and device, video playing method and device, electronic equipment and storage medium

Abstract

The invention provides a video recommendation method and device, a video playing method and device, an electronic device and a storage medium, the video recommendation method is applied to a server side and comprises the steps that a recommendation request length sent by a client side is received, and the recommendation request comprises a user identifier; obtaining an interest portrait of the user based on the user identifier; based on the interest portrait of the user, recommended videos and preview segments of the videos are determined from a video content library, and each video comprises a plurality of preview segments of different styles; and sending the determined preview segment of the video to the client, so that the client displays the preview segment of the video. According to the method and the device, the preview segments of the recommended video are determined based on the interest portraits of the users, so that differentiated pushing of the video is realized, the preview segments of different styles in the video are recommended to different users, the recommendation efficiency of the video is improved, and the potential interest and experience of the users in browsing the video are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of video processing, and in particular to a video recommendation method, a video playing method, a device, an electronic device and a readable storage medium. BACKGROUND

[0002] In related technologies, content discovery and distribution of a long video platform usually adopts a "shelving type" or a "poster wall type" or a "direct search type", that is, video posters are arranged in a grid shape on a platform homepage, and are classified into fixed or semi-fixed categories such as "movies", "television series", "latest online", "popular ranking", etc. Users need to select by browsing these arranged video posters, reading the text introduction or clicking to watch the trailer. This way, it is easy to cause user decision fatigue and low video discovery efficiency, that is, there are too many intermediate links from seeing the poster of the long video to actually deciding to watch, such as clicking to view details -> reading the introduction -> watching the trailer -> returning, etc. Similarly, this way, the traffic and exposure opportunities of the platform are highly concentrated on a few top "blockbuster" films and television series, while the inventory old dramas and non-hot dramas with good reputation cannot be effectively exposed, which seriously wastes the content library resources.

[0003] With the rise of short videos, the "immersive sliding" mode of short videos adopts a revolutionary "single-column vertical screen, up-down sliding switching" infinite information flow mode. This mode greatly reduces the content discovery cost through instant playback, providing strong user stickiness. However, if the immersive sliding mode of short videos is directly copied to long videos, it will not work. This is because a film or television series lasting for tens of minutes or hours may have its exciting part at any time point, not at the beginning. Simply cutting the first 30 seconds of the video as a preview clip may be a dull prologue or setup, which cannot attract users and may even have a negative effect. Therefore, there is an essential difference between long videos and short videos in terms of content rhythm and value density.

[0004] Therefore, how to recommend long videos to users and improve the playing efficiency of long videos is a problem to be solved at present. SUMMARY

[0005] The present application provides a video recommendation method, a video playing method, a device, an electronic device and a readable storage medium to at least solve the problem of user decision fatigue and low video playing efficiency caused by too many intermediate links of previewing videos in related technologies. The technical solutions of the present application are as follows: According to a first aspect of an embodiment of the present application, a video recommendation method is provided, comprising: receiving a recommendation request sent by a client, the recommendation request comprising a user identifier; obtaining an interest profile of the user based on the user identifier; determining recommended videos and preview clips of the videos from a video content library based on the user's interest profile, each of the videos comprising a plurality of preview clips of different styles; sending the determined preview clips of the videos to the client to enable the client to display the preview clips of the videos.

[0006] Optionally, the determining the recommended videos and the preview clips of the videos from the video content library based on the user's interest profile comprises: selecting, based on the user's interest profile, a video with the highest matching degree to the interest profile from the video content library within a time corresponding to a front term of a recommendation ratio in a video recommendation strategy, or selecting a video with a matching degree lower than the highest matching degree from the video content library within a time corresponding to a rear term of the recommendation ratio in the video recommendation strategy, as the recommended video.

[0007] Optionally, the determining the preview clips of the videos based on the user's interest profile comprises: selecting, from the recommended video, a preview clip with the highest matching degree to the user's interest profile as the preview clip of the recommended video.

[0008] Optionally, the method further comprises: obtaining a user group feature of the user; the selecting, from the recommended video, the preview clip with the highest matching degree to the user's interest profile as the preview clip of the recommended video comprises: selecting, from the recommended video, a preview clip with the highest matching degree to a combination of the user group feature of the user and the user's interest profile as the preview clip of the recommended video.

[0009] Optionally, after the sending the determined preview clips of the videos to the client, the method further comprises: obtaining user behavior data of the client playing the preview clips of the videos; determining a feedback intensity of the user to the video content according to the user behavior data; updating a label weight in the user's interest profile based on the feedback intensity, and / or adjusting a recommendation ratio of a subsequent recommendation of the video based on the feedback intensity and the user behavior data.

[0010] Optionally, before the receiving the recommendation request sent by the client, the method further comprises: pre-creating a plurality of preview clips of different styles of each video in the video content library in the following manner: acquire a multi-dimensional heat map of each video in a plurality of videos; select a plurality of preview clips corresponding to a time axis of the video based on the multi-dimensional heat map of each video, and mark the plurality of preview clips; store the video marked with the plurality of preview clips into a video content library.

[0011] According to a second aspect of the embodiments of the present application, a video playing method is provided, comprising: receiving a preview clip of a recommended video from a server side, wherein the preview clip of the recommended video is selected from a plurality of preview clips of the video based on a user portrait; playing the recommended preview clip of the video.

[0012] Optionally, the method further comprises: In the process of playing the recommended preview clip of the video, in response to a first gesture instruction of the preview clip, jumping to a playing page of a video corresponding to the preview clip to play the video, or in response to a second gesture instruction of the preview clip, jumping to a preview clip of a next video.

[0013] Optionally, before receiving the preview clip of the recommended video from the server side, the method further comprises: sending a recommendation request of a video to the server side, wherein the recommendation request comprises a user identifier.

[0014] Optionally, after playing the preview clip of the video, the method further comprises: recording user behavior data of playing the preview clip of the video, wherein the user behavior data comprises a time length of playing the preview clip of the video and user interaction data; sending the user behavior data to the server side, so that the server side determines a feedback intensity of the user to the video content based on the user behavior data, and updates a label weight in the user portrait based on the feedback intensity and / or adjusts a recommendation ratio of subsequently recommending the video based on the feedback intensity and the user behavior data.

[0015] According to a third aspect of the embodiments of the present application, a video recommendation device is provided, which is applied to a server side and comprises: a receiving module configured to receive a recommendation request sent by a client side, wherein the recommendation request comprises a user identifier; a first acquiring module configured to acquire a user interest portrait based on the user identifier; a determining module configured to determine a recommended video and a preview clip of the video from a video content library based on the user interest portrait, wherein each of the video comprises a plurality of preview clips of different styles. The recommendation module is configured to send the preview segment of the video determined by the determination module to the client, so that the client displays the preview segment of the video.

[0016] Optionally, the determination module comprises: The video selection module is configured to select, based on the interest profile of the user, a video with the highest matching degree to the interest profile from a video content library within a time corresponding to the front term of the recommendation ratio in the video recommendation strategy, or select a video with a matching degree lower than the highest matching degree from the video content library within a time corresponding to the rear term of the recommendation ratio in the video recommendation strategy, as a recommended video.

[0017] Optionally, the determination module further comprises: The recommended segment determination module is configured to select, from the recommended video, a preview segment with the highest matching degree to the interest profile of the user as a preview segment of the recommended video.

[0018] Optionally, the apparatus further comprises: The second acquisition module is configured to acquire user group features of the user. The recommendation module is further configured to select, from the recommended video, a preview segment with the highest matching degree to the combination of the user group features of the user and the interest profile of the user as a preview segment of the recommended video.

[0019] Optionally, the apparatus further comprises: The third acquisition module is configured to acquire user behavior data of the client playing the preview segment of the video after the recommendation module sends the preview segment of the video determined by the determination module to the client. The intensity determination module is configured to determine the feedback intensity of the user to the video content according to the user behavior data. The update module is configured to update the label weight in the interest profile of the user based on the feedback intensity.

[0020] Optionally, the apparatus further comprises a creation module configured to pre-create a plurality of preview segments of different styles for each video in the video content library before the reception module receives the recommendation request sent by the client.

[0021] Optionally, the creation module comprises: The fourth acquisition module is configured to acquire a multi-dimensional heat map of each video in a plurality of videos. The selection and marking module is configured to select and mark a plurality of preview segments of different styles corresponding to the playing time axis of the video based on the multi-dimensional heat map of each video. A first storage module is configured to store the video marked with the preview segments of the plurality of different styles into a video content library.

[0022] According to a fourth aspect of the embodiments of the present application, a video playing device is provided, which is applied to a client and comprises: A receiving module is configured to receive the preview segment of the recommended video sent by the server, wherein the preview segment of the recommended video is selected from the preview segments of the plurality of different styles of the video based on the user portrait. A first playing module is configured to play the recommended preview segment of the video.

[0023] Optionally, the device further comprises a second playing module and / or a jumping module, wherein: The second playing module is configured to, in the process of playing the recommended preview segment of the video by the first playing module, jump to a playing page of the video corresponding to the preview segment and play the video in response to a first gesture instruction of the preview segment. The jumping module is configured to, in the process of playing the recommended preview segment of the video by the first playing module, jump to a preview segment of a next video in response to a second gesture instruction of the preview segment.

[0024] Optionally, the device further comprises: A first sending module is configured to send a video recommendation request to the server before the receiving module receives the preview segment of the recommended video from the server, wherein the recommendation request comprises a user identifier.

[0025] Optionally, the device further comprises: A behavior data determining module is configured to record user behavior data of playing at least one preview segment of the video after the first playing module plays the preview segment of the video, wherein the user behavior data comprises a time length of playing the preview segment of the video and user interaction data. A second sending module is configured to send the user behavior data to the server, so that the server determines a feedback intensity of the user to the video content according to the user behavior data and updates a label weight in the user portrait based on the feedback intensity.

[0026] According to a fifth aspect of the embodiments of the present application, an electronic device is provided, which comprises: A processor; A memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the video recommendation method or the video playing method.

[0027] According to a sixth aspect of the embodiments of the present application, a readable storage medium is provided, when instructions in the readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the video recommendation method or the video playing method as described above.

[0028] According to a seventh aspect of the embodiments of the present application, a computer program product is provided, including a computer program or instructions, when the computer program or instructions are executed by a processor of an electronic device, the video recommendation method or the video playing method as described above is implemented.

[0029] The technical solutions provided by the embodiments of the present application at least have the following beneficial effects: In the embodiments of the present application, a recommendation request sent by a client is received, the recommendation request including a user identifier; an interest profile of the user is obtained based on the user identifier; a recommended video and a preview segment of the video are determined from a video content library based on the interest profile of the user, each of the videos including a plurality of preview segments of different styles; and the determined preview segment of the video is sent to the client, so that the client displays the preview segment of the video. That is, in the embodiments of the present application, the preview segment of the recommended video is determined based on the interest profile of the user, thereby realizing differentiated pushing of the video, and the preview segments of different styles in the video are recommended to different users, which not only improves the recommendation efficiency of the video, but also improves the potential interest and viewing time of the user in watching the video.

[0030] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0031] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the present application, and do not constitute an improper limitation on the present application. In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0032] Figure 1 is a flowchart of a video recommendation method provided by the embodiments of the present application.

[0033] Figure 2 is a flowchart of a video playing method provided by the embodiments of the present application.

[0034] Figure 3 is a flowchart of an application example of a video recommendation method provided by an embodiment of the present application.

[0035] Figure 4 is a block diagram of a video recommendation device provided by an embodiment of the present application.

[0036] Figure 5 is another block diagram of a video recommendation device provided by an embodiment of the present application.

[0037] Figure 6 is a block diagram of a video playback device provided by an embodiment of the present application.

[0038] Figure 7 is a schematic diagram of a video recommendation system provided by an embodiment of the present application.

[0039] Figure 8 is a block diagram of an electronic device provided by an embodiment of the present application.

[0040] Figure 9 is a block diagram of a device for video recommendation or video playback provided by an embodiment of the present application. DETAILED DESCRIPTION

[0041] In order for those skilled in the art to better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings.

[0042] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. The implementation described in the following exemplary embodiments does not represent all implementations consistent with the present application. Rather, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0043] Please refer to Figure 1 is a flowchart of a video recommendation method provided by an embodiment of the present application, as shown in Figure 1 the method is applied to a server side, comprising the following steps: Step 101: receiving a recommendation request sent by a client, the recommendation request comprising a user identifier.

[0044] Step 102: obtaining an interest profile of the user based on the user identifier.

[0045] Step 103: determining recommended videos and preview segments of the videos from a video content library based on the user's interest profile, each of the videos including a plurality of preview segments of different styles.

[0046] Step 104: sending the determined preview segments of the videos to the client to enable the client to display the preview segments of the videos.

[0047] In the embodiment of the present application, the preview segments of the recommended videos are determined based on the user's interest profile, thereby realizing differentiated pushing of the videos and recommending the preview segments of different styles in the video to different users, which not only improves the video recommendation efficiency, but also improves the potential interest and experience of the user in browsing the video.

[0048] The video recommendation method described in the present application can be applied to terminals, servers, etc., and is not limited herein. The terminal implementation device can be a smart phone, a notebook computer, a tablet computer, a desktop computer, a personal digital assistant (PDA, Personal Digital Assistant), a wearable device, etc. The server can be a standalone server, a server cluster, or a server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, intermediate services, domain name services, security services, content distribution networks, or big data and artificial intelligence platforms, etc., and is not limited herein.

[0049] The specific implementation steps of the video recommendation method provided in the embodiment of the present application will be described in detail below. Figure 1

[0050] In step 101, a recommendation request sent by a client is received, and the recommendation request includes a user identifier.

[0051] In this step, first, the user opens an application on the client, enters a sliding discovery interface, and triggers a video recommendation request to the server side (also referred to as a server or a service side, etc., hereinafter) through the sliding discovery interface, and the recommendation request includes a user identifier. The recommendation request can be a first recommendation request or can not be a first recommendation request, and the present embodiment is not limited thereto. The recommendation request includes a user identifier, and in actual applications, other parameters can also be included, and the present embodiment is not limited thereto.

[0052] The video in the present embodiment can include a movie, a TV series, a documentary, or a long variety show, etc. These videos can be referred to as long videos, and are not limited thereto, and other types of videos can also be included, and the present embodiment is not limited thereto.

[0053] ​It should be noted that the sliding discovery interface in the embodiment is self-defined, and is a full-screen, single-column, and infinite sliding information stream interface similar to a short video. A user can switch and watch preview clips (such as Highlight Clips) of different videos by using a simple left / right or up / down sliding gesture, which can also be referred to as a sliding immersive playback interface. In this embodiment, the preview clip can be a climax clip, a Highlight Clip, a wonderful clip, a highlight clip, a highlight clip, a selected clip, a famous scene clip, or a clip with the most comments, such as a clip with the most comments on character portrayal, plot structure, dialogue, photography, music, editing, art design, and the like.

[0054] Then, the server end receives a recommendation request in which a user identifier is included, by using a dynamic recommendation engine, in response to a request for video recommendation by the user through the sliding discovery interface.

[0055] In step 102, an interest profile of the user is obtained based on the user identifier.

[0056] In this step, the server end can obtain the interest profile of the user from a user profile library, wherein different users correspond to different interest profiles. The user profile library includes interest profiles of a large number of users. The server end updates the interest profile of the corresponding user in real time or at a fixed time according to the user behavior data of the user watching videos.

[0057] That is, in this embodiment, the server end constructs and updates the interest profile of the user according to the user behavior data of the user watching the preview clips of the historical videos (long videos or short videos). The user behavior data can include the watching time (i.e., the stay duration) of watching the preview clips and interaction data (as key indicators for measuring the interest degree of the user). For example, if the watching time is less than 2 seconds, it is considered that the user is not interested, and if the watching time is greater than 8 seconds, it is considered that the user is very interested. The interest profile of the user is updated in real time or periodically based on this. It should be noted that the 2 seconds and 8 seconds in this embodiment are only used as examples, and are not limited thereto in actual applications.

[0058] In step 103, recommended videos and preview clips of the videos are determined from a video content library based on the interest profile of the user, and each of the videos includes a plurality of preview clips of different styles.

[0059] This step includes: selecting a corresponding video from the video content library as a recommended video based on the interest profile of the user and a recommendation ratio in a video recommendation strategy; and selecting a preview clip with the highest matching degree with the interest profile of the user from the recommended video as a preview clip of the recommended video.

[0060] According to the video recommendation strategy, a video corresponding to a type is selected from the video content library as a recommended video based on the user's interest profile. Specifically, the method comprises the following steps: According to the video recommendation strategy, a video corresponding to a type is selected from the video content library as a recommended video based on the user's interest profile. Specifically, the method comprises the following steps:

[0061] In this embodiment, the recommendation ratio in the video recommendation strategy is pre-set. In this example, the video recommendation strategy uses the recommendation ratio of "exploitation and exploration" as an example. For example, the recommendation ratio is 8:2. Of course, it can also be set to 7:3 or 6:4, and the like. This embodiment is not limited. According to the recommendation ratio in the recommendation strategy, the type of the recommended video is determined.

[0062] In this embodiment, the recommendation ratio in the recommendation strategy of "exploitation and exploration" is taken as an example of a dynamic ratio of 8:2. The value 8 in the recommendation ratio represents that 80% of the time is used to push similar or related content that the user may highly like according to the user's interest profile. This embodiment is called an exploitation strategy. The value 2 in the recommendation ratio represents that 20% of the time is used to push some generalized or new content to the user to explore the user's potential interest and prevent information cocooning as an exploration strategy. Other ratios are similar and will not be described here.

[0063] Information cocooning refers to the fact that people in a network environment, due to the influence of algorithm recommendation and personalized services, only contact and accept information that they are interested in, and ignore or reject different or opposite information, thereby forming a self-closed information environment.

[0064] Therefore, in this embodiment, if the recommendation ratio is taken as an example of 8:2, if it is "exploitation", a video highly related to the user's interest profile is searched from the video content library as a recommended video. If it is "exploration", new content that the user may not have contacted, i.e., a new type of video, is selected from the video content library as a recommended video.

[0065] It should be noted that the recommended video includes a video episode ID. One video episode ID can include multiple preview segments of different content styles, which can also be referred to as different style preview segments.

[0066] Among the recommended videos, one preview segment with the highest matching degree to the user's interest profile is selected as the preview segment of the recommended video.

[0067] In this step, based on the selected recommended video, one preview segment with the highest matching degree to the user's interest profile is selected from the selected video as the preview segment of the video. For example, if the user's interest profile is inclined to action type, a preview segment including action type in the video is selected as the recommended preview segment, etc. That is, one preview segment with the highest matching degree to the user's interest profile is selected from different style preview segments of the recommended video as the preview segment of the recommended video.

[0068] In step 104, the determined preview segment of the video is sent to the client to display the preview segment of the video on the client.

[0069] In this embodiment, the preview segment with the highest matching degree to the interest profile in the video can be sent to the client through the sliding discovery interface. The preview segment sent to the client can be the first preview segment according to the play time axis of the video, or a certain preview segment in the middle, or a certain preview segment at the end, etc. This embodiment is not limited, so that the client can play the preview segment of the recommended video through the sliding discovery interface.

[0070] In the embodiment of the application, a recommendation request sent by the client is received, the recommendation request including a user identifier; the interest profile of the user is obtained based on the user identifier; recommended videos and preview segments of the videos are determined from a video content library based on the interest profile of the user, each of the videos having multiple preview segments of different styles; and the determined preview segments of the videos are sent to the client to display the preview segments of the videos on the client. That is, the preview segments of the recommended videos are determined based on the interest profile of the user, so that the differentiated push of the videos is realized, and different users are recommended with preview segments of different styles in the videos, which not only improves the recommendation efficiency of the videos, but also improves the potential interest and experience of the users in browsing the videos.

[0071] Optionally, in another embodiment, the method further includes obtaining user group characteristics of the user; and the step of selecting one preview segment with the highest matching degree to the interest profile of the user from the recommended videos as the preview segment of the recommended video includes selecting one preview segment with the highest matching degree to the combination of the user group characteristics of the user and the interest profile of the user from the recommended videos as the preview segment of the recommended video.

[0072] That is, in this embodiment, the user group characteristics of the user are combined with the interest profile of the user, and the most matching preview segment is selected from the determined video. Based on the interest profile of the user combined with the current group characteristics of the user (such as the 18-25 male group, or the 30-40 female group, or the "photography enthusiast" group, the action enthusiast group, the romance drama enthusiast group, etc.), a preview segment that is most likely to attract attention is matched, and the preview segment is recommended to the user.

[0073] In the embodiment of the application, based on the user group characteristics of the user combined with the interest profile of the user, the most matching preview segment is selected from the video, which realizes the purpose of differentiated pushing for different groups of people, such as pushing different preview segments of the same video to different users, such as pushing the fighting scene in the video to action movie enthusiasts, and pushing the emotional entanglement scene in the video to romance drama enthusiasts, so as to realize precise and efficient distribution of thousands of people with different faces. Not only the recommendation efficiency of the video is improved, but also the potential interest and viewing time of the user watching the video are improved.

[0074] Optionally, in another embodiment, the embodiment is based on the above-mentioned embodiment, after sending the determined preview segment of the video to the client, the method can further include: obtaining user behavior data after the client plays the preview segment of the video; determining the feedback intensity of the user to the video content according to the user behavior data; updating the label weight in the interest profile of the user based on the feedback intensity; and / or adjusting the recommendation ratio of recommending the video based on the feedback intensity and the user behavior data. The specific updating process can be performed by a test system, etc., and the present embodiment is not limited.

[0075] In this embodiment, when the user plays the preview segment of the recommended video through the sliding discovery interface, that is, when the user watches the preview segment, the system records the watching time (i.e. the stay time) of the user and the interaction data (such as likes, comments, etc.) of the user. The recorded stay time and interaction information are collectively referred to as user behavior data in this embodiment, and the user behavior data is sent to the server end for corresponding processing by the server end.

[0076] The server end analyzes the stay time in the received user behavior data, and quantifies it as the feedback intensity to the content, such as determining negative feedback if the stay time is less than 2 seconds, strong positive feedback if the stay time is greater than 8 seconds, etc.

[0077] Based on feedback intensity: on one hand, updating the user portrait of the user based on the feedback intensity to improve the accuracy of the next recommendation; on the other hand, adjusting the recommendation ratio of the subsequent recommendation of the video based on the feedback intensity and the user behavior data, that is, testing the feedback intensity and the user behavior data using an A / B test system to improve the optimization operation of the A / B test system on the recommendation ratio of the video.

[0078] Optionally, in this embodiment, when it is detected that the user performs a left-right sliding operation again, the user identifier of the user requesting video recommendation through the sliding discovery interface is continuously acquired, that is, a new round of recommendation-feedback cycle is started, and a closed loop of continuous learning and optimization is formed.

[0079] It should be noted that the short-term interaction of the user with each preview segment (such as a highlight segment Highlight Clip) of the video on the sliding discovery interface is referred to as micro-interaction (Micro-interaction). Each left sliding (negative feedback), dwell (positive feedback), and click-to-watch-full (strong positive feedback) behavior of the user is captured by the system in real time. Among them, the captured effective dwell time is the most critical and direct quantitative indicator for measuring the instant attraction of the user to a preview segment (i.e., specific content) of the video.

[0080] Correspondingly, macro-distribution refers to how the platform efficiently pushes the tens of thousands of videos in the video content library to a large number of users. Based on this, the embodiment of the application provides a high-speed and self-optimizing closed loop feedback process from micro-interaction to macro-distribution. The purpose is different from the traditional way that the user "guesses" whether the video is good or not by static text and images. In the embodiment of the application, a preview segment that is most likely to attract the user is directly recommended to the user, so that the user can decide to continue watching the preview segment of the video and then watch the video within a few seconds, thereby improving the interest and duration of the user in watching.

[0081] In addition, an instant feedback learning principle is also proposed, that is, based on the collected sliding and dwell behaviors of the user, a most direct "yes / no" vote on the preview segment is constituted. The recommendation system no longer needs to wait for the user to watch a movie or a TV series to update the portrait, but can learn and iterate in real time and high frequency according to the feedback of the user in the few seconds of watching the preview segment, so as to give more accurate recommendations in the next sliding, thereby improving the recommendation efficiency of the video.

[0082] In the embodiments of the present application, a dynamic A / B test optimization principle is also provided, that is, the present application regards all preview segment streams of each video as a huge and perpetual A / B test platform. Through small-flow testing of different preview segments, different covers and different scripts of the same video, the system can scientifically find the most effective interests and hobbies of different crowds in a data-driven manner, thereby maximizing the conversion potential of each video and ultimately realizing the flow optimization of the entire video content library.

[0083] Optionally, in another embodiment, which is based on the above-mentioned embodiments, before receiving the recommendation request sent by the client, the method can further include: creating multiple preview segments of different styles for each video in the video content library in the following manner: Obtaining a multi-dimensional heat map of each video in the multiple videos; based on the multi-dimensional heat map of each video, selecting a corresponding multiple preview segments of different styles according to the time axis of the video and marking; and storing the video marked with the multiple preview segments of different styles into the video content library.

[0084] In this embodiment, the multi-dimensional heat map (also referred to as multi-dimensional heat map) of the video is usually a data visualization content that shows the video heat condition or its change trend in the form of a chart. It usually includes the following information: 1) Time dimension: show the change curve of the data such as the play count, like count and comment count of the video in different time periods (such as hours, days and weeks). For example, the start and end time of the recommendation heat B.

[0085] 2) Content dimension: it can be marked which segments (such as highlights and key scenes) in the video are the "hot areas" with repeated viewing and more interactions of users.

[0086] 3) Association dimension: present the heat association of other related content (such as similar videos, recommended videos and different age stages) of the video.

[0087] On the basis of understanding the multi-dimensional heat map, based on the obtained multi-dimensional heat map of each video, a corresponding multiple preview segments are selected according to the play time axis of the video, and the content of a certain play time period is marked as a preview segment; and the video marked with the multiple preview segments of different styles is stored into the video content library.

[0088] Among them, in this embodiment, multiple preview segments of different styles of the video are selected, which can also be determined on the basis of the multi-dimensional heat map and combined with the analysis of the characteristics of the user group. That is, the user behavior data of the obtained user group is analyzed, and the viewing data of a large number of users is mined to automatically locate and intercept the "high-energy" segments that are "reviewed" by the most users, "paused" for detailed viewing, or trigger a burst of bullet screen comments and reviews, as preview segments.

[0089] Of course, AI content understanding can also be combined to determine it, that is, computer vision and audio analysis technology is used to identify video segments with intense action, key expressions, climax emotional music, or important plot turning points as multiple preview segments.

[0090] In the embodiment of the application, the style type of the preview segment is usually centered on the content characteristics of the preview segment and assisted by multi-dimensional heat data, which can be realized by standardized means such as "feature extraction, label matching and accurate determination", but in the specific implementation process, it is not limited thereto, and the specific implementation process includes: 11) Extracting the two-dimensional core features of each preview segment.

[0091] In this step, the explicit content features and the implicit heat-related features of each preview segment selected from the multi-dimensional heat map can be extracted synchronously to form a complete feature data set for style determination, and the two features can be mutually corroborated to improve the accuracy of determination.

[0092] Among them, the explicit content features: directly extracting the quantifiable and identifiable content attributes of the preview segment itself, which can specifically include picture features (such as color tone, shot switching frequency, live-action / animation ratio, etc.), content theme (such as plot narration, landscape display, dry goods output, funny jokes, and warm healing), presentation form (such as voiceover, mashup, pure picture, and combination of picture and text), audio features (such as background music style, presence or absence of dialogue, dialogue density, etc.), and all features are quantified (such as warm color tone ratio ≥ 75%, shot switching ≤ 1 time / s).

[0093] The implicit heat-related features can be understood as auxiliary basis: the multi-dimensional heat data of the original video corresponding to the preview segment, which inversely corroborates the style highlights, such as that the "interaction heat peak" in the heat map of a certain preview segment is derived from a dense bullet screen mocking, which can assist in determining "funny style"; the "retention heat full score" is derived from a slow-paced landscape picture, which can assist in determining "healing style", and so on, so that the style determination is in line with the real preferences of users.

[0094] 12) Build a preset standardized style label judgment standard, that is, set a standardized style determination standard.

[0095] In this step, a hierarchical and scalable style label can be built in advance as the style standard for all preview clips. The style label corresponds to the feature dimensions extracted in the previous step one by one, avoiding confusion in the judgment standard. In general, the label is set to two levels.

[0096] The first level is the style category (coarse classification): covering the mainstream user preference dimensions, such as plot, healing, humor, dry goods, scenery, technology, and life. Each category has clear core feature thresholds (e.g., healing = warm tone ratio ≥ 70% + lens switching frequency ≤ 1 time / s), but in actual application, it is not limited to this.

[0097] The second level of style subdivision (fine classification): The first level of style category is finely split to solve the problem of ambiguous style boundaries, such as slow-paced scenery healing and warm narrative healing for healing, plot reversal humor and monologue humor for humor, and knowledge point dissection and practical step demonstration for dry goods. Each second-level label has a clear feature matching list, forming a style label and feature threshold comparison table.

[0098] 13) Double-mode intelligent judgment (final labeling) In this embodiment, a "feature precise matching as the main, intelligent clustering as the auxiliary" double judgment mode can be used to match the feature data set of the preview clip with the preset style label system, output a unique and accurate style type result, and adapt to all preview clip scenarios.

[0099] For example, through feature precise matching, the two-dimensional feature data of the preview clip is compared with the "style label-feature threshold comparison table" one by one. When the feature matching degree ≥ 80% (the threshold here is for illustration purposes, and can be adjusted according to actual needs), the corresponding first-level + second-level style label is directly assigned to the clip (e.g., "healing-slow-paced scenery healing"), and the style matching degree score is also marked.

[0100] In the above manner, each preview clip (such as highlight clips and wonderful clips) in the video corresponds to a style type, forming a complete file of preview clip ID-style label-feature details-matching degree, and is marked according to the playback time axis of the corresponding video.

[0101] In the embodiments of the present application, the matching core logic of the style type and the user interest portrait can be realized by user portrait disassembly, style weight matching, and quantitative calculation of matching degree and optimal clip screening to ensure that the highest matching degree preview clip is accurately matched for different user portraits. Specifically, it includes: 21) Structured disassembly of user interest portrait, i.e., extraction of matching dimensions.

[0102] In this step, the user interest profile is first disassembled, and the interest dimensions strongly related to the preview segment style are extracted to form a standardized user interest feature library, ensuring consistency with the style judgment dimensions of the preview segment to reduce mismatching bias.

[0103] Among them, the core interest label is extracted: based on user historical behavior (watching time, completion rate, likes, collections, forwarding), the user is given a style label homologous to the segment (such as a user frequently watches slow-paced landscape videos, and is given a "healing direction - slow-paced landscape healing" interest label), for example, a main interest label combined with two to three secondary interest labels.

[0104] Among them, the interest preference weight is given: according to the user's preference for each style, the interest label is assigned a weight (the sum of the weights = 1), the main interest label has the highest weight (such as 0.6, etc.), and the secondary interest label has a decreasing weight (such as 0.3, 0.1, etc.), the higher the weight, the stronger the user's preference for the style.

[0105] Further, the user's behavior characteristics can also be extracted, and the user's auxiliary preferences (such as preference for 15-second / 30-second segments, preference for high-quality segments) can also be extracted as additional conditions for subsequent optimal segment screening.

[0106] 22) Build a style-image matching weight matrix (clear matching rules) A unified matching weight matrix is built to clearly define the corresponding matching weight of "segment style label" and "user interest label", solving the problem of matching standardization between different labels, the core rule is "the highest weight for accurate matching of the same label, the second for associated matching of similar labels, and no matching value for unrelated labels".

[0107] Among them, the matching can include: accurate matching, associated matching: cross-class matching, etc.

[0108] 23) Quantitative calculation of the accurate matching degree of the preview segment and the user.

[0109] In this step, for each user, the style profile of all preview segments is retrieved, combined with the user's interest label and weight, and the matching degree is calculated for each segment through the weighted sum formula, making the matching result quantifiable, comparable, and without subjective judgment space, the calculation formula is: The matching degree of each preview segment and the user = Σ (the corresponding matching weight of the preview segment style label and the user interest label × the preference weight of the corresponding user interest label) 24) Screening of optimal preview segments and dynamic iterative optimization.

[0110] In this step, the optimal preview segment is screened based on the matching degree score, and a feedback mechanism is established to continuously optimize the matching accuracy, ensuring long-term adaptation to user preference changes.

[0111] First, the optimal preview segment is preferentially screened, that is, all preview segments are sorted in descending order of matching degree score, and the preview segment with the highest matching degree is selected as the first recommended preview segment; if there are multiple highest matching degree segments, secondary screening is performed in combination with user auxiliary preferences (such as segment length, picture quality) to determine the final optimal preview segment. Second, multi-style alternative supplement, that is, 2-3 different style segments with the second highest matching degree are simultaneously recommended for each user to meet the user's diversified needs.

[0112] Finally, iterative optimization, that is, real-time collection of user feedback data (such as complete play rate, skipping, secondary viewing, collection, etc.) on the recommended segment, reverse updating of the labels and weights of user interests, and optimization of the feature threshold of the style label and the matching weight matrix to make the subsequent matching results continuously fit the user's real preferences.

[0113] Optionally, in another embodiment, the method further comprises: testing the video marked with multiple preview segments in the video content library on different user groups; automatically selecting the optimal preview segment of the video watched by different user groups based on the effective dwell rate and the positive film conversion rate in the test results; labeling the optimal preview segment of the video corresponding to different user groups; storing the user groups corresponding to the optimal preview segment of the video in the video content library to improve the accuracy of recommendation.

[0114] In this embodiment, the video marked with multiple preview segments in the video content library is A / B tested, and the preview segment of different user groups is selected according to the test results. The A / B testing process of multiple preview segments of different styles includes: using an A / B testing system to put the video marked with multiple preview segments of different styles into different user groups for testing, automatically selecting the optimal highlight version according to indexes such as "effective dwell rate" and "positive film conversion rate", and feeding back the optimization of the video label of the corresponding video according to the test results to realize accurate recommendation of different user groups, that is, "grass planting". The video label is some keywords about the video, which can be used to improve the search ranking.

[0115] Among them, the effective dwell rate (Effective Dwell Rate) is the proportion of users whose dwell time Dwell_Time watching the video segment is greater than the set threshold (such as 8s). And the positive film conversion rate (CTR_ViewFull) is the proportion of users who click the "watch the positive film" button.

[0116] The embodiment of the present application provides a brand-new, efficient and immersive video recommendation method. The method can significantly reduce the decision cost of a user, and changes the laborious process of finding a video into the easy experience of browsing a video, thereby activating the full video content library of a video platform, improving the video recommendation efficiency, and prolonging the total watching time of the user.

[0117] Also refer to Figure 2 The video playing method provided by the embodiment of the present application is applied to a client and includes the following steps. Step 201: receiving a preview segment of a recommended video from a server end, wherein the preview segment of the recommended video is one selected from a plurality of preview segments of different styles of the video based on a user portrait.

[0118] In this step, the user can receive the preview segment of the recommended video from the server end through a sliding discovery interface, and the preview segment of the recommended video is one selected from a plurality of preview segments of different styles of the video based on a user portrait. The selection process of the preview segment of the video is specifically described in the above embodiment, and is not described here again.

[0119] Step 202: playing the preview segment of the recommended video.

[0120] In this step, the preview segment of the recommended video can be played through a sliding discovery interface on the client. The preview segment of the recommended video that is played can be the first preview segment played by a play time axis of the video, or a middle preview segment, or a last preview segment played, and the like.

[0121] In the embodiment of the present application, the client receives the preview picture of the video recommended by the server end according to the user interest portrait to play, which not only improves the interest and experience of the user in watching the video, but also prolongs the watching time.

[0122] Optionally, in another embodiment, the method can further include the following steps: in the process of playing the preview segment of the recommended video, in response to a first gesture instruction of the preview segment, jumping to a play page of a video corresponding to the preview segment to play the video.

[0123] In this step, when the client detects the gesture instruction of the sliding discovery interface, for example, detects the operation key of the user clicking the play feature film, jumps to the play page of the video corresponding to the preview segment to play the video.

[0124] In the embodiment of the application, the client plays the preview segment of the received recommended video, and in the process of playing, in response to the first gesture instruction of the preview segment, jumps to the playing page of the video corresponding to the preview segment and plays the video. Not only the interest and experience of the user in watching the video are improved, but also the watching time is prolonged.

[0125] Optionally, in another embodiment, the method further comprises, in response to a second gesture instruction of the preview segment, jumping to the preview segment of the next video in the process of playing the preview segment of the recommended video.

[0126] For example, after the user watches the preview segment of the video and does not like the preview segment, the user can slide the discovery interface through a downward sliding gesture operation, so as to jump to the next video.

[0127] Optionally, in another embodiment, the method further comprises, before receiving the preview segment of the video recommended by the server, sending a recommendation request of the video to the server, wherein the recommendation request comprises a user identifier.

[0128] In this step, the user opens an application and enters a sliding discovery interface, and triggers a recommendation request of a video to the server through the sliding discovery interface. The recommendation request can be a first recommendation request or can not be a first recommendation request, which is not limited in the embodiment. The recommendation request comprises a user identifier, and in actual application, can further comprise other parameters, which are not limited in the embodiment.

[0129] Optionally, in another embodiment, the method further comprises, after playing the preview segment of the recommended video, recording user behavior data of playing the preview segment of the video, wherein the user behavior data comprises a time length of playing the preview segment of the video and user interaction data (such as likes, comments, stays, etc.), and sending the user behavior data to the server, so that the server determines feedback intensity of the user to the video content according to the user behavior data, updates a label weight of the video in an interest portrait of the user, and / or adjusts a recommendation ratio of subsequent recommendation of the video based on the feedback intensity and the user behavior data, thereby improving the recommendation accuracy of the video.

[0130] Also refer to Figure 3 An application example of a video recommendation method provided by the embodiment of the application is shown in the figure. The method is applied to the interactive communication between a user terminal (such as a client, etc.) and a server, and the method comprises: Step 301, the user enters the sliding discovery interface of the user terminal, initiates a first recommendation request, and the recommendation request includes a user identifier.

[0131] Step 302, the dynamic recommendation engine on the server side receives the recommendation request, reads the user interest portrait based on the user identifier in the recommendation request.

[0132] In this step, the server side can read the user interest portrait from the user portrait library.

[0133] Step 303, the server side determines the video for this recommendation according to the preset "exploitation and exploration" recommendation strategy (for example, a ratio of 8:2).

[0134] Step 304: if it is "exploitation" (i.e. 80%), go to step 304a to search for video content highly related to the user interest portrait; if it is "exploration" (i.e. 20%), go to step 304b to search for generalized or lower matching degree than the highest matching degree described above video, that is, to select new video content that the user may not have contacted.

[0135] Step 305, the server side matches a preview segment most likely to attract based on the video recommended in step 304a or step 304b, based on the user's interest portrait and the current user's group characteristics (i.e. the behavior data of different user groups).

[0136] In this step, based on the user's interest portrait and the current user's group characteristics, a preview segment is selected from the multiple preview segments of different styles marked on the playback timeline of the recommended video.

[0137] Step 306, the server side sends (i.e. recommends) the preview segment of the recommended video to the user terminal, so that the user terminal displays the preview segment of the video to the user.

[0138] Step 307, the user watches the preview segment of the recommended video through the user terminal and interacts (such as sliding or staying, etc.), and records the behavior data of the interaction.

[0139] For example, left swipe triggers the next video recommendation request, and returns to execute step 302.

[0140] Step 308, the user terminal collects user behavior data (such as dwell time, likes, comments, etc.) of the user watching the preview segment of the video.

[0141] Step 309, the user terminal sends the collected user behavior data to the server side; Step 310, the server side judges the feedback intensity of the user according to the dwell time in the received user behavior data, and then executes steps 311 and 312.

[0142] That is, the server side analyzes the stay duration and quantifies it as the feedback intensity of the content, for example, when judging that the stay duration is shorter than 2 seconds, it is determined as negative feedback, and when judging that the stay duration is longer than 8 seconds, it is determined as strong positive feedback.

[0143] Wherein, the steps 311 and 312 are not limited in the order of execution when they are executed in detail, and can be performed simultaneously.

[0144] Step 311, on the one hand, the server side updates the label weight of the video in the user interest portrait in the user portrait library based on the feedback intensity in real time, so as to improve the accuracy of the next portrait recommendation.

[0145] Step 312, on the other hand, the server side adjusts the recommendation ratio of the video in subsequent recommendation based on the feedback intensity and user behavior data.

[0146] Specifically, the feedback intensity and user behavior data can be aggregated into an A / B test system to optimize the preview segment itself, that is, to optimize the recommendation ratio in the recommendation strategy, so as to improve the next recommendation accuracy of the test system.

[0147] When the user left slides again, return to step 302 to start a new round of recommendation-feedback cycle to form a closed loop of continuous learning and optimization.

[0148] It should be noted that for the method embodiment, in order to simply describe, it is expressed as a series of action combinations, but those skilled in the art should know that the present disclosure is not limited by the described action order, because according to the present application, certain steps can be performed in other order or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions involved are not necessarily required by the present application.

[0149] Please also refer to Figure 4 A video recommendation device block diagram provided by the embodiment of the present application. The device comprises: a receiving module 401, a first acquisition module 402, a determination module 403 and a recommendation module 404, wherein, The receiving module 401 is configured to receive a recommendation request sent by a client, wherein the recommendation request comprises a user identifier; The first acquisition module 402 is configured to acquire the interest portrait of the user based on the user identifier; The determination module 403 is configured to determine the recommended video and the preview segment of the video from the video content library based on the interest portrait of the user, wherein each video comprises a plurality of preview segments of different styles; The recommendation module 404 is used to send the determined preview segment of the video to the client so that the client can display the preview segment of the video.

[0150] Optionally, in another embodiment, based on the above embodiments, the determining module includes: The video selection module is used to select, based on the user's interest profile, the video with the highest matching degree from the video content library according to the first corresponding time period of the recommendation ratio in the video recommendation strategy, or the video with a lower matching degree than the highest matching degree mentioned above according to the second corresponding time period of the recommendation ratio in the video recommendation strategy, as the recommended video.

[0151] Optionally, the determining module further includes: The recommended segment determination module is used to select the preview segment that best matches the user's interest profile from the recommended videos, and use it as the preview segment of the recommended video.

[0152] Optionally, in another embodiment, based on the above embodiments, the device further includes: a second acquisition module 501, the structural block diagram of which is shown below. Figure 5 As shown, where, The second acquisition module 501 is used to acquire the user group characteristics of the user. The recommendation module 404 is further configured to select the preview segment with the highest matching degree from the recommended videos, which combines the user group characteristics and the user's interest profile, as the preview segment of the recommended video.

[0153] Optionally, in another embodiment, based on the above embodiments, the apparatus further includes: The third acquisition module is used to acquire user behavior data of the client playing the preview segment of the video after the recommendation module sends the determined preview segment of the video to the client; An intensity determination module is used to determine the intensity of the user's feedback to the video content based on the user behavior data; An update module is used to update the tag weights of the user's interest profile and the video in real time based on the feedback intensity; and / or adjust the recommendation ratio of the video in subsequent recommendations based on the feedback intensity and the user behavior data.

[0154] Optionally, in another embodiment, based on the above embodiments, the apparatus further includes: a creation module, configured to pre-create multiple preview segments of different styles for each video in the video content library before the receiving module receives the recommendation request sent by the client.

[0155] Optionally, in another embodiment, the embodiment is based on the above-mentioned embodiments, and the creating module comprises: a fourth obtaining module for obtaining a multi-dimensional heat map of each video in a plurality of videos; a selection and marking module for selecting a plurality of preview segments of different styles according to a playing time axis of the video based on the multi-dimensional heat map of each video and marking the video; a first storage module for storing the video marked with the plurality of preview segments of different styles in a video content library.

[0156] Optionally, in another embodiment, the embodiment is based on the above-mentioned embodiments, and the device further comprises: a testing module for testing the video stored in the video content library and marked with the plurality of preview segments on different user groups; a segment selection module for automatically selecting an optimal preview segment of the video watched by different user groups based on an effective stay rate and a positive conversion rate in the test result; a marking module for marking the optimal preview segment of the video corresponding to different user groups; a second storage module for storing the user groups corresponding to the marked preview segment of the video in the video content library.

[0157] Also see Figure 6 a block diagram of a video playing device provided by an embodiment of the application, the device being applied to a client and comprising a receiving module 601 and a first playing module 602, wherein, the receiving module 601 is configured to receive a preview segment of a recommended video sent by a server end, the preview segment of the recommended video being one selected from a plurality of preview segments of different styles of the video based on a user portrait.

[0158] the first playing module 602 is configured to play the recommended preview segment of the video.

[0159] Optionally, the device further comprises a second playing module and / or a jump module, wherein, the second playing module is configured to, in a process of playing the recommended preview segment of the video by the first playing module, jump to a playing page of a video corresponding to the preview segment and play the video in response to a first gesture instruction of the preview segment; the jump module is configured to, in the process of playing the recommended preview segment of the video by the first playing module, jump to a preview segment of a next video in response to a second gesture instruction of the preview segment.

[0160] Optionally, in another embodiment, the device further comprises: The first sending module is configured to send a recommendation request of a video to the server side before the receiving module receives the preview segment of the recommended video on the server side, wherein the recommendation request comprises a user identifier.

[0161] Optionally, in another embodiment, the device further comprises: The behavior data determination module is configured to record user behavior data of playing at least one preview segment of the video after the first playing module plays the preview segment of the video, wherein the user behavior data comprises a duration of playing the preview segment of the video and user interaction data. The second sending module is configured to send the user behavior data to the server side, so that the server side determines a feedback intensity of the user to the video content according to the user behavior data, and updates a label weight in a user portrait based on the feedback intensity and / or adjusts a recommendation ratio of subsequently recommending the video based on the feedback intensity and the user behavior data.

[0162] Also refer to Figure 7 An application example diagram of a video recommendation system is provided, and the video recommendation system comprises a user terminal 71 and a server side 72. Wherein, The user terminal 71 comprises a sliding discovery interface (711) and a behavior data acquisition module (712). The sliding discovery interface (711) is configured to receive a sliding operation of a user, initiate a recommendation request of a video to the server side 72, and receive a preview segment of the recommended video, wherein the recommendation request comprises a user identifier, the preview segment of the recommended video is selected from a plurality of preview segments of different styles of the video based on a user portrait, the preview segment of the recommended video is played, and a first gesture instruction of the preview segment is responded to jump to a playing page of a video corresponding to the preview segment, and the video is played. The data acquisition module (712) is configured to record an interactive behavior of a client playing a preview segment of a video in real time, such as a stay duration, and upload the interactive behavior to the server side 72.

[0163] The server side 72 comprises an intelligent preview segment generation system (721), a dynamic recommendation engine (722), a user portrait database (723), and a video content library and a label library (724), An intelligent preview segment generation system (721) for pre-creating and optimizing multiple preview segments of different styles for a video marked along the play time axis. The system comprises a user behavior analysis module (721a), an AI content understanding module (721b) and an A / B testing and optimization module (721c) working collaboratively, whose output of preview segments and optimization strategies are fed to a video content library and tag library (724) and a dynamic recommendation engine (722) respectively.

[0164] A dynamic recommendation engine (722) is the core of the recommendation decision, which receives recommendation requests from user terminals and user behavior data of users watching preview segments of recommended videos, combines information in a user portrait database (723) and a video content library and tag library (724), executes a recommendation strategy, and decides which preview segment of a video to issue to a user. The issued preview segment is one of multiple preview segments of different styles for the video.

[0165] A user portrait database (723) is used to store and manage the interest portrait and corresponding historical behavior data of each user.

[0166] A video content library and tag library (724) is used to store original videos, multiple preview segments of different styles marked in the videos along the play time axis, and corresponding metadata and dynamic optimization tags.

[0167] It should be noted that the functions and effects of each module are specifically described in the implementation process of the above-mentioned corresponding embodiments, and will not be repeated here. The data flow and control flow shown by the arrows in the figure interact to form a closed-loop system from user interaction to video content recommendation to preview segment optimization.

[0168] In this embodiment, the data process of each module interaction is as follows: The user slides on the sliding interaction interface (711) to trigger a recommendation request, which includes a user identifier.

[0169] The dynamic recommendation engine (722) receives the recommendation request, reads the user interest portrait from the user portrait database (723) according to the user identifier in the recommendation request, and formulates a corresponding recommendation strategy (such as 8:2 utilization and exploration, etc., which is specifically described above and will not be repeated here) according to the user interest portrait.

[0170] The dynamic recommendation engine (722) selects a video from the video content library and tag library (724) according to the recommendation strategy, matches the best preview segment for it, and sends the best preview segment in the video to the user terminal (711). The user terminal (711) plays the preview segment by sliding the discovery interface 711, and during the playback of the preview segment, responds to the first gesture command of the preview segment and jumps to the playback page of the video corresponding to the preview segment to play the video.

[0171] The behavior data acquisition module (712) records user behavior data of the user watching the preview segment of the video. The user behavior data includes interaction data such as dwell time, and sends the behavior data back to the server 72.

[0172] The server-side module 72 includes: a user behavior analysis module (721a), an AI content understanding module (721b), and an A / B testing and optimization module (721c). The user behavior analysis module (721a) on the server side determines the intensity of the user's feedback to the video content based on the received user behavior data; Based on the feedback intensity, update the tag weights of the video in the user's interest profile; and / or, The feedback intensity and user behavior data are used to adjust the recommendation ratio of subsequent recommended videos. In other words, the feedback intensity and user behavior data are integrated into the testing system so that the testing system can conduct recommendation tests and optimize the recommendation ratio of subsequent recommended videos.

[0173] This user behavior data is used by the dynamic recommendation engine (722) to update the next recommendation. Simultaneously, the feedback intensity and user behavior data are fed into the A / B testing module (721c) and user behavior analysis module (721a) of the intelligent preview segment generation system (721) for long-term optimization of the preview segment quality and the accuracy of the user profile. Specifically: The user behavior analysis module (721a) determines the intensity of the user's feedback to the video content based on the received user behavior data; and updates the tag weights of the video in the user's interest profile based on the feedback intensity.

[0174] The A / B testing module (721c) integrates the feedback intensity and user behavior data into the testing system so that the testing system can optimize the recommendation ratio of the video in subsequent recommendations.

[0175] It should be noted that the video recommendation system provided in this embodiment includes a video recommendation device and a video playback device. The video recommendation device may include the various modules of the server-side described above, and the video playback device may include the various modules of the user terminal.

[0176] As to the apparatus in the above-mentioned embodiments, the specific manners in which various modules perform operations have been described in detail in the embodiments of the method, and thus will not be described in detail here.

[0177] The apparatus embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0178] For ease of understanding, please refer to the following application examples. For example, a new science fiction movie Star Trek: Unknown Frontier is newly launched on the platform.

[0179] Step 1: Offline generation of preview clips This is an offline task performed by the intelligent preview clip generation system (721) after the video is ingested according to the multi-dimensional heat map (i.e., multi-dimensional heat map) of the obtained video content.

[0180] Input: Complete video file and subtitle file of the movie Star Trek.

[0181] Objective: Generate multiple (three in this embodiment) candidate preview clips (such as high-light clips Candidate Clips, etc.) of different styles according to the multi-dimensional heat map.

[0182] Processed by the AI content understanding module (721b): Algorithm: A pre-trained 3D-CNN model can be used for video action recognition, and a BERT model can be used for semantic and emotional analysis of subtitles.

[0183] The process includes: The model scans the entire video and identifies visually impactful clips (such as spaceship curve flying, space battle explosion scenes), which are marked as Clip_A (action type).

[0184] The model analyzes the subtitles and finds dialogues containing key plot twists or strong emotional conflicts (such as the confrontation between the main character and the villain), which are marked as Clip_B (plot type).

[0185] The model identifies shots that show grand world views or exotic alien landscapes, which are marked as Clip_C (wonder type).

[0186] Output: Three candidate preview clips Clip_A, Clip_B, Clip_C, each with a duration of 15 seconds. These three clips are marked on the corresponding video according to the play timeline, and the video marked with multiple preview clips is stored in the video content library and the tag library (724).

[0187] Processing (for old videos that have online play data) by user behavior analysis module (721a): Algorithm: Use the time window moving average algorithm to analyze the play logs of all users of this video.

[0188] Processing parameters: Rewind_Count: number of rewinds.

[0189] Comment_Density: number of comments / bullets per unit time.

[0190] Process: Find the Top-K time points with the highest Rewind_Count and Comment_Density per unit time, and generate candidate clips by taking 7.5 seconds before and after each time point.

[0191] For the new movie Star Trek, this user behavior analysis module does not work at the beginning, but will continue to collect user behavior data for watching preview clips of the movie after the movie is online. This behavior data is used to generate better "data-driven" preview clips in the future.

[0192] It should be noted that if it is a TV series, based on the multi-dimensional heat map of each episode of the TV series, a plurality of preview clips corresponding to the play timeline of each episode of the TV series are selected and marked; and each episode of the TV series marked with multiple preview clips is stored in the video content library.

[0193] Step 2: User request and initial recommendation (User Request and Initial Recommendation) Scenario: User Xiaoming (a user with historical data showing a preference for action movies) opens the App and enters the new "slide discovery" function, i.e., enters the slide discovery interface.

[0194] The user terminal (71) initiates a recommendation request to the server side 72, and the request carries the user ID.

[0195] The dynamic recommendation engine (722) in the server side 72 receives the recommendation request. First, according to the user ID, Xiaoming's interest profile is read from the user profile database (723), and it is found that the "action_movie_interest_score" parameter in the interest profile is very high.

[0196] Dynamic recommendation engine (712) executes the "exploitation and exploration" strategy. Assume this hit "80% exploitation" branch.

[0197] Dynamic recommendation engine (712) retrieves a video from the video content library and tag library (724) that is highly relevant to the "action movie" tag and has not been watched by Xiao Ming, i.e., the new movie "Star Trek" is selected.

[0198] A / B testing and optimization module (721c) intervention: At this time, the system needs to select a preview clip of "Star Trek" for Xiao Ming. Due to the early stage of the preview of this movie, the A / B testing system randomly (or according to a certain strategy) assigns a version, such as selecting a preview clip in the video that matches the user portrait the most, of course, the first preview clip in the video according to the timeline can also be recommended to the user, etc., and then a preview clip in the corresponding playback time period can be recommended to the user according to the user's left and right sliding operation on the video timeline.

[0199] Assume: The system selects Clip_A (action type) for Xiao Ming.

[0200] The server returns the video and the playback address and related information of Clip_A in the video to the user terminal.

[0201] Step three: User interaction and real-time feedback (User Interaction and Real-time Feedback) Xiao Ming's phone sliding discovery interface (711) starts playing the 15-second space battle clip (Clip_A) of "Star Trek".

[0202] The behavior data collection module (712) starts a timer in the background to record the dwell time (Dwell Time) and sends the collected dwell time to the server 72.

[0203] Scenario one (strong positive feedback): Xiao Ming is attracted by the battle scene and watches the entire 15-second preview clip.

[0204] Processing: Dwell_Time = 15s. The server determines that the feedback intensity is strong positive feedback.

[0205] Parameter update: After receiving this strong positive feedback, the dynamic recommendation engine (722) will strengthen the weights of the "Star Trek" related tags in Xiao Ming's user behavior portrait, such as "sci-fi_interest_score" += 0.5, "space_battle_interest_score" += 1.0.

[0206] Next recommendation: Xiao Ming's next swipe has a high probability of triggering the "utilize" logic, and the system will recommend another sci-fi blockbuster or a movie related to the actor.

[0207] Scenario two (negative feedback): Xiao Ming is not interested in space battles and swipes away to the next video after 2 seconds of playback.

[0208] Processing: Dwell_Time = 2s. The system determines it as negative feedback.

[0209] Parameter update: The dynamic recommendation engine (722) will slightly reduce the weight of "sci-fi_interest_score" in Xiao Ming's interest profile or mark that Xiao Ming is not interested in "Star Trek" and will not push any preview clips of this movie in the short term.

[0210] Next recommendation: The system may attempt "explore" logic and push a video of a completely different genre (such as a comedy) and its preview clip.

[0211] Step four: Data aggregation and model optimization of A / B testing This is a continuous background process.

[0212] Data aggregation: The A / B testing and optimization module (7201c) continuously collects all user behavior data for the three preview clips of "Star Trek" (Clip_A, Clip_B, Clip_C).

[0213] Main metrics: Effective dwell rate: The proportion of users with Dwell_Time>8s.

[0214] Full movie conversion rate (CTR_ViewFull): The proportion of users who click the "Watch Full Movie" button.

[0215] Data analysis and decision-making: Assume that after a week, the system has the following data: Clip_A (action type): In the "18-25 year old male" group, the effective dwell rate and conversion rate are the highest.

[0216] Clip_B (drama type): Performs best in the "30-40 year old female" group.

[0217] Clip_C (spectacle type): Overall performance is mediocre, but has a good dwell rate among "photography enthusiasts" tag groups.

[0218] Model and strategy optimization: Recommended strategy optimization: The dynamic recommendation engine (7202) updates its recommendation rules. When the next user is "18-25 year old male", the system will preferentially push Clip_A. When the user is "30-40 year old female", Clip_B will be preferentially pushed. This realizes the accurate recommendation of thousands of people and thousands of faces.

[0219] Content tag optimization: In the video content library and tag library (724), the dynamic tags of Star Trek are updated. The system adds more refined tags such as "action scenes suitable for young men" and "dramatic conflicts suitable for mature women" to it, which can be used in other recommendation scenarios in the future.

[0220] Elimination and iteration: Clip_C may be reduced in push weight or even eliminated by the system because of poor overall effect. At the same time, the user behavior analysis module (721a) may have generated a new and more popular Clip_D according to the real play data of the online one week, and added it to the A / B test pool to start a new round of optimization cycle.

[0221] Through the above steps, the application constructs a video recommendation system that can self-learn and self-evolve. According to the user interest portrait, the preview segment of the video interested by the user is pushed to the corresponding user, which not only improves the recommendation efficiency, but also improves the user's viewing time, thereby systematically improving the operation efficiency of the entire platform.

[0222] Optionally, the embodiment of the application further provides an electronic device, comprising: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the video recommendation method or the video playback method as described above.

[0223] Optionally, the embodiment of the application further provides a readable storage medium, when the instructions in the readable storage medium are executed by the processor of the electronic device, the electronic device can execute the video recommendation method or the video playback method as described above.

[0224] Optionally, the embodiment of the application further provides a computer program product, comprising a computer program or instructions, which are executed by the processor of the electronic device to implement the video recommendation method or the video playback method as described above.

[0225] Also see Figure 8FIG. 8 is a block diagram of an electronic device 800, according to an embodiment of the present disclosure. The electronic device 800 can be a mobile terminal or a server. For example, the electronic device 800 can be a mobile phone, a computer, a digital broadcast terminal, a message device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0226] Referring to Figure 8 The electronic device 800 can include one or more of the following components: a processing component 802, a memory 804, a power component 806, a multimedia component 808, an audio component 810, an input / output (I / O) interface 812, a sensor component 814, and a communication component 816.

[0227] The processing component 802 usually controls overall operations of the electronic device 800, such as operations associated with displaying, making phone calls, data communications, camera operations, and recording operations. The processing component 802 can include one or more processors 820 to execute instructions to complete all or part of steps of the methods described above. In addition, the processing component 802 can include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 can include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0228] The memory 804 is configured to store various types of data to support operations of the electronic device 800. Examples of these data include instructions for any application or method operating on the electronic device 800, contact data, phonebook data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or nonvolatile memory, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disc or optical disc.

[0229] The power component 806 provides power to various components of the electronic device 800. The power component 806 can include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.

[0230] The multimedia component 808 includes a screen to provide an output interface between the electronic device 800 and a user. In some embodiments, the screen can include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense a touch, a slide, or a gesture on the touch panel. The touch sensor can not only sense a boundary of a touching or a sliding action, but also detect duration and pressure related to the touching or sliding action. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. The front camera and / or the rear camera can receive external multimedia data when the electronic device 800 is in an operating mode, such as a shooting mode or a video mode. Each of the front and rear camera can be a fixed optical lens system or have a focal length and optical zooming capability.

[0231] The audio component 810 is configured to output and / or input an audio signal. For example, the audio component 810 includes a microphone (MIC) to receive an external audio signal when the electronic device 800 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the memory 804 or transmitted via the communication component 816. In some embodiments, the audio component 810 further includes a speaker to output an audio signal.

[0232] The I / O interface 812 provides an interface between the processing component 802 and peripheral interface modules, which can be a keypad, a click wheel, buttons, and the like. The buttons can include, but are not limited to, a home button, a volume button, a start button, and a lock button.

[0233] The sensor component 814 includes one or more sensors to provide various state assessments for the electronic device 800. For example, the sensor component 814 can detect an open / closed state of the device 800, relative positioning of components, such as a display and a keypad of the electronic device 800, a change in position of the electronic device 800 or a component of the electronic device 800, presence or absence of user contact with the electronic device 800, an orientation or acceleration / deceleration of the electronic device 800, and a temperature change of the electronic device 800. The sensor component 814 can include a proximity sensor configured to detect presence of a nearby object without any physical touch. The sensor component 814 can further include a light sensor such as a CMOS or CCD image sensor for use in an imaging application. In some embodiments, the sensor component 814 can further include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0234] The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 can access a wireless network based on a communication standard, such as WiFi, a cellular network (e.g., 2G, 3G, 4G, or 5G), or a combination thereof. In an example embodiment, the communication component 816 receives broadcast signals or broadcast-related information from external broadcast management systems via a broadcast channel. In an example embodiment, the communication component 816 also includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) techniques, infrared data association (IrDA) techniques, ultra-wideband (UWB) techniques, Bluetooth (BT) techniques, and other techniques.

[0235] In an embodiment, the electronic device 800 can be implemented with one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors, or other electronic elements, to perform the video recommendation method or the video playback method as described above.

[0236] In an embodiment, a readable storage medium is also provided, which, when instructions in the readable storage medium are executed by a processor of an electronic device, enables the electronic device 800 to perform the video recommendation method or the video playback method as described above. For example, the readable storage medium can be a computer-readable storage medium, a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0237] In an embodiment, a computer program product is also provided, which includes a computer program or instructions, which, when executed by the processor 820 of the electronic device 800, enables the electronic device 800 to perform the video recommendation method or the video playback method as described above.

[0238] In the embodiments described above, the entire or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, the entire or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded into and executed by a computer, the entire or part of the processes or functions according to the embodiments of the present application are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatuses. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (for example, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (for example, infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD), or a semiconductor medium (for example, Solid State Disk (SSD)) and the like.

[0239] Figure 9 is a block diagram of an apparatus 900 for video recommendation method or video playing provided by an embodiment of the present application. For example, the apparatus 900 can be provided as a server. Referring to Figure 9 , the apparatus 900 includes a processing component 922, which further includes one or more processors, and a memory resource represented by a memory 932, for storing instructions executable by the processing component 922, such as an application program. The application program stored in the memory 932 can include one or more than one module each corresponding to a set of instructions. In addition, the processing component 922 is configured to execute the instructions to perform the method described above.

[0240] The apparatus 900 can also include a power supply component 926 configured to perform power management of the apparatus 900, a wired or wireless network interface 950 configured to connect the apparatus 900 to a network, and an input / output (I / O) interface 958. The apparatus 900 can operate based on an operating system stored in the memory 932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.

[0241] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.

[0242] It is to be understood that the application is not limited to the precise construction herein described and as shown in the attached drawings, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the application. The scope of the application is to be indicated by the appended claims, rather than the description and examples.

Claims

1. A video recommendation method, characterized by, The method is applied to a server side, and comprises: receiving a recommendation request sent by a client, wherein the recommendation request comprises a user identifier; obtaining an interest profile of the user based on the user identifier; determining recommended videos and preview segments of the videos from a video content library based on the interest profile of the user, wherein each of the videos comprises a plurality of preview segments of different styles; sending the determined preview segments of the videos to the client to enable the client to display the preview segments of the videos.

2. The video recommendation method of claim 1, wherein, The step of determining the recommended videos and the preview segments of the videos from the video content library based on the interest profile of the user comprises: selecting, based on the interest profile of the user, a video with the highest matching degree from the video content library within a time corresponding to a front term of a recommendation ratio in a video recommendation strategy, or selecting a video with a matching degree lower than the highest matching degree from the video content library within a time corresponding to a rear term of the recommendation ratio in the video recommendation strategy, as the recommended video.

3. The video recommendation method of claim 2, wherein, The step of determining the preview segments of the videos based on the interest profile of the user comprises: selecting, from the recommended video, a preview segment with the highest matching degree to the interest profile of the user as the preview segment of the recommended video.

4. The video recommendation method of claim 3, wherein, The method further comprises: obtaining a user group feature of the user; The step of selecting, from the recommended video, a preview segment with the highest matching degree to the interest profile of the user as the preview segment of the recommended video comprises: selecting, from the recommended video, a preview segment with the highest matching degree to a combination of the user group feature of the user and the interest profile of the user as the preview segment of the recommended video. 5.The video recommendation method of any one of claims 1 to 4, characterized in that, After the step of sending the determined preview segments of the videos to the client, the method further comprises: obtaining user behavior data after the client plays the preview segments of the videos; determining a feedback intensity of the user to the video content according to the user behavior data; updating a label weight of the video in the interest profile of the user based on the feedback intensity. 6.The video recommendation method of any one of claims 1 to 4, characterized in that, Before the step of receiving the recommendation request sent by the client, the method further comprises: pre-creating a plurality of preview segments of different styles for each video in a video content library in the following manner: obtaining a multi-dimensional heat map of each video in a plurality of videos; selecting, based on the multi-dimensional heat map of each video, a plurality of preview segments of different styles corresponding to a play time axis of the video and marking the preview segments; storing the video marked with the plurality of preview segments of different styles into the video content library.

7. A video playing method, characterized in that, The method is applied to a client, and comprises: receiving a preview segment of a recommended video sent by a server side, wherein the preview segment of the recommended video is selected from a plurality of preview segments of different styles of the video based on a user profile; playing the preview segment of the recommended video.

8. The video playback method of claim 7, wherein, The method further comprises: In the process of playing the recommended preview segment of the video, in response to a first gesture instruction of the preview segment, jumping to a playing page of the video corresponding to the preview segment and playing the video; or in response to a second gesture instruction of the preview segment, jumping to a preview segment of a next video.

9. The video playback method of claim 7, wherein, Before receiving the preview segment of the server-end recommended video, the method further comprises: sending a recommendation request of a video to a server end, the recommendation request comprising a user identifier.

10. The video playing method of any of claims 7-9, wherein, After playing the recommended preview segment of the video, the method further comprises: recording user behavior data of playing the preview segment of the video, the user behavior data comprising: a time length of playing the preview segment of the video, and user interaction data; sending the user behavior data to the server end, so that the server end determines feedback intensity of the user to the video content according to the user behavior data, and updates a label weight in a user portrait based on the feedback intensity and / or adjusts a recommendation ratio of subsequently recommending the video based on the feedback intensity and the user behavior data.

11. A video recommendation apparatus, comprising: The device applied to a server end comprises: a receiving module configured to receive a recommendation request sent by a client end, the recommendation request comprising a user identifier; a first obtaining module configured to obtain an interest portrait of the user based on the user identifier; a determining module configured to determine, based on the interest portrait of the user, a recommended video and a preview segment of the video from a video content library, each of the videos comprising a plurality of preview segments of different styles; a recommendation module configured to send the determined preview segment of the video to the client end, so that the client end displays the preview segment of the video.

12. A video playback device, comprising: The device applied to a client end comprises: a receiving module configured to receive a preview segment of a recommended video sent by a server end, the preview segment of the recommended video being one selected from a plurality of preview segments of different styles of the video based on a user portrait; a first playing module configured to play the recommended preview segment of the video.

13. An electronic device, comprising: comprise: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the video recommendation method of any one of claims 1 to 6 or the video playing method of any one of claims 7 to 10.

14. A readable storage medium, characterized by, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is enabled to perform the video recommendation method of any one of claims 1 to 6 or the video playing method of any one of claims 7 to 10.

Images