WO2021082603A1 - 视频推送方法、装置、计算机设备和存储介质 - Google Patents

视频推送方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2021082603A1
WO2021082603A1 PCT/CN2020/107976 CN2020107976W WO2021082603A1 WO 2021082603 A1 WO2021082603 A1 WO 2021082603A1 CN 2020107976 W CN2020107976 W CN 2020107976W WO 2021082603 A1 WO2021082603 A1 WO 2021082603A1
Authority
WO
WIPO (PCT)
Prior art keywords
video file
candidate
candidate video
video
click
Prior art date
Application number
PCT/CN2020/107976
Other languages
English (en)
French (fr)
Inventor
李宏强
黎猛
Original Assignee
百果园技术(新加坡)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 百果园技术(新加坡)有限公司 filed Critical 百果园技术(新加坡)有限公司
Publication of WO2021082603A1 publication Critical patent/WO2021082603A1/zh

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/23Processing of content or additional data; Elementary server operations; Server middleware
    • H04N21/239Interfacing the upstream path of the transmission network, e.g. prioritizing client content requests
    • H04N21/2393Interfacing the upstream path of the transmission network, e.g. prioritizing client content requests involving handling client requests
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/251Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/258Client or end-user data management, e.g. managing client capabilities, user preferences or demographics, processing of multiple end-users preferences to derive collaborative data
    • H04N21/25866Management of end-user data
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/262Content or additional data distribution scheduling, e.g. sending additional data at off-peak times, updating software modules, calculating the carousel transmission frequency, delaying a video stream transmission, generating play-lists

Definitions

  • the present disclosure relates to video processing technology, for example, to a video push method, device, computer equipment, and storage medium.
  • platforms that provide these videos usually push some videos that users are interested in to users, so that users can obtain more information and use the services provided by the platform.
  • the platform pushes videos to users in the following two ways:
  • the user's geographic location is acquired in real time, and videos sent by users around the geographic location are pushed to the user.
  • Collect historical data of the user's browsing videos and use the historical data to train a personalized model, input the characteristics of the video and the user into the model to predict the user's preference for the video, and push the video with a high degree of preference to the user.
  • this method of pushing based on geographic location is limited to being able to obtain the user's geographic location in real time, and there are users sending videos around the geographic location, which limits the application range.
  • this method of pushing based on user historical behavior requires offline batch training of models, which has a large computational cost, slow model update speed, and poor real-time performance.
  • the present disclosure provides a video push method, device, computer equipment, and storage medium to solve the problem of limited application range of push based on geographic location, high computational cost and poor real-time performance of push based on user's historical behavior.
  • a video push method including:
  • a video push device including:
  • a candidate video file selection module configured to select at least part of original video files from a preset first video pool as candidate video files
  • a candidate video file push module configured to push the candidate video file to the first client for display
  • a feedback operation data statistics module configured to count the feedback operation data of the first client on the candidate video file
  • a candidate video file transfer module configured to transfer at least part of the candidate video file to a preset second video pool according to the feedback operation data
  • a target video file selection module configured to select at least part of the candidate video files from the second video pool as target video files
  • the target video file push module is configured to push the target video file to the second client for display.
  • a computer device is also provided, and the computer device includes:
  • One or more processors are One or more processors;
  • Memory set to store one or more programs
  • the one or more processors When the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned video push method.
  • a computer-readable storage medium is also provided, and a computer program is stored, and when the computer program is executed by a processor, the above-mentioned video push method is realized.
  • FIG. 1 is a flowchart of a video push method provided by Embodiment 1 of the present invention.
  • FIG. 2 is an exemplary diagram of a video pool provided by an embodiment of the present invention.
  • FIG. 3 is a flowchart of a video push method provided by Embodiment 2 of the present invention.
  • FIG. 4 is a flowchart of a video push method provided by Embodiment 3 of the present invention.
  • FIG. 5 is a schematic structural diagram of a video push device provided by Embodiment 4 of the present invention.
  • FIG. 6 is a schematic structural diagram of a computer device according to Embodiment 5 of the present invention.
  • FIG. 1 is a flow chart of a video push method provided by Embodiment 1 of the present invention. This embodiment is applicable to a situation where a small amount of user feedback is detected through a video file, and then all users are pushed.
  • This method can be executed by a video push device
  • the video push device can be implemented by software and/or hardware, and can be configured in a computer device. For example, the method includes the following steps:
  • the original video file belongs to the video (vedio), which has multiple frames of continuous image data.
  • the continuous image data changes more than 24 frames per second, according to the principle of persistence of vision, the human eye cannot distinguish a single static picture. It looks like a smooth and continuous visual effect.
  • the original video file can be obtained in advance by capturing video files from the network, accumulating video files uploaded by users, and downloading video files from a public database.
  • the format and form of the original video file are different, which is not limited in this embodiment.
  • the format of the video file may include Moving Picture Experts Group (MPEG), RealMedia Variable Bitrate (RMVB), Audio Video Interleaved (AVI), Streaming media format (Flash Video, FLV), etc.
  • MPEG Moving Picture Experts Group
  • RMVB RealMedia Variable Bitrate
  • AVI Audio Video Interleaved
  • FLV Flash Video
  • the format of the original video file may include short video, live video, movie, TV series, and so on.
  • the first video pool 201 and the second video pool 202 are constructed in the computer device in advance.
  • the first video pool 201 may refer to an area where original video files are stored, such as a database, a cache queue, and so on.
  • the second video pool 202 may refer to an area where candidate video files are stored, such as a database, a cache queue, and so on.
  • the summary information of the original video file is stored in the first video pool 201, but the original video file itself is not necessarily stored.
  • the summary information of the candidate video file is stored in the second video pool 202, but the candidate video file itself is not necessarily stored.
  • the copy description can be initialized with the description of the video file, and the operator can configure the corresponding copy description after watching the video file;
  • the cover uses the automatic interception algorithm to intercept the image data for initialization, and the operator can select the appropriate one after watching the video file
  • the image data is used as the cover of the video.
  • the original video files stored in the first video pool 201 may be the originally obtained video files, or the original obtained video files that meet certain conditions (for example, the length of time exceeds a preset threshold, the resolution exceeds a preset threshold). Threshold, etc.) and the recalled video files, this embodiment does not impose restrictions on this.
  • At least part of the original video files may be selected from the first video pool as candidate video files to be pushed to the user 203.
  • the computer device may push the candidate video file to at least the first client terminal.
  • the client displays the candidate video file for the user to browse.
  • pushing the candidate video file to the first client refers to pushing the summary information of the candidate video file to the first client, and does not necessarily refer to pushing the candidate video file itself to the first client.
  • the first client terminal When displaying the summary information of the candidate video file, the first client terminal loads the candidate video file itself and plays it if it receives a click operation.
  • the pushed candidate video file can be regarded as a guessing user Preference information.
  • the pushed candidate video file may be regarded as an advertisement.
  • regional information and language information can be set for candidate video files in advance, and regional information and language information can be set for users.
  • These users can be cached in the queue, and every certain time interval, new users are added, and the users in the queue are updated.
  • the geographic information is used to indicate countries, regions, and other regions, such as China, Singapore, and South Korea.
  • the language information is used to indicate languages, such as Chinese, English, Portuguese, etc.
  • the region information and language information set for the candidate video file are determined, and on the other hand, the first user ID that marks the region information and language information is searched.
  • the first user identity may refer to information that identifies the user, such as a user account, user ID, device identity (such as International Mobile Equipment Identity (IMEI)), etc.
  • IMEI International Mobile Equipment Identity
  • the candidate video file is pushed to the first client that marks the first user ID for display.
  • the corresponding detection time and detection quantity can be set for the candidate video file in advance, and based on the detection time and detection quantity, it is determined whether to push the candidate video file to the destination. Said the first client to display.
  • the detection time and the detection quantity set for the candidate video files are determined; on the other hand, the number of candidate video files that have been pushed is counted as the push quantity.
  • the user 203 in the process of the user 203 browsing the candidate video file, the user 203 can click, share, delete the candidate video file, etc. to give feedback, so as to test the user's feedback on the video file stored in the first video pool 201.
  • the reaction of the candidate video file As shown in FIG. 2, in the process of the user 203 browsing the candidate video file, the user 203 can click, share, delete the candidate video file, etc. to give feedback, so as to test the user's feedback on the video file stored in the first video pool 201. The reaction of the candidate video file.
  • the computer device receives the operation on the candidate video file sent by the first client, thereby generating feedback operation data.
  • the feedback operation data may include operation data indicating positive feedback, such as clicking on a candidate video file, sharing a link of the candidate video file, etc., and may also include operation data indicating negative feedback, such as deleting a candidate video file.
  • S104 Transfer at least part of the candidate video file to a preset second video pool according to the feedback operation data.
  • the candidate video file stored in the first video pool 201 is pushed to at least a number of users 203 for testing, and the feedback operation data of the user 203 is collected, which can represent the candidate video file to a certain extent. Relative to the quality of the user 203, a candidate video file with a higher quality is transferred to the second video pool 202.
  • At least part of candidate video files may be selected from the second video pool 202 as target video files to be pushed to the user 203.
  • users who push candidate video files are test users with a small number, while users who push target video files are normal users with a large number, which can be expressed as the number of second clients is generally much larger than The number of first clients.
  • the computer device may push the target video file to the full amount of the second client.
  • the client displays the target video file for the user to browse.
  • pushing the target video file to the second client terminal refers to pushing the summary information of the target video file to the second client terminal, and does not necessarily mean pushing the target video file itself to the second client terminal.
  • the second client terminal When displaying the summary information of the target video file, the second client terminal loads the target video file itself and plays it if it receives a click operation.
  • the pushed candidate video file can be regarded as a guessing user Preference information.
  • the second client is not compatible with the computer device, for example, the computer device is a server of a short video platform, and the second client is a browser, the pushed target video file can be regarded as an advertisement.
  • the first client and the second client have the same attributes, which ensures that the test effect of the video file can continue to the normal push.
  • the target video file can be set with regional information and language information in advance, and the user can be set with regional information and language information.
  • These users can be cached in the queue, and every certain time interval, new users are added, and the users in the queue are updated.
  • the geographic information is used to indicate countries, regions, and other regions, such as China, Singapore, and South Korea.
  • the language information is used to indicate languages, such as Chinese, English, Portuguese, etc.
  • the geographic information and language information set for the target video file are determined, and on the other hand, the second user ID that marks the geographic information and language information is searched.
  • the second user identification may refer to information identifying the user, such as a user account, a user ID, a device identification (such as an IMEI code), and so on.
  • the video file (candidate video file, target video file)
  • the video file is used as the target video file, and the user can be deduplicated. Avoid repeatedly pushing the same video file to the same user to ensure user experience.
  • the candidate video files are pushed to the first client for display, and the first client’s statistics on the candidate video files Feedback operation data, transfer at least part of the candidate video files to the preset second video pool according to the feedback operation data, select at least part of the candidate video files from the second video pool as the target video file, and push the target video file to the second video pool.
  • the client displays.
  • the video files pushed in this embodiment are not limited to geographic areas. The video files in the video pool can be directly pushed to the user, so there is no need to obtain the user's geographic location.
  • this embodiment tests the feedback of a small number of first clients on candidate video files, and filters out the better quality. High candidate video files are pushed to the second client as the target video file, without collecting user historical data to train personalized models, reducing the complexity of calculation, simplifying the process of pushing, thereby reducing calculation costs and ensuring real-time Sex.
  • FIG. 3 is a flowchart of a video pushing method provided in Embodiment 2 of the present invention. This embodiment uses the foregoing embodiment as a basis to illustrate the processing operation of selecting candidate video files based on Thompson Sampling.
  • the trial in the process of testing the user's reaction, in order to quickly test out candidate video files with a high click rate from the first video pool 201 and add them to the second video pool 202, the trial can be modeled.
  • the number of trial impressions is limited to M, and the number of trials is at least K. If the number of pushes exceeds K and the number of impressions exceeds M, it can be considered that the click rate of the trial has converged and different users can be eliminated And the deviation caused by different sending time, then, the number of pushes for a single trial is:
  • Each user selects one of the n candidate video files for testing, which is a typical Bernoulli Multi-Armed Bandit problem. If the user clicks on the tested candidate video file, it will get a positive one. Score, the goal of the model is to maximize the cumulative benefit of the trial under the limit of the limited number of trials M.
  • A represents a set of a series of actions, each action represents that the user selects one of the n candidate video files for trial.
  • a] ⁇ .
  • a represents the selected candidate video file, a ⁇ A.
  • E represents expectation
  • r represents the feedback value
  • a] represents the expectation of feedback r after selecting action a.
  • the income of the candidate video file corresponding to the optimal selection method is:
  • the loss function is the loss value of not selecting the best candidate video file:
  • the loss function is used to compare with the income.
  • T represents the total number of decisions, Represents the cumulative sum of the feedback value after each decision.
  • the strategy for selecting candidate video files each time can be expressed as:
  • the above formula indicates which candidate video file among the first n candidate video files is selected for trial at time t.
  • a represents the selected candidate video file, a ⁇ A, and A represents the set of the first n candidate video files.
  • h t represents the state at time t, including the beta distribution of the current n candidate video files.
  • h t ) represents the probability of choosing a in the h t state at time t.
  • Q(a) is the same as above, which represents the income of choosing a this time.
  • r t represents the feedback value after the t-th decision.
  • the method includes the following steps:
  • multiple original video files are stored in the first video pool, and the original video files are configured with a click rate.
  • the so-called click rate may refer to the probability that the user clicks on the candidate video file when the original video file is pushed to the first client as a candidate video file.
  • the original video files can be selected through random, sorting, etc., to determine whether the original video file can be pushed as a candidate video file.
  • the original video file can be searched in the preset first video pool, and at least part of the original video file can be selected based on the click rate of the original video file.
  • n is a positive integer and belongs to an adjustable parameter.
  • the click rate of the original video file belongs to the probability, and the probability distribution of the probability is calculated to obtain the probability distribution of the click rate.
  • the probability distribution includes the beta distribution, and the domain of the beta distribution is both [0,1], that is, the beta distribution describes the probability distribution of a random variable defined in the interval [0,1], by
  • the two parameters ⁇ >0 and ⁇ >0 are determined, usually denoted as ⁇ Beta( ⁇
  • Beta distribution is usually used as the probability distribution, and the Beta distribution describes the univariate distribution. Therefore, the Beta distribution can be used as the prior probability of the binomial distribution.
  • the original video file is configured with a display quantity.
  • a preset region such as the k-th country, calculate the average of the click-through rates of candidate video files as the average click-through rate, and calculate the initial click-through rate of the original video file based on the average click-through rate.
  • the average click-through rate can be multiplied by a specified coefficient to obtain the initial click-through rate of the original video file, where the coefficient is greater than 1, such as 2.
  • the newly configured original video file Since the newly configured original video file is not displayed and clicked, by initializing the number of impressions and click-through rate, for example, increasing the initial click-through rate, the newly configured original video file can initially be given priority to be pushed as a candidate video file.
  • the minimum value of the display quantity of the candidate video file is calculated as the minimum display quantity, and the initial display quantity of the original video file is calculated based on the minimum display quantity.
  • the product of the click rate of the original video file and the number of display of the original video file can be calculated as the number of times the original video file is clicked.
  • a beta distribution is generated for the click rate of the original video file.
  • a random value is generated through the probability distribution, and a click rate that conforms to the probability distribution can be obtained as the expected click rate of the original video file.
  • S304 Select at least part of the original video file as candidate video files based on the expected click rate.
  • n expected click rates can be known, from which a suitable original video file is selected as a candidate video file.
  • S308 Calculate a click rate of the candidate video file based on the number of clicks and the number of displays.
  • the feedback operation data includes the number of impressions and the click-through rate.
  • the number of clicks refers to the total number after accumulating the number of clicks during this push
  • the number of impressions refers to the total number after accumulating the number of displays during this push.
  • the ratio between the number of clicks and the number of impressions can be calculated as the click rate of the candidate video file.
  • the ratio between the number of clicks and the number of impressions can be calculated, and the comparison value of the smoothing factor can be used to adjust to obtain the candidate video
  • the click rate of the file is adjusted by the smoothing factor.
  • it can also smooth the click rate to prevent the click rate from fluctuating too much.
  • the smoothing factors are all constants, which can be set according to the number of users tentatively exploring different regions.
  • click-through rate number of clicks/number of impressions+number of smoothed clicks/(number of impressions+number of smoothed impressions), where the number of smoothed clicks and the number of smoothed impressions are smoothing factors.
  • a quantity threshold may be preset, and the display quantity of candidate video files can be compared with the quantity threshold.
  • the candidate video file is transferred to the second video pool.
  • the candidate video files in the second video pool are configured with a click rate, through which a suitable candidate video file can be selected as the target video file.
  • the click rate of the target video file may be updated according to the operation of the second client on the target video file.
  • the original video file is selected in the preset first video pool, the probability distribution is generated for the click rate of the original video file, and the probability value that conforms to the probability distribution is sampled as the expected click rate of the original video file, based on the expected click rate.
  • the click rate selects at least part of the original video files as candidate video files.
  • FIG. 4 is a flowchart of a video push method provided in Embodiment 3 of the present invention. This embodiment uses the foregoing embodiment as a basis to illustrate the processing operation of push monitoring. The method includes the following steps:
  • the candidate video files pushed over a period of time can be monitored, and if significant differences are found, alarm information is generated and sent to the corresponding operating personnel.
  • the processor On the one hand, it counts the first push indicators for pushing candidate video files in the first time period, such as the number of pushes, the number of impressions, and the number of clicks.
  • the second time period and the first time period are time periods that belong to different periods and are in the same position, and achieve a year-on-year comparison.
  • the first period for example, day
  • the first time period for example, hour
  • the interval in which the first period of time is located in the first period is determined.
  • the second time period and the first time period belong to the same period and are in a time period in a different position to achieve a chain comparison. Then in this case, determine the first period (e.g. day) in which the first time period (e.g. hour) is located, and in the first period, set the time period before the first time period as the second time period (such as hours).
  • the first period e.g. day
  • the first time period e.g. hour
  • the first push indicator in the first time period is compared with the second push indicator in the second time period. If the difference between the two is greater than the preset indicator threshold, it is confirmed that a significant difference has occurred.
  • the candidate video file referred to in this embodiment may generally refer to any candidate video file pushed to the first client during the first time period and the second time period, and does not refer to a specific candidate video file .
  • S408 Transfer at least part of the candidate video file to a preset second video pool according to the feedback operation data.
  • the first push index for pushing candidate video files in the first time period is calculated, the second time period relative to the first time period is determined, and the second time period that pushes the candidate video file in the second time period is obtained.
  • the difference between the first push index and the second push index is greater than the preset index threshold, an alarm message is generated, and an alarm is issued when a significant difference is detected to ensure the safety of the push.
  • FIG. 5 is a schematic structural diagram of a video push device provided by Embodiment 4 of the present invention.
  • the device may include the following modules:
  • the candidate video file selection module 501 is configured to select at least part of original video files from a preset first video pool as candidate video files;
  • the candidate video file pushing module 502 is configured to push the candidate video file to the first client for display;
  • the feedback operation data statistics module 503 is configured to count the feedback operation data of the first client on the candidate video file
  • a candidate video file transfer module 504 configured to transfer at least part of the candidate video file to a preset second video pool according to the feedback operation data
  • the target video file selection module 505 is configured to select at least part of the candidate video files from the second video pool as target video files;
  • the target video file pushing module 506 is configured to push the target video file to the second client for display.
  • the video pushing device provided in the embodiment of the present invention can execute the video pushing method provided in any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects of the execution method.
  • FIG. 6 is a schematic structural diagram of a computer device according to Embodiment 5 of the present invention.
  • the computer device includes a processor 600, a memory 601, a communication module 602, an input device 603, and an output device 604; the number of processors 600 in the computer device can be one or more.
  • the processor 600 is taken as an example; the processor 600, the memory 601, the communication module 602, the input device 603, and the output device 604 in the computer equipment may be connected by a bus or other methods. In FIG. 6, the connection by a bus is taken as an example.
  • the computer device provided in this embodiment can execute the video push method provided in any embodiment of the present disclosure, and has corresponding functions and beneficial effects.
  • the sixth embodiment of the present invention also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, a video push method is implemented, and the method includes:
  • the computer program of the computer-readable storage medium provided by the embodiment of the present invention is not limited to the method operations described above, and may also perform related operations in the video push method provided by any embodiment of the present disclosure.
  • the present disclosure can be implemented by software and necessary general-purpose hardware, and can also be implemented by hardware.
  • the present disclosure can be embodied in the form of a software product.
  • the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, Read-Only Memory (ROM), and Random Access Memory (Random Access Memory). , RAM), flash memory (FLASH), hard disk or optical disk, etc., including a variety of instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute the methods described in the multiple embodiments of the present disclosure.
  • the multiple units and modules included are only divided according to the functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be realized; in addition, the multiple functional units
  • the names are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the present disclosure.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computer Graphics (AREA)
  • Computing Systems (AREA)
  • Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
  • Information Transfer Between Computers (AREA)
  • Television Signal Processing For Recording (AREA)

Abstract

本文公开了一种视频推送方法、装置、计算机设备和存储介质,该视频推送方法包括:从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件;将所述候选视频文件推送至第一客户端进行展示;统计所述第一客户端对所述候选视频文件的反馈操作数据;根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中;从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件;将所述目标视频文件推送至第二客户端进行展示。

Description

视频推送方法、装置、计算机设备和存储介质
本申请要求在2019年10月31日提交中国专利局、申请号为201911051675.8的中国专利申请的优先权,该申请的全部内容通过引用结合在本申请中。
技术领域
本公开涉及视频处理技术,例如涉及一种视频推送方法、装置、计算机设备和存储介质。
背景技术
随着网络技术的快速发展,不断有基于数字技术的新媒体涌现,如直播、短视频,等等。这些新媒体在网络技术的支持下,往往具备创新性和混合媒介的特性,得以这些数字媒体为代表的视频成为网络又一热点领域。
为了提高用户体验,提供这些视频的平台,通常会推送一些用户感兴趣的视频给用户,以使得用户获得更多的资讯,使用该平台提供的服务。
平台推送视频给用户使用如下两种方式:
1、基于地理位置推送
在此种方式中,实时获取用户的地理位置,给用户推送该地理位置周边用户发送的视频。
2、基于用户历史行为推送
采集用户浏览视频的历史数据,并使用该历史数据训练个性化模型,将视频和用户的特征输入该模型预测该用户对该视频的喜好程度,将喜好程度高的视频推送给用户。
但是,基于地理位置推送这种方式受限于能够实时获取用户的地理位置,且该地理位置的周边有用户发送视频,应用范围受限。
此外,基于用户历史行为推送这种方式需要离线批量训练模型,计算成本较大,模型更新速度慢,实时性较差。
发明内容
本公开提供一种视频推送方法、装置、计算机设备和存储介质,以解决基于地理位置推送应用范围受限,基于用户历史行为推送计算成本较大、实时性较差的问题。
提供了一种视频推送方法,包括:
从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件;
将所述候选视频文件推送至第一客户端进行展示;
统计所述第一客户端对所述候选视频文件的反馈操作数据;
根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中;
从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件;
将所述目标视频文件推送至第二客户端进行展示。
还提供了一种视频推送装置,包括:
候选视频文件选择模块,设置为从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件;
候选视频文件推送模块,设置为将所述候选视频文件推送至第一客户端进行展示;
反馈操作数据统计模块,设置为统计所述第一客户端对所述候选视频文件的反馈操作数据;
候选视频文件转移模块,设置为根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中;
目标视频文件选择模块,设置为从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件;
目标视频文件推送模块,设置为将所述目标视频文件推送至第二客户端进行展示。
还提供了一种计算机设备,所述计算机设备包括:
一个或多个处理器;
存储器,设置为存储一个或多个程序;
当所述一个或多个程序被所述一个或多个处理器执行,使得所述一个或多个处理器实现上述的视频推送方法。
还提供了一种计算机可读存储介质,存储有计算机程序,该计算机程序被处理器执行时实现上述的视频推送方法。
附图说明
图1为本发明实施例一提供的一种视频推送方法的流程图;
图2为本发明实施例提供的一种视频池的示例图;
图3是本发明实施例二提供的一种视频推送方法的流程图;
图4是本发明实施例三提供的一种视频推送方法的流程图;
图5为本发明实施例四提供的一种视频推送装置的结构示意图;
图6为本发明实施例五提供的一种计算机设备的结构示意图。
具体实施方式
下面结合附图和实施例对本公开进行说明。附图中仅示出了与本公开相关的部分而非全部结构。在不冲突的情况下,本发明实施例及实施例中的特征可以相互组合。
实施例一
图1为本发明实施例一提供的一种视频推送方法的流程图,本实施例可适用于通过视频文件探测少量用户反馈,进而再全量用户推送的情况,该方法可以由视频推送装置来执行,该视频推送装置可以由软件和/或硬件实现,可配置在计算机设备中,例如,该方法包括如下步骤:
S101、从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件。
原始视频文件,属于视频(vedio),其具有多帧连续的图像数据,连续的图像数据变化每秒超过24帧(frame)时,根据视觉暂留原理,人眼无法辨别单幅的静态画面,看上去是平滑连续的视觉效果。
在本实施例中,可预先通过从网络抓取视频文件、累积用户上传的视频文件、从公开的数据库中下载视频文件等方式,获取原始视频文件。
针对不同的业务场景,该原始视频文件的格式、形式有所不同,本实施例对此不加以限制。
在一实施例中,该视频文件的格式可以包括动态图像专家组(Moving Picture Experts Group,MPEG)、可变比特率(RealMedia Variable Bitrate,RMVB)、音频视频交错格式(Audio Video Interleaved,AVI)、流媒体格式(Flash Video,FLV),等等。
该原始视频文件的格式可以包括短视频、直播视频、电影、电视剧,等等。
在本实施例中,如图2所示,预先在计算机设备构建第一视频池201、第二视频池202。
第一视频池201可以指存储原始视频文件的区域,如数据库、缓存队列等。
第二视频池202可以指存储候选视频文件的区域,如数据库、缓存队列等。
可选的,在第一视频池201中存储原始视频文件的摘要信息,如文案描述、封面、地址等,而并不一定存储原始视频文件本身。
同样地,在第二视频池202中存储候选视频文件的摘要信息,如文案描述、封面、地址等,而并不一定存储候选视频文件本身。
其中,文案描述可以使用视频文件的描述初始化,运营人员在观看视频文件之后可再配置相应的文案描述;封面采用自动截取算法截取图像数据进行初始化,运营人员在观看视频文件之后可再选择合适的图像数据作为视频封面。
此外,第一视频池201中存储的原始视频文件可以为原始获取到的视频文件,也可以为原始获取到的视频文件符合一定条件(如时间长度超过预设的阈值、分辨率超过预设的阈值等)而召回的视频文件,本实施例对此不加以限制。
在一实施例中,可以从第一视频池中选择至少部分原始视频文件,作为候选视频文件,待推送至用户203。
S102、将所述候选视频文件推送至第一客户端进行展示。
在一实施例中,若第一客户端满足预设的条件,如暂停播放视频、进入直播间等,则计算机设备可以将该候选视频文件推送(push)至少量的第一客户端,第一客户端展示该候选视频文件供用户浏览。
在一实施例中,所谓将候选视频文件推送至第一客户端,是指将候选视频文件的摘要信息推送至第一客户端,并不一定指将候选视频文件本身推送至第一客户端。
第一客户端在展示候选视频文件的摘要信息时,若接收到点击操作,则加载该候选视频文件本身进行播放。
可选的,若第一客户端与计算机设备配套,例如,该计算机设备为短视频平台的服务器、该第一客户端为短视频应用,则该被推送的候选视频文件可以被视为猜测用户喜好的信息。
若第一客户端与计算机设备不配套,例如,该计算机设备为短视频平台的服务器、该第一客户端为浏览器,则该被推送的候选视频文件可以被视为广告。
在一种推送方式中,可以预先对候选视频文件设置地域信息和语言信息,以及,对用户设置地域信息和语言信息。
这些用户可以缓存在队列中,每间隔一定的时间,新增用户,更新该队列中的用户。
该地域信息用于表示国家、地区等地域,如中国、新加坡、韩国等。
该语言信息用于表示语言,如中文、英文、葡萄牙语等。
在此推送方式中,一方面,确定对候选视频文件设置的地域信息和语言信息,另一方面,查找标记地域信息与语言信息的第一用户标识。
其中,第一用户标识可以指标识用户的信息,如用户账号、用户ID、设备标识(如国际移动设备识别码(International Mobile Equipment Identity,IMEI))等。
将候选视频文件推送至标记第一用户标识的第一客户端进行展示。
此外,针对候选视频文件不同的属性(如地域、语言),可以预先针对该候选视频文件设置相应的探测时间和探测数量,基于该探测时间和探测数量确定是否将所述候选视频文件推送至所述第一客户端进行展示。
在挑选候选视频文件之后,一方面,确定对候选视频文件设置的探测时间、探测数量,另一方面,统计已推送候选视频文件的数量,作为推送数量。
将该探测数量与该推送数量进行比较。若推送数量小于或等于探测数量,则在到达探测时间时,将候选视频文件推送至第一客户端进行展示。若推送数量大于所述探测数量,则禁止将候选视频文件推送至第一客户端进行展示。
S103、统计所述第一客户端对所述候选视频文件的反馈操作数据。
如图2所示,在用户203浏览候选视频文件的过程中,用户203可以对该候选视频文件进行点击、分享、删除等进行反馈的操作,从而可以试探用户对第一视频池201中存储的候选视频文件的反应。
计算机设备接收第一客户端发送的、针对候选视频文件的操作,从而生成反馈操作数据。
可选的,该反馈操作数据可以包括表示正向反馈的操作数据,如点击候选视频文件、分享候选视频文件的链接等,也可以包括表示负向反馈的操作数据,如删除候选视频文件等。
S104、根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中。
在一实施例中,如图2所示,将第一视频池201中存储的候选视频文件推送至少量用户203进行试探,收集用户203的反馈操作数据,可以在一定程度上表示该候选视频文件相对于该用户203的质量,从而将质量较高的候选视频 文件转移至第二视频池202中。
S105、从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件。
在一实施例中,如图2所示,可以从第二视频池202中选择至少部分候选视频文件,作为目标视频文件,待推送至用户203。
可选的,推送候选视频文件的用户,属于测试性质的用户,数量较少,而推送目标视频文件的用户,属于正常性质的用户,数量较多,可表现为第二客户端的数量一般远大于第一客户端的数量。
S106、将所述目标视频文件推送至第二客户端进行展示。
在一实施例中,若第二客户端满足预设的条件,如暂停播放视频、进入直播间等,则计算机设备可以将该目标视频文件推送(push)至全量的第二客户端,第二客户端展示该目标视频文件供用户浏览。
其中,所谓将目标视频文件推送至第二客户端,是指将目标视频文件的摘要信息推送至第二客户端,并不一定指将目标视频文件本身推送至第二客户端。
第二客户端在展示目标视频文件的摘要信息时,若接收到点击操作,则加载该目标视频文件本身进行播放。
可选的,若第二客户端与计算机设备配套,例如,该计算机设备为短视频平台的服务器、该第二客户端为短视频应用,则该被推送的候选视频文件可以被视为猜测用户喜好的信息。
若第二客户端与计算机设备不配套,例如,该计算机设备为短视频平台的服务器、该第二客户端为浏览器,则该被推送的目标视频文件可以被视为广告。
一般情况下,针对同一视频文件(候选视频文件、目标视频文件),第一客户端与第二客户端具有相同的属性,保证该视频文件在测试的效果可以延续至正常的推送中。
在一种推送方式中,可以预先对目标视频文件设置地域信息、语言信息,以及,对用户设置地域信息、语言信息。
这些用户可以缓存在队列中,每间隔一定的时间,新增用户,更新该队列中的用户。
该地域信息用于表示国家、地区等地域,如中国、新加坡、韩国等。
该语言信息用于表示语言,如中文、英文、葡萄牙语等。
在此推送方式中,一方面,确定对目标视频文件设置的地域信息和语言信 息,另一方面,查找标记地域信息与语言信息的第二用户标识。
其中,第二用户标识可以指标识用户的信息,如用户账号、用户ID、设备标识(如IMEI码)等。
将目标视频文件推送至标记第二用户标识的第二客户端进行展示。
此外,针对同一视频文件(候选视频文件、目标视频文件),若该视频文件在先作为候选视频文件已推送至一个用户,则该视频文件作为目标视频文件,则可以对用户进行去重处理,避免重复将同一视频文件推送至相同的用户,保证用户体验。
在本实施例中,从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件,将候选视频文件推送至第一客户端进行展示,统计第一客户端对候选视频文件的反馈操作数据,根据反馈操作数据将至少部分候选视频文件转移至预置的第二视频池中,从第二视频池中选择至少部分候选视频文件,作为目标视频文件,将目标视频文件推送至第二客户端进行展示,相对于基于地理位置推送的方式,本实施例所推送的视频文件并不受限于地域,可将视频池中的视频文件直接推送给用户,因此,无需获取用户的地理位置,也无需该地理位置周边有用户发送视频,大大提高了应用范围;相对于基于用户历史行为推送的方式,本实施例通过试探少量第一客户端对候选视频文件的反馈,筛选出质量较高的候选视频文件,作为目标视频文件推送至第二客户端,无需采集用户的历史数据训练个性化模型,降低了计算的复杂度,简化了推送的过程,从而降低了计算成本,保证了实时性。
实施例二
图3为本发明实施例二提供的一种视频推送方法的流程图,本实施例以前述实施例为基础,说明基于汤普森采样(Thompson Sampling)选择候选视频文件的处理操作。
如图2所示,在试探用户反应的过程中,为了快速从第一视频池201中试探出高点击率的候选视频文件加入到第二视频池202中,可对试探进行建模。
假设单次试探的用户数量为N,每条试探展示次数限制为M,且至少试探次数为K,如果推送次数超过K并且展示次数超过M,可认为试探的点击率已经收敛,能够消除不同用户和不同发送时间导致的偏差,那么,单次试探的推送条数为:
Figure PCTCN2020107976-appb-000001
每一个用户选择n条候选视频文件中的一条进行试探属于典型的伯努利多臂赌博机问题(Bernoulli Multi-Armed Bandit),如果用户对试探的候选视频文件进行点击,就算得到一个正向的得分,模型的目标是在有限的试探次数M限制下,使试探的累计收益最大。
将试探问题简化为没有状态的马尔科夫决策过程(Markov decision process),表示为<A,R>,每次从n条候选视频文件中选择一条进行试探,n条候选视频文件的点击率表示为{θ 12,…θ n}。
A表示一系列动作的集合,每一个动作表示用户选择n条候选视频文件中的一条进行试探。在选择动作的时候,计算选择这条候选视频文件试探之后的预估收益,表示为Q(a)=E[r|a]=θ。其中,a表示选择的候选视频文件,a∈A。E表示期望,r表示反馈值,E[r|a]表示在选择动作a后,反馈r的期望。如果在第t次试探中选择了第i条候选视频文件,那么第t次试探的预估收益为Q(a t)=θ i。R表示反馈函数,第t次试探过程中,真实的反馈值r t=R(a t),有Q(a t)的概率为1。
模型的目标是最大化累计收益
Figure PCTCN2020107976-appb-000002
理想情况下最优的选择方式对应的候选视频文件的收益为:
Figure PCTCN2020107976-appb-000003
因此,损失函数为不选择最优候选视频文件的损失值:
Figure PCTCN2020107976-appb-000004
其中,该损失函数用于与收益进行对比。其中,T表示决策的总次数,
Figure PCTCN2020107976-appb-000005
表示对每次决策后的反馈值进行累计求和。
在本实施例中,采用汤普森采样(Thompson Sampling)求解,假设Q(a)满 足贝塔分布(Beta distribution),点击率θ满足伯努利分布(Bernoulli distribution)。Beta(α,β)范围为[0,1],其中α,β分别是表示点击和没有点击的次数。
每次选择候选视频文件的策略可表示为:
Figure PCTCN2020107976-appb-000006
其中,上述公式表示在t时刻,选择前n条候选视频文件中的哪条候选视频文件进行试探。
a表示选择的候选视频文件,a∈A,A表示前n条候选视频文件的集合。
h t表示t时刻的状态,包括当前n条候选视频文件的贝塔分布。
π(a|h t)表示在t时刻h t状态下,选择a的概率。
Q(a)同上,表示本次选择a的收益。
Figure PCTCN2020107976-appb-000007
表示在t时刻h t状态下,选择a比选择其他所有候选视频文件收益大的概率。
a t表示第t次选择的候选视频文件,1(a t=argmax a∈AQ(a))表示如果第t次选择的候选视频文件是收益最大的候选视频文件,就返回1,否则返回0。
Figure PCTCN2020107976-appb-000008
表示在t时刻h t状态下,累计反馈的期望值。
在每次选择候选视频文件时,对每个候选视频文件对应的Beta(α ii)分布分别进行采样得到Q(a),a∈A,选择Q(a)的最大值对应的候选视频文件进行试探,其候选视频文件表示为a t=argmax a∈AQ(a)。
当用户的反馈操作数据回流之后,更新每条候选视频文件的贝塔分布:
α i←α i+r t1(a t=a i)
β i=β i+(1-r t)1(a t=a i)
其中,1(a t=a i)表示指示器函数,如果a t=a i就返回1,否则返回0。r t表示第t次决策之后的反馈值。
参见图3,该方法包括如下步骤:
S301、在预置的第一视频池中选择原始视频文件。
在本实施例中,第一视频池中存储有多个原始视频文件,该原始视频文件配置有点击率。
所谓点击率,可以指该原始视频文件作为候选视频文件推送至第一客户端时,用户点击该候选视频文件的概率。
由于原始视频文件的数量较多,可以通过随机、排序等方式,从中选择原始视频文件,判定该原始视频文件是否可以作为候选视频文件推送。
在一种选择方式中,可以在预置的第一视频池中查找原始视频文件,基于原始视频文件的点击率选择至少部分原始视频文件。
例如,为了快速筛选出点击率高的候选视频文件,可以截取点击率从高到低排序前n条,即对点击率进行倒序排序,排序越高,则点击率越大,反之,排序越低,则点击率越小。
选择排序最高的n个点击率所属的原始视频文件,其中,n为正整数,属于可调的参数。
S302、对所述原始视频文件的点击率生成概率分布。
在一实施例中,原始视频文件的点击率属于概率,计算该概率的分布,可获得点击率的概率分布。
在一种概率分布的示例中,概率分布包括贝塔分布,贝塔分布的定义域均为[0,1],即贝塔分布描述的是定义在区间[0,1]上随机变量的概率分布,由两个参数α>0和β>0决定,通常记为μ~Beta(μ|α,β)。
在实际使用中,通常将贝塔分布作为概率的分布,贝塔分布描述的是单变量分布,因此,Beta分布可作为二项分布的先验概率。
可选的,原始视频文件配置有展示数量。
在初始时,在预设的地域内,如第k个国家,计算候选视频文件的点击率的平均值,作为平均点击率,基于平均点击率计算原始视频文件初始的点击率。
例如,可在平均点击率的基础上乘以指定的系数,获得原始视频文件初始的点击率,其中,系数大于1,如2。
由于新配置的原始视频文件没有展示和点击,通过初始化展示数量和点击率,例如提高初始的点击率,可以使得新配置的原始视频文件初始获得优先作为候选视频文件进行推送的机会。
此外,在预设的地域内,如第k个国家,计算候选视频文件的展示数量的最小值,作为最低展示数量,基于最低展示数量计算原始视频文件初始的展示数量。
例如,将最低展示数量直接赋值为原始视频文件初始的展示数量。
在本示例中,可计算原始视频文件的点击率与原始视频文件的展示数量之间的乘积,作为原始视频文件被点击的次数。
基于原始视频文件的点击率计算原始视频文件的未点击率,即1-原始视频文件的点击率,即为原始视频文件的未点击率。
计算原始视频文件的未点击率与原始视频文件的展示数量之间的乘积,作为原始视频文件未被点击的次数。
以被点击的次数作为阿尔法α、未被点击的次数作为贝塔β,对原始视频文件的点击率生成贝塔分布。
假设平均点击率为C k,最低展示次数为D k,对于贝塔分布:
α=D k×2C k
β=D k×(1-2C k)
S303、采样符合所述概率分布的概率值,作为所述原始视频文件的期望点击率。
通过该概率分布产生一个随机值,可以得到符合这个概率分布的一个点击率,作为原始视频文件的期望点击率。
S304、基于所述期望点击率选择至少部分所述原始视频文件,作为候选视频文件。
对于n条原始视频文件,可知n个期望点击率,从中选择合适的原始视频文件,作为候选视频文件。
一般而言,可以选择期望点击率最高的原始视频文件,作为候选视频文件。
也可以选择期望点击率第二高或第三高的原始视频文件,作为候选视频文件,本实施例对此不加以限制。
S305、将所述候选视频文件推送至第一客户端进行展示。
S306、统计所述候选视频文件在所述第一客户端中被点击的次数,作为点击数量。
S307、统计所述候选视频文件在所述第一客户端中展示的数量,作为展示数量。
S308、基于所述点击数量与所述展示数量计算所述候选视频文件的点击率。
在本实施例中,反馈操作数据包括展示数量、点击率。
在用户的操作回流之后,一方面,统计候选视频文件在第一客户端中被点击的次数,作为点击数量,另一方面,统计候选视频文件在第一客户端中展示的数量,作为展示数量。
可选的,该点击数量指累积本次推送时被点击的数量之后的总数量,该展示数量指累积本次推送时被展示的数量之后的总数量。
基于点击数量与展示数量计算候选视频文件的点击率,此时,概率分布相应更新。
可选的,可以计算点击数量与展示数量之间的比值,作为候选视频文件的点击率。
为了在初始时出现点击率为0等无意义的情况,并且,初始时的点击率置信度较低,可以计算点击数量与展示数量之间的比值,使用平滑因子对比值进行调整,获得候选视频文件的点击率,通过平滑因子修整点击率,除了避免点击率为0等无意义的情况,降低初始时置信度较低的点击率的影响,还可以平滑点击率,防止点击率波动过大。
其中,平滑因子都是常数,可以根据对不同地域试探的用户数量设置。
在一个示例中,点击率=点击数量/展示数量+平滑点击数量/(展示数量+平滑展示数量),其中,平滑点击数量与平滑展示数量为平滑因子。
S309、若所述候选视频文件的展示数量超过预设的数量阈值,则将所述候选视频文件转移至预置的第二视频池中。
在本实施例中,可以预先设置数量阈值,将候选视频文件的展示数量与数量阈值进行比较。
若候选视频文件的展示数量超过数量阈值,则将候选视频文件转移至第二视频池中。
此外,若候选视频文件的点击率未超过预设的点击率阈值,则禁止将候选视频文件推送至第一客户端。
S310、在所述第二视频池中查找候选视频文件。
S311、基于所述点击率选择至少部分所述候选视频文件,作为目标视频文件。
在本实施例中,第二视频池中的候选视频文件配置有点击率,通过该点击 率可以选择合适的候选视频文件,作为目标视频文件。
为了快速筛选出点击率高的目标视频文件,可以截取点击率从高到低排序的前m条,即对点击率进行倒序排序,排序越高,则点击率越大,反之,排序越低,则点击率越小。
选择排序最高的m个点击率所属的候选视频文件,作为目标视频文件,其中,m为正整数,属于可调的参数。
可选的,将目标视频文件推送至第二客户端展示之后,可根据第二客户端对目标视频文件的操作,更新目标视频文件的点击率。
S312、将所述目标视频文件推送至第二客户端进行展示。
在本实施例中,在预置的第一视频池中选择原始视频文件,对原始视频文件的点击率生成概率分布,采样符合概率分布的概率值,作为原始视频文件的期望点击率,基于期望点击率选择至少部分原始视频文件,作为候选视频文件,通过对点击率的概率分布进行采样,计算速度快,可快速过滤出质量较高的候选视频文件,实时定位用户的兴趣,提高用户整体的点击率。
实施例三
图4为本发明实施例三提供的一种视频推送方法的流程图,本实施例以前述实施例为基础,说明推送监控的处理操作,该方法包括如下步骤:
S401、从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件。
S402、将所述候选视频文件推送至第一客户端进行展示。
S403、统计在第一时间段中推送所述候选视频文件的第一推送指标。
S404、确定与所述第一时间段相对的第二时间段。
S405、获取在所述第二时间段中推送所述候选视频文件的第二推送指标。
S406、若所述第一推送指标与所述第二推送指标之间的差值大于预设的指标阈值,则生成报警信息。
在本实施例中,可以对一段时间内推送的候选视频文件进行监控,如果发现显著差异,则生成报警信息,发送至相应的运营人员。
一方面,统计在第一时间段中推送候选视频文件的第一推送指标,如推送数量、展示数量、点击数量等。
另一方面,确定可与第一时间段进行比较的第二时间段,统计在第二时间 段中推送候选视频文件的第二推送指标,如推送数量、展示数量、点击数量等。
一种情况中,该第二时间段与第一时间段是属于不同周期、以及处于相同位置的时间段,实现同比。则在此情况中,确定第一时间段(如小时)所处的第一周期(如天),确定第一时间段在第一周期所处的区间。
确定位于第一周期之前的第二周期(如天),将第二周期中处于区间的周期设置为第二时间段(如小时)。
在另一种情况中,该第二时间段与第一时间段属于相同周期、处于不同位置的时间段,实现环比。则在此情况中,确定第一时间段(如小时)所处的第一周期(如天),在该第一周期中,将位于第一时间段之前的时间段设置为第二时间段(如小时)。
将第一时间段内的第一推送指标与第二时间段内的第二推送指标进行比较,若两者之间的差值大于预设的指标阈值,即确认发生显著差异。
可选的,本实施例中所指的候选视频文件,可泛指在该第一时间段、第二时间段内任一推送至第一客户端的候选视频文件,并非指一个特定的候选视频文件。
S407、统计所述第一客户端对所述候选视频文件的反馈操作数据。
S408、根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中。
S409、从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件。
S410、将所述目标视频文件推送至第二客户端进行展示。
在本实施例中,统计在第一时间段中推送候选视频文件的第一推送指标,确定与第一时间段相对的第二时间段,获取在第二时间段中推送候选视频文件的第二推送指标,若第一推送指标与第二推送指标之间的差值大于预设的指标阈值,则生成报警信息,在监测到显著差异时进行报警,保证推送的安全性。
实施例四
图5为本发明实施例四提供的一种视频推送装置的结构示意图,该装置可以包括如下模块:
候选视频文件选择模块501,设置为从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件;
候选视频文件推送模块502,设置为将所述候选视频文件推送至第一客户端进行展示;
反馈操作数据统计模块503,设置为统计所述第一客户端对所述候选视频文件的反馈操作数据;
候选视频文件转移模块504,设置为根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中;
目标视频文件选择模块505,设置为从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件;
目标视频文件推送模块506,设置为将所述目标视频文件推送至第二客户端进行展示。
本发明实施例所提供的视频推送装置可执行本公开任意实施例所提供的视频推送方法,具备执行方法相应的功能模块和有益效果。
实施例五
图6为本发明实施例五提供的一种计算机设备的结构示意图。如图6所示,该计算机设备包括处理器600、存储器601、通信模块602、输入装置603和输出装置604;计算机设备中处理器600的数量可以是一个或多个,图6中以一个处理器600为例;计算机设备中的处理器600、存储器601、通信模块602、输入装置603和输出装置604可以通过总线或其他方式连接,图6中以通过总线连接为例。
本实施例提供的计算机设备,可执行本公开任一实施例提供的视频推送方法,具备相应的功能和有益效果。
实施例六
本发明实施例六还提供一种计算机可读存储介质,存储有计算机程序,该计算机程序被处理器执行时实现一种视频推送方法,该方法包括:
从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件;
将所述候选视频文件推送至第一客户端进行展示;
统计所述第一客户端对所述候选视频文件的反馈操作数据;
根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中;
从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件;
将所述目标视频文件推送至第二客户端进行展示。
本发明实施例所提供的计算机可读存储介质,其计算机程序不限于如上所述的方法操作,还可以执行本公开任意实施例所提供的视频推送方法中的相关操作。
通过以上关于实施方式的描述,本公开可借助软件及必需的通用硬件来实现,也可以通过硬件实现。本公开可以以软件产品的形式体现出来,该计算机软件产品可以存储在计算机可读存储介质中,如计算机的软盘、只读存储器(Read-Only Memory,ROM)、随机存取存储器(Random Access Memory,RAM)、闪存(FLASH)、硬盘或光盘等,包括多种指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本公开多个实施例所述的方法。
上述视频推送装置的实施例中,所包括的多个单元和模块只是按照功能逻辑进行划分的,但并不局限于上述的划分,只要能够实现相应的功能即可;另外,多个功能单元的名称也只是为了便于相互区分,并不用于限制本公开的保护范围。

Claims (20)

  1. 一种视频推送方法,包括:
    从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件;
    将所述候选视频文件推送至第一客户端进行展示;
    统计所述第一客户端对所述候选视频文件的反馈操作数据;
    根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中;
    从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件;
    将所述目标视频文件推送至第二客户端进行展示。
  2. 根据权利要求1所述的方法,其中,所述从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件,包括:
    在预置的第一视频池中选择所述原始视频文件,所述原始视频文件配置有点击率;
    对所述原始视频文件的点击率生成概率分布;
    采样符合所述概率分布的概率值,作为所述原始视频文件的期望点击率;
    基于所述期望点击率选择至少部分所述原始视频文件,作为所述候选视频文件。
  3. 根据权利要求2所述的方法,其中,所述在预置的第一视频池中选择所述原始视频文件,包括:
    在预置的第一视频池中查找所述原始视频文件;
    基于所述点击率选择至少部分所述原始视频文件。
  4. 根据权利要求3所述的方法,其中,所述基于所述点击率选择至少部分原始视频文件,包括:
    对所述点击率进行倒序排序;
    选择排序最高的n个点击率所属的原始视频文件。
  5. 根据权利要求2所述的方法,其中,所述原始视频文件配置有展示数量,所述概率分布包括贝塔分布;
    所述对所述原始视频文件的点击率生成概率分布,包括:
    计算所述原始视频文件的点击率与所述原始视频文件的展示数量之间的乘积,作为所述原始视频文件被点击的次数;
    基于所述原始视频文件的点击率计算所述原始视频文件的未点击率;
    计算所述原始视频文件的未点击率与所述原始视频文件的展示数量之间的乘积,作为所述原始视频文件未被点击的次数;
    以所述被点击的次数作为阿尔法、所述未被点击的次数作为贝塔,对所述原始视频文件的点击率生成贝塔分布。
  6. 根据权利要求2-5中任一项所述的方法,其中,所述从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件,还包括:
    在预设的地域内,计算所述候选视频文件的点击率的平均值,作为平均点击率;
    基于所述平均点击率计算所述原始视频文件初始的点击率;
    或者,
    在预设的地域内,计算所述候选视频文件的展示数量的最小值,作为最低展示数量;
    基于所述最低展示数量计算所述原始视频文件初始的展示数量。
  7. 根据权利要求6所述的方法,其中,所述基于所述平均点击率计算所述原始视频文件初始的点击率,包括:
    在所述平均点击率的基础上乘以指定的系数,获得所述原始视频文件初始的点击率,其中,所述系数大于1。
  8. 根据权利要求1-5中任一项所述的方法,其中,
    所述将所述候选视频文件推送至第一客户端进行展示,包括:
    确定对所述候选视频文件设置的地域信息和语言信息;
    查找标记所述地域信息与所述语言信息的第一用户标识;
    将所述候选视频文件推送至标记所述第一用户标识的第一客户端进行展示;
    所述将所述目标视频文件推送至第二客户端进行展示,包括:
    确定对所述目标视频文件设置的地域信息和语言信息;
    查找标记所述地域信息与所述语言信息的第二用户标识;
    将所述目标视频文件推送至标记所述第二用户标识的第二客户端进行展示。
  9. 根据权利要求1-5中任一项所述的方法,其中,所述将所述候选视频文件推送至第一客户端进行展示,包括:
    确定对所述候选视频文件设置的探测时间和探测数量;
    统计已推送所述候选视频文件的数量,作为推送数量;
    在所述推送数量小于或等于所述探测数量且在到达所述探测时间的情况下,将所述候选视频文件推送至所述第一客户端进行展示;
    在所述推送数量大于所述探测数量的情况下,禁止将所述候选视频文件推送至所述第一客户端进行展示。
  10. 根据权利要求1-5中任一项所述的方法,其中,所述反馈操作数据包括展示数量和点击率;
    所述统计所述第一客户端对所述候选视频文件的反馈操作数据,包括:
    统计所述候选视频文件在所述第一客户端中被点击的次数,作为点击数量;
    统计所述候选视频文件在所述第一客户端中展示的数量,作为展示数量;
    基于所述点击数量与所述展示数量计算所述候选视频文件的点击率。
  11. 根据权利要求10所述的方法,其中,所述基于所述点击数量与所述展示数量计算所述候选视频文件的点击率,包括:
    计算所述点击数量与所述展示数量之间的比值;
    使用平滑因子对所述比值进行调整,获得所述候选视频文件的点击率。
  12. 根据权利要求10所述的方法,其中,所述根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中,包括:
    在所述候选视频文件的展示数量超过预设的数量阈值的情况下,将所述候选视频文件转移至所述预置的第二视频池中。
  13. 根据权利要求10所述的方法,还包括:
    在所述候选视频文件的点击率未超过预设的点击率阈值的情况下,禁止将所述候选视频文件推送至所述第一客户端。
  14. 根据权利要求1-5中任一项所述的方法,其中,所述从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件,包括:
    在所述第二视频池中查找候选视频文件,所述候选视频文件配置有点击率;
    基于所述点击率选择至少部分所述候选视频文件,作为所述目标视频文件。
  15. 根据权利要求14所述的方法,其中,所述基于所述点击率选择至少部分所述候选视频文件,作为所述目标视频文件,包括:
    对所述点击率进行倒序排序;
    选择排序最高的m个点击率所属的候选视频文件,作为所述目标视频文件。
  16. 根据权利要求1-5中任一项所述的方法,还包括:
    统计在第一时间段中推送所述候选视频文件的第一推送指标;
    确定与所述第一时间段相对的第二时间段;
    获取在所述第二时间段中推送所述候选视频文件的第二推送指标;
    在所述第一推送指标与所述第二推送指标之间的差值大于预设的指标阈值的情况下,生成报警信息。
  17. 根据权利要求16所述的方法,其中,所述确定与所述第一时间段相对的第二时间段,包括:
    确定所述第一时间段所处的第一周期;
    确定所述第一时间段在所述第一周期所处的区间;
    确定位于所述第一周期之前的第二周期;
    将所述第二周期中处于所述区间的周期设置为所述第二时间段;
    或者,
    确定所述第一时间段所处的第一周期;
    在所述第一周期中,将位于所述第一时间段之前的时间段设置为所述第二时间段。
  18. 一种视频推送装置,包括:
    候选视频文件选择模块,设置为从预置的第一视频池中选择至少部分原始视频文件,作为候选视频文件;
    候选视频文件推送模块,设置为将所述候选视频文件推送至第一客户端进行展示;
    反馈操作数据统计模块,设置为统计所述第一客户端对所述候选视频文件的反馈操作数据;
    候选视频文件转移模块,设置为根据所述反馈操作数据将至少部分所述候选视频文件转移至预置的第二视频池中;
    目标视频文件选择模块,设置为从所述第二视频池中选择至少部分所述候选视频文件,作为目标视频文件;
    目标视频文件推送模块,设置为将所述目标视频文件推送至第二客户端进行展示。
  19. 一种计算机设备,包括:
    一个或多个处理器;
    存储器,设置为存储一个或多个程序;
    所述一个或多个程序被所述一个或多个处理器执行,使得所述一个或多个处理器实现如权利要求1-17中任一项所述的视频推送方法。
  20. 一种计算机可读存储介质,存储有计算机程序,所述计算机程序被处理器执行时实现如权利要求1-17中任一项所述的视频推送方法。
PCT/CN2020/107976 2019-10-31 2020-08-07 视频推送方法、装置、计算机设备和存储介质 WO2021082603A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911051675.8 2019-10-31
CN201911051675.8A CN110769283B (zh) 2019-10-31 2019-10-31 一种视频推送方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2021082603A1 true WO2021082603A1 (zh) 2021-05-06

Family

ID=69334940

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/107976 WO2021082603A1 (zh) 2019-10-31 2020-08-07 视频推送方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN110769283B (zh)
WO (1) WO2021082603A1 (zh)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110769283B (zh) * 2019-10-31 2021-08-27 广州市网星信息技术有限公司 一种视频推送方法、装置、计算机设备和存储介质
CN111464864B (zh) * 2020-04-02 2022-12-06 Oppo广东移动通信有限公司 倒序视频获取方法、装置、电子设备以及存储介质
CN111726649B (zh) * 2020-06-28 2021-12-28 百度在线网络技术(北京)有限公司 视频流处理方法、装置、计算机设备和介质
CN112579822A (zh) * 2020-12-25 2021-03-30 百果园技术(新加坡)有限公司 一种视频数据的推送方法、装置、计算机设备和存储介质
CN112788702B (zh) * 2021-01-07 2022-07-22 西安电子科技大学 一种用于卫星物联网移动终端的映射表项缓存更新方法
CN112836085A (zh) * 2021-02-08 2021-05-25 深圳市欢太科技有限公司 一种权重调整方法及装置、存储介质
CN113722615B (zh) * 2021-08-19 2023-11-21 北京智慧星光信息技术有限公司 互联网数据精准推送方法、系统、电子设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160381419A1 (en) * 2013-08-26 2016-12-29 Google Inc. Segmentation of a video based on user engagement in respective segments of the video
CN106294775A (zh) * 2016-08-11 2017-01-04 合智能科技(深圳)有限公司 内容推送方法及装置
CN107592572A (zh) * 2017-09-21 2018-01-16 广州华多网络科技有限公司 视频推荐方法、装置及其设备
CN110413895A (zh) * 2019-07-31 2019-11-05 腾讯科技(深圳)有限公司 一种信息推送的方法以及相关装置
CN110769283A (zh) * 2019-10-31 2020-02-07 广州市网星信息技术有限公司 一种视频推送方法、装置、计算机设备和存储介质
CN110809177A (zh) * 2019-10-22 2020-02-18 腾讯科技(深圳)有限公司 内容的处理方法、装置、服务器及存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107526758B (zh) * 2016-10-25 2020-07-28 腾讯科技(深圳)有限公司 消息推送方法和装置
CN106779985A (zh) * 2017-02-24 2017-05-31 武汉奇米网络科技有限公司 一种个性化商品排序的方法及系统
US10635707B2 (en) * 2017-09-07 2020-04-28 Xerox Corporation Contextual memory bandit for proactive dialogs

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160381419A1 (en) * 2013-08-26 2016-12-29 Google Inc. Segmentation of a video based on user engagement in respective segments of the video
CN106294775A (zh) * 2016-08-11 2017-01-04 合智能科技(深圳)有限公司 内容推送方法及装置
CN107592572A (zh) * 2017-09-21 2018-01-16 广州华多网络科技有限公司 视频推荐方法、装置及其设备
CN110413895A (zh) * 2019-07-31 2019-11-05 腾讯科技(深圳)有限公司 一种信息推送的方法以及相关装置
CN110809177A (zh) * 2019-10-22 2020-02-18 腾讯科技(深圳)有限公司 内容的处理方法、装置、服务器及存储介质
CN110769283A (zh) * 2019-10-31 2020-02-07 广州市网星信息技术有限公司 一种视频推送方法、装置、计算机设备和存储介质

Also Published As

Publication number Publication date
CN110769283B (zh) 2021-08-27
CN110769283A (zh) 2020-02-07

Similar Documents

Publication Publication Date Title
WO2021082603A1 (zh) 视频推送方法、装置、计算机设备和存储介质
EP2479684B1 (en) Method and evaluation server for evaluating a plurality of videos
CN104584571B (zh) 在机顶盒处产生音频指纹序列
CN103686237B (zh) 推荐视频资源的方法及系统
KR101840420B1 (ko) 챗봇 플랫폼 제공 방법 및 장치
US20170142177A1 (en) Method and system for network dispatching
US20150039601A1 (en) Pre-delivery of content to a user device
CN104853230B (zh) 一种热点视频推送方法及装置
CN110366043B (zh) 视频处理方法、装置、电子设备及可读介质
CN103763585B (zh) 一种用户特征信息获取方法、装置和终端设备
CN103595576B (zh) 一种基于内容提供商标识的互联口icp流量统计系统及方法
EP1971075A1 (en) An information issuing system, a public media information issuing system and an issuing method
JP2002279026A (ja) 番組関心度提示サーバ、番組関心度提示方法、および番組関心度提示プログラム
CA2945801A1 (en) Systems and methods for generating network intelligence through real-time analytics
CN103997662A (zh) 节目推送的方法及系统
CN106797327A (zh) 使用与自适应比特率流传输相关联的消息执行对移动平台的媒体监视
TW201735654A (zh) 多媒體資源的品質評估方法和裝置
CN109688217B (zh) 一种消息推送方法、装置及电子设备
KR102327383B1 (ko) 소셜 빅데이터를 활용한 콘텐츠 마케팅 효과 분석 시스템
WO2019062404A1 (zh) 应用程序的处理方法、装置、存储介质及电子设备
CN103945239B (zh) 影音节目收视信息的实时记录与分析系统及方法
CN104702981A (zh) 一种计算数字电视目标受众的方法和系统
CN109960748A (zh) 基于知识图谱的个人直播用户观看主播预测的方法及系统
CN107404659B (zh) 网络视频显示方法及装置
JP2014110001A (ja) サーバシステム、情報処理システム及びプログラム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20882952

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20882952

Country of ref document: EP

Kind code of ref document: A1