An intelligent recommendation method and system for a video website based on edge computing

Abstract

The application provides a video website intelligent recommendation method and system based on edge computing, and the method comprises the following steps: watching a video on a video website through an electronic device, obtaining authorization, collecting frame images in a random sampling and / or non-random sampling manner in the background, and saving the frame images to a local storage; obtaining a label image when the device is in a low-power consumption state, and uploading the label image to a server; analyzing the label information corresponding to the label image by using a preset image recognition algorithm on the server; recording and classifying the label information, combining the video classification attributes of the video website to generate one or more favorite labels of a user; and pushing and displaying the video on the homepage of the electronic device when the video is pushed. The video website intelligent recommendation method provided by the application is applied to a video website, and the video type watched by a user is analyzed, recorded, saved, analyzed and pushed, so that the video content is related to the user, and the growing video recommendation demand of the user is met.

Description

Technical Field

[0001] This invention relates to the field of algorithm recommendation, and more specifically, to an intelligent recommendation method and system for video websites based on edge computing. Background Technology

[0002] On August 27, 2021, the China Internet Network Information Center (CNNIC) released its 48th Statistical Report on Internet Development in China in Beijing. The report shows that as of June 2021, my country's internet user base reached 1.011 billion, an increase of 21.75 million from December 2020, with an internet penetration rate of 71.6%. This massive user base has created the world's largest and most vibrant digital society. As of December 2020, my country's online video user base reached 927 million, an increase of 76.33 million from March 2020, accounting for 93.7% of all internet users. Among them, short video users reached 873 million, an increase of 100 million from March 2020, accounting for 88.3% of all internet users.

[0003] Online video refers to dynamic images transmitted over the internet in video file formats such as WMV, RM, RMVB, FLV, and MOV. It includes various films and television programs, news, advertisements, Flash animations, self-shot DV videos, chat videos, game videos, and surveillance videos. As the personal values ​​and online behaviors of internet users become increasingly complex and diverse, their video consumption structure will also exhibit diversified characteristics. This diversification of video consumption demand will drive the competitive landscape of China's online video market towards two directions: pursuing scale and pursuing differentiation.

[0004] With the widespread adoption of the internet, online video has experienced explosive growth, with video update rates increasing daily. Major video websites offer a wide variety of videos updated rapidly. However, existing video websites struggle to accurately recommend video content that matches user preferences. This is due to two main reasons: firstly, the faster video update speed means users' tastes evolve accordingly; secondly, current recommendation algorithms often rely on keyword recognition, resulting in a narrower scope. Therefore, existing website video recommendation methods are no longer sufficient to meet users' ever-growing recommendation needs.

[0005] Therefore, there is an urgent need for a solution to the above problems regarding website video recommendation methods. Summary of the Invention

[0006] To address the problems existing in current technologies, this invention provides a video website intelligent recommendation method and system based on edge computing. The specific solution is as follows:

[0007] A video website intelligent recommendation method based on edge computing includes the following:

[0008] After a user watches a video on a video website using an electronic device and obtains authorization, the background collects frame images of the current video using random sampling and / or non-random sampling and saves them to local storage.

[0009] The background preprocesses the frame images in the local memory into tag images and uploads them to the server only when the electronic device is in a low-power state.

[0010] The server receives the tag image and applies a preset image recognition algorithm to analyze the tag information corresponding to the tag image.

[0011] Record and categorize the tag information, and combine it with the video category attributes of the video website to generate one or more user preference tags;

[0012] When pushing content, the server combines the preference tags to deliver relevant video content and displays it on the homepage of the electronic device.

[0013] In one specific embodiment, the current preference tag will overwrite the past preference tag;

[0014] The server only stores a preset number of preference tags within a preset period.

[0015] In one specific embodiment, frame images of the current video are acquired using a non-random sampling method;

[0016] The background obtains the total duration of the current video and pre-sets the number of frames to be captured;

[0017] The acquisition frequency of the acquired frame images is calculated based on the number of frame images and the total duration, as well as the acquisition time of the current video frame image, so that the background can acquire frame images at each acquisition time.

[0018] In one specific embodiment, frame images of the current video are acquired by combining random sampling and non-random sampling. Frame images acquired by random sampling are used as random frame images, and frame images acquired by non-random sampling are used as non-random frame images.

[0019] The background obtains the total duration of the current video and pre-sets the number of frames to be captured for each video;

[0020] The number of frame images is divided into a number of random frame images and a number of non-random frame images according to a preset ratio;

[0021] The acquisition frequency of non-random frame images is calculated based on the number of non-random frame images and the total duration, as well as the non-random sampling time for acquiring non-random frame images, so that the background can acquire non-random frame images at each non-random sampling time.

[0022] In addition to the non-random sampling times, random sampling times are randomly set so that the background can collect random frame images at each random sampling time.

[0023] In one specific embodiment, the state of the electronic device being in a low-power state includes, but is not limited to, the following:

[0024] The electronic device is powered on and idle, or the electronic device is in standby mode.

[0025] In one specific embodiment, the tag information includes: video classification attributes, images of people in the video, devices in the video, regions involved in the video, and languages ​​involved in the video.

[0026] A video website intelligent recommendation system based on edge computing includes electronic devices and a server, wherein the electronic devices include,

[0027] Acquisition Unit: Used to acquire frame images of the current video in a random sampling and / or non-random sampling manner after obtaining authorization when the user watches video on a video website through an electronic device, and save them to local storage;

[0028] Transmission unit: When the electronic device is in a low-power state, it is used to preprocess the frame image in the local memory into a tag image and upload it to the server.

[0029] The server includes,

[0030] Analysis unit: used to receive the label image and apply a preset image recognition algorithm to analyze the label information corresponding to the label image;

[0031] Tag unit: Used to record and classify the tag information, and generate one or more user preference tags by combining the video classification attributes of the video website;

[0032] Push Unit: When pushing content, the server combines the preference tags to deliver relevant video content and displays it on the homepage of the electronic device.

[0033] In one specific embodiment, in the tag unit, the current preference tag will overwrite the past preference tags, and only the preference tags within a preset period and / or a preset number will be saved.

[0034] In one specific embodiment, the acquisition unit acquires frame images of the current video in a non-random sampling manner;

[0035] The background obtains the total duration of the current video and pre-sets the number of frames to be captured for each video;

[0036] The acquisition frequency of the acquired frame images is calculated based on the number of frame images and the total duration, as well as the acquisition time of the current video frame image, so that the background can acquire frame images at each acquisition time.

[0037] In one specific embodiment, the acquisition unit acquires frame images of the current video in a combination of random sampling and non-random sampling, using the frame images acquired by random sampling as random frame images and the frame images acquired by non-random sampling as non-random frame images.

[0038] The background obtains the total duration of the current video and pre-sets the number of frames to be captured for each video;

[0039] The number of frame images is divided into a number of random frame images and a number of non-random frame images according to a preset ratio;

[0040] The acquisition frequency of non-random frame images is calculated based on the number of non-random frame images and the total duration, as well as the non-random sampling time for acquiring non-random frame images, so that the background can acquire non-random frame images at each non-random sampling time.

[0041] In addition to the non-random sampling times, random sampling times are randomly set so that the background can collect random frame images at each random sampling time.

[0042] Beneficial effects:

[0043] This invention provides an intelligent recommendation method and system for video websites based on edge computing. Applied to video websites, it analyzes, records, saves, and pushes relevant video content based on the types of videos watched by users, thereby achieving relevance-based video content delivery and meeting the growing demand for video recommendations. Based on deep learning algorithms, it uses local electronic devices to collect video frame images, and then processes, analyzes, and understands these images in the cloud to identify various patterns of targets and objects, thus determining user preferences and achieving precise recommendations. This intelligent recommendation method, which accurately recommends video content matching user preferences, differs from traditional video websites that rely on keyword recognition. It has a wider coverage, and the collected images provide more comprehensive sample information than text, resulting in preference tags that better reflect users' true tastes.

[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of the intelligent recommendation method for video websites according to Embodiment 1 of the present invention;

[0047] Figure 2 This is a schematic diagram illustrating the principle of the intelligent recommendation method for video websites in Embodiment 1 of the present invention;

[0048] Figure 3 This is a schematic diagram of the intelligent recommendation system for video websites according to Embodiment 2 of the present invention.

[0049] Figure label:

[0050] 1-Electronic device; 2-Server; 11-Collection unit; 12-Transmission unit; 21-Analysis unit; 22-Tag unit; 23-Push unit. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1

[0053] This embodiment proposes an intelligent recommendation method for video websites based on edge computing. The process steps are as shown in the appendix of the instruction manual. Figure 1 The specific solution is as follows:

[0054] A video website intelligent recommendation method based on edge computing includes the following steps:

[0055] 101. When a user watches a video on a video website through an electronic device and obtains authorization, the background collects frame images of the current video in a random sampling and / or non-random sampling manner and saves them to local storage.

[0056] 102. The background preprocesses the frame images in the local storage into tag images and uploads them to the server only when the electronic device is in a low power state.

[0057] 103. The server receives the tag image and uses a preset image recognition algorithm to analyze the tag information corresponding to the tag image;

[0058] 104. Record and categorize tag information, and generate one or more user preference tags by combining them with the video category attributes of the video website;

[0059] 105. When pushing content, the server combines preference tags to deliver relevant video content and displays it on the homepage of electronic devices.

[0060] The intelligent recommendation method for video websites provided in this embodiment is applied to video websites and can analyze, record, save, and push relevant information about the types of videos watched by users. Based on deep learning algorithms, it uses local electronic devices to capture frame images of videos, and utilizes the cloud to process, analyze, and understand the images to identify targets and objects of various patterns, thereby determining user preferences and achieving accurate recommendations. The complete process principle is shown in the attached manual. Figure 2 As shown.

[0061] Specifically, in step 101, after a user watches a video on a video website through an electronic device and obtains authorization, the background collects frame images of the current video in a random sampling and / or non-random sampling manner and saves them to local storage.

[0062] When a user is detected accessing a video website via an electronic device, the device will request authorization to read the currently viewed video content and capture frame images. The device will only be allowed to capture images if the user grants authorization.

[0063] A frame image is a single-frame image. Video is broken down into individual frames, and these frames are used as samples to analyze user preferences. Frame image acquisition methods include random sampling, non-random sampling, or a combination of both. To avoid the potential impact of frequent acquisition on system performance, the acquisition time is set according to the video duration. For longer videos, the number of frames acquired can be increased, and the acquisition frequency reduced. For shorter videos, the acquisition frequency can be increased, and the number of frames acquired can be reduced.

[0064] The acquisition time refers to the point in time when the background captures frame images. At each acquisition time, one or more frame images can be captured. The number of frame images that can be captured varies at different acquisition times; some acquisition times require only one frame image, while others can capture multiple frame images. Since consecutive frame images may capture more useful information, multiple frame images can be captured at some acquisition times, such as the middle part of a video, which is often the core part of the video. Conversely, only one frame image can be captured at some acquisition times, such as the beginning and end of a video. Since the beginning and end of a video often contain introductory information rather than substantial video content, a small number of frame images can be captured in these sections. Preferably, the acquisition times are arranged chronologically, and the number of frame images captured at each acquisition time follows a normal distribution.

[0065] For example, frames of the current video are captured using a non-random sampling method. The background obtains the total duration of the current video and pre-sets the number of frames to be captured for each video. Based on the number of frames and the total duration, the background calculates the frame capture frequency and the capture times for the current video frames, enabling the background to capture frames at each capture time. For instance, if the sampling number for each video is set to 20, and the current playback time is 10 minutes, a frame of the current video will be captured every 30 seconds.

[0066] For example, frame images of the current video are captured using random sampling. The total duration of the current video is obtained, and sampling times are randomly set within the total duration, so that the background captures frame images at each sampling time.

[0067] For example, frame images of the current video are acquired using a combination of random and non-random sampling. The backend obtains the total duration of the current video and pre-sets the number of frame images to be acquired. Frame images acquired through random sampling are named "random frame images," and frame images acquired through non-random sampling are named "non-random frame images." The number of random and non-random frame images is divided according to a preset ratio. The acquisition frequency of non-random frame images and the various non-random sampling times for acquiring non-random frame images of the current video are calculated based on the number of non-random frame images and the total duration, so that the backend acquires non-random frame images at each non-random sampling time. In addition to non-random sampling times, random sampling times are set, so that the backend acquires random frame images at each random sampling time.

[0068] Specifically, in step 102, if and only if the electronic device is in a low-power state, the background preprocesses the frame image in the local memory into a tag image and uploads it to the server.

[0069] After acquiring the frame image, it is saved to the local storage of the electronic device and is not directly uploaded to the server. This is because real-time uploading could impact the user's network bandwidth and cause video loading stuttering. To avoid this, the intelligent recommendation method in this embodiment chooses to upload the frame image when the device is in a low-power state. Since the intelligent recommendation method does not have high real-time requirements, delayed uploading will not affect the acquisition of preference tags.

[0070] In this embodiment, low-power states include standby state and idle state. During this time, the user is not using the device, and the background system can perform preprocessing such as compression on the frame image to obtain the tag image.

[0071] Specifically, in step 103, the server receives the label image and applies a preset image recognition algorithm to analyze the label information corresponding to the label image. The server analyzes the characteristics of the uploaded label image, and the image recognition algorithm uses a deep learning algorithm.

[0072] Image recognition algorithms can identify features such as text, images, and buildings in images. Pre-defined image recognition algorithms have undergone extensive training and are capable of recognizing relevant features. Tag information includes: video classification attributes, people in the video, devices in the video, regions involved in the video, and languages ​​used in the video. The scope of tag information acquired depends on the recognition capabilities of the image recognition algorithm.

[0073] Specifically, step 104 involves recording and classifying tag information, and generating one or more user preference tags by combining them with the video classification attributes of the video website. In this embodiment, each tag information is classified into multiple items, including video categories in regular video websites, such as suspense, romance, drama, science fiction, and animation, as well as regions, eras, and lead actors. For example, image recognition algorithms can identify public figures involved in a video and classify them into the category corresponding to that public figure. Since user preferences change, and with the rise of short videos, user preferences iterate and update faster, the acquired preference tags will also be updated accordingly. Current preference tags will overwrite past preference tags, and the server only saves preference tags within a preset period and / or a preset number. For example, the server only saves the 20 most recently acquired preference tags, or only the preference tags from the last half month. The acquisition of preference tags is based on machine learning; the longer a user uses the method of this embodiment, the more accurate and relevant the obtained preference tags will be to the user's actual preferences. For example, if a user watches a wide range of types of content, the algorithm will gradually summarize this pattern over time, thereby pushing a wider range of video types.

[0074] 105. During push notifications, the server combines user preference tags to deliver relevant video content, which is then displayed on the electronic device's homepage. When the server pushes homepage information to the electronic device, it filters content based on the currently logged-in account's preference tags to ensure relevance. After a period of use, the device gradually learns about the user and can directly push videos closest to their preferences to the homepage without requiring manual searching. Users can then play videos according to their preferences. Furthermore, the longer the device is used, the more accurate the collected tag images become, the more closely the preference tags match the user, and the more precise the content pushed.

[0075] This embodiment provides an intelligent recommendation method for video websites based on edge computing. Applied to video websites, it analyzes, records, saves, and pushes relevant video content based on the types of videos watched by users, thereby achieving relevance-based video content delivery and meeting the growing demand for video recommendations. Based on deep learning algorithms, it uses local electronic devices to collect video frame images, and then processes, analyzes, and understands these images in the cloud to identify various patterns of targets and objects, thus determining user preferences and achieving precise recommendations. This intelligent recommendation method, which accurately recommends video content matching user preferences, differs from traditional video websites that rely on keyword recognition. It has a wider coverage, and the collected images provide more comprehensive sample information than text, resulting in preference tags that better reflect users' true tastes.

[0076] Example 2

[0077] This embodiment, based on Embodiment 1, proposes an intelligent recommendation system for video websites based on edge computing, systematizing the intelligent recommendation method of Embodiment 1. The system's modules are listed in the appendix to the specification. Figure 3 The specific solution is as follows:

[0078] An intelligent recommendation system for video websites based on edge computing includes an electronic device and a server. The electronic device 1 includes a data acquisition unit 11 and a transmission unit 12, while the server 2 includes an analysis unit 21, a tagging unit 22, and a push unit 23. When it detects that a user has logged into a relevant video website and is watching a video through the electronic device 1, the electronic device 1 will request authorization to read the currently viewed video content and capture frame images of the video. The electronic device 1 will only be allowed to capture images if the user agrees to the authorization.

[0079] Acquisition Unit 11: When a user watches a video on a video website through an electronic device, after obtaining authorization, the background acquires frame images of the current video in a random sampling and / or non-random sampling manner, and saves them to local storage;

[0080] Transmission unit 12: When the electronic device is in a low-power state, it is used to preprocess the frame image in the local memory into a tag image and upload it to the server.

[0081] Analysis unit 21: Used to receive tag images and apply a preset image recognition algorithm to analyze the tag information corresponding to the tag images;

[0082] Tag Unit 22: Used to record and classify tag information, and generate one or more user preference tags by combining the video classification attributes of the video website;

[0083] Push Unit 23: When pushing content, the server combines preference tags to deliver relevant video content and displays it on the homepage of electronic devices.

[0084] In tag unit 22, the current preference tag will overwrite the past preference tags, and only the preference tags within the preset period and / or the preset number will be saved.

[0085] For example, the acquisition unit 11 acquires frame images of the current video in a non-random sampling manner; the background obtains the total duration of the current video and pre-sets the number of frame images to be acquired for each video; the acquisition frequency of the frame images and the acquisition time of the current video frame images are calculated based on the number of frame images and the total duration, so that the background acquires frame images at each acquisition time.

[0086] For example, the acquisition unit 11 acquires frame images of the current video using a combination of random sampling and non-random sampling. Frame images acquired through random sampling are used as random frame images, and frame images acquired through non-random sampling are used as non-random frame images. The background obtains the total duration of the current video and pre-sets the number of frame images to be acquired for each video. The number of frame images is divided into the number of random frame images and the number of non-random frame images according to a preset ratio. The acquisition frequency of non-random frame images and the various non-random sampling times for acquiring non-random frame images are calculated based on the number of non-random frame images and the total duration, so that the background acquires non-random frame images at each non-random sampling time. In addition to non-random sampling times, random sampling times are randomly set, so that the background acquires random frame images at each random sampling time.

[0087] This embodiment provides an intelligent recommendation system for video websites based on edge computing, which systematizes the intelligent recommendation method of Embodiment 1 and makes it more practical.

[0088] This invention provides an intelligent recommendation method and system for video websites based on edge computing. Applied to video websites, it analyzes, records, saves, and pushes relevant video content based on the types of videos watched by users, thereby achieving relevance-based video content delivery and meeting the growing demand for video recommendations. Based on deep learning algorithms, it uses local electronic devices to collect video frame images, and then processes, analyzes, and understands these images in the cloud to identify various patterns of targets and objects, thus determining user preferences and achieving precise recommendations. This intelligent recommendation method, which accurately recommends video content matching user preferences, differs from traditional video websites that rely on keyword recognition. It has a wider coverage, and the collected images provide more comprehensive sample information than text, resulting in preference tags that better reflect users' true tastes.

[0089] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0090] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

[0091] The above-disclosed examples are only a few specific implementation scenarios of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for intelligent recommendation of a video website based on edge computing, characterized in that, The method comprises the following steps: A user watches a video on a video website through an electronic device, and after obtaining authorization, a background collects frame images of the current video in a random sampling and / or non-random sampling manner and saves them to a local storage; The background pre-processes the frame images in the local storage into label images and uploads them to a server only when the electronic device is in a low-power consumption state; The server receives the label images, applies a preset image recognition algorithm to analyze the label information corresponding to the label images; The label information is recorded and classified, and one or more favorite labels of the user are generated in combination with the video classification attributes of the video website; When performing a push, the server combines the favorite labels to deliver related video content and performs a push display through a home page of the electronic device; The collection of frame images comprises: collecting frame images of the current video in a combination of random sampling and non-random sampling, taking the frame images obtained through random sampling as random frame images, and taking the frame images obtained through non-random sampling as non-random frame images; the background obtains the total duration of the current video, pre-sets the number of frame images to be collected for each video; divides the number of frame images into the number of random frame images and the number of non-random frame images according to a preset ratio; calculates the collection frequency of non-random frame images and each non-random sampling time for collecting non-random frame images according to the number of non-random frame images and the total duration, so that the background collects non-random frame images at each non-random sampling time; in addition to the non-random sampling time, randomly set random sampling time, so that the background collects random frame images at each random sampling time; each collection time is in chronological order, and the number of frame images collected at each collection time conforms to a normal distribution. 2.The method of claim 1, wherein, The current favorite label will cover the past favorite label; The server only saves the favorite labels within a preset period and / or a preset number. 3.The method of claim 1, wherein, Collect frame images of the current video in a non-random sampling manner; The background obtains the total duration of the current video and pre-sets the number of frame images to be collected; According to the number of frame images and the total duration, calculate the collection frequency of frame images and each collection time for collecting frame images of the current video, so that the background collects frame images at each collection time. 4.The method of claim 1, wherein, Only when the electronic device is in a low-power consumption state, specifically comprising: The electronic device is in a powered-on and idle state, or the electronic device enters a standby state.

5. The method of claim 1, wherein, The label information includes: video classification attributes, characters in the video, devices in the video, regions involved in the video, and languages involved in the video.

6. An edge computing-based intelligent recommendation system for a video website, characterized in that, The method comprises an electronic device and a server, wherein the electronic device comprises, The collection unit is used to collect frame images of a current video in a random sampling and / or non-random sampling manner after authorization is obtained when a user watches the video on a video website through an electronic device, and save the frame images to a local storage; the collection of the frame images includes: collecting frame images of the current video in a combination of random sampling and non-random sampling, taking the frame images obtained by random sampling as random frame images, and taking the frame images obtained by non-random sampling as non-random frame images; the background obtains a total duration of the current video, and a number of frame images to be collected for each video is preset; the number of frame images is divided into a number of random frame images and a number of non-random frame images according to a preset ratio; a collection frequency of collecting non-random frame images and each non-random sampling time for collecting non-random frame images are calculated according to the number of non-random frame images and the total duration, so that the background collects non-random frame images at each non-random sampling time; random sampling times are randomly set except for the non-random sampling times, so that the background collects random frame images at each random sampling time; the number of frame images collected at each collection time is in a normal distribution according to a time sequence; The transmission unit is used to pre-process the frame images in the local storage into label images and upload the label images to the server when and only when the electronic device is in a low-power consumption state. The server includes, The analysis unit is used to receive the label images, and analyze label information corresponding to the label images by applying a preset image recognition algorithm; The label unit is used to record and classify the label information, and generate one or more favorite labels of a user in combination with video classification attributes of the video website; The push unit is used to push related video content in combination with the favorite labels, and push and display the video content on a home page of the electronic device. 7.The video website intelligent recommendation system of claim 6, wherein, In the label unit, a current favorite label covers a past favorite label, and only favorite labels within a preset period and / or a preset number are saved. 8.The video website intelligent recommendation system of claim 6, wherein, The collection unit collects frame images of the current video in a non-random sampling manner; The background obtains a total duration of the current video, and a number of frame images to be collected for each video is preset; A collection frequency of collecting frame images and each collection time for collecting frame images of the current video are calculated according to the number of frame images and the total duration, so that the background collects frame images at each collection time.

Images