A real-time information flow management and optimization system based on AI intelligent matching

Abstract

The present invention belongs to the technical field of information flow management, and specifically is a real-time information flow management and optimization system based on AI intelligent matching. The system captures the hot spot feature information of the target platform on that day, identifies the browsing preference feature information of the designated user, and determines the updateable video capacity of the dedicated push video micro-library of the designated user on that day. The information flow frame extraction method is applied to the target platform video resource library to screen the adapted push videos of the designated user on that day, which not only effectively reduces the calculation complexity and time cost, but also ensures that the pushed content is always synchronized with the current popular trends and user interests, and realizes the update of the dedicated push video micro-library of the designated user on that day. Then, by analyzing the overall frame extraction state accuracy and the overall push state accuracy of the adapted push video, the accuracy of the update mechanism of the dedicated push video micro-library of the designated user on the target platform on that day is evaluated to feedback the optimization space, so as to promote the progressive development of the target platform video push.

Description

Technical Field

[0001] The present invention belongs to the technical field of information flow management, and specifically is a real-time information flow management and optimization system based on AI intelligent matching. Background Art

[0002] Information flow refers to the collection of various information that continuously flows and spreads in the digital space, covering multiple forms such as text, images, audio, and video. As an important component of information flow, video plays an irreplaceable role on major digital platforms with its vivid and intuitive expression form and powerful information carrying capacity.

[0003] However, in today's high-speed digital age, the explosive growth of information poses a major challenge to the public in accurately screening and obtaining the content they need. Faced with the vast amount of video resources, users often fall into the dilemma of "information overload" and find it difficult to quickly identify and access high-quality content that truly meets their interests and needs. For video platforms, how to quickly achieve accurate and personalized video push and improve user satisfaction and retention has become a key problem that needs to be overcome.

[0004] In the prior art, there are also some related solutions that can be applied to platform video push, for example, the short video advertising push method and system with Chinese patent publication number CN117893260A, which analyzes the short video behavior data of the target user, constructs a target attention bias behavior directed graph containing attention bias content, extracts attention bias feature data and performs keyword analysis, and determines the target keyword with a frequency greater than the preset frequency, so as to push short video advertising data to the target user. Although it can be used in platform video push to help improve the push reliability, it still has the limitations of the prior art, specifically: 1. The prior art mainly relies on analyzing the user platform video browsing behavior data to determine the attention bias keywords, and uses this as the only basis for video push. The lack of a real-time hot spot capture mechanism on the platform makes it impossible for users to respond to real-time hot events on the platform in a timely manner, resulting in the pushed video content possibly lagging behind the current hot topics and changes in user interests.

[0005] 2. The existing technology fails to pay attention to the limited capacity of the user video push micro-library, and further fails to address the issue of determining the updateable capacity of the video push micro-library, resulting in the inability to timely guarantee the quantity and quality of user pushed videos.

[0006] 3. The existing technology adopts an overall video comparison and analysis method for the matching process of user-adapted pushed videos, which not only requires powerful computing power and time cost, but also has the risk of high algorithm error rate and poor stability, which imposes a great burden on the platform operation cost.

[0007] 4. The existing technology lacks a comprehensive evaluation and feedback mechanism for the effect of user video push, which limits the improvement of push accuracy and easily causes blind push, thereby damaging the user experience and even user loss. Summary of the invention

[0008] In order to overcome the shortcomings of the background technology, an embodiment of the present invention provides a real-time information flow management and optimization system based on AI intelligent matching, which can effectively solve the problems involved in the above-mentioned background technology.

[0009] The purpose of the present invention can be achieved through the following technical solutions: a real-time information flow management and optimization system based on AI intelligent matching, including: a video update activation module, a daily hot spot capture module, a user preference identification module, a video push update module, a push effect evaluation module and a cloud database.

[0010] The video update activation module is connected to the daily hot spot capture module and the user preference identification module respectively, the daily hot spot capture module and the user preference identification module are both connected to the video push update module, the video push update module is connected to the push effect evaluation module, and the cloud database is connected to the user preference identification module.

[0011] The video update activation module is used to activate the exclusive push video micro-library update mechanism for a registered user on the target platform at a preset time point on the same day, and record the registered user as the designated user.

[0012] The daily hotspot capture module is used to capture the hotspot feature information of the target platform on that day.

[0013] The user preference identification module is used to analyze the video browsing behavior of a specified user within a preset historical period and identify the browsing preference feature information of the specified user.

[0014] The video push update module is used to determine the updateable video capacity of the designated user's exclusive push video micro-library on the day, combine the target platform's hot feature information on the day and the designated user's browsing preference feature information, perform information flow frame analysis on each licensed push video in the target platform's video resource library, and select each adapted push video to add to the designated user's exclusive push video micro-library on the day to achieve update.

[0015] The push effect evaluation module is used to randomly extract a preset number of adapted push videos for full-frame analysis of the information flow, compare them with the corresponding information flow frame analysis results, and combine the actual platform video browsing data of the specified user on that day to evaluate the accuracy of the exclusive push video micro-library update mechanism for the specified user on the target platform on that day, so as to provide feedback for optimization space.

[0016] Cloud database, used to store the target platform's various video content theme standard keyword sets.

[0017] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention screens the hot videos in the video resource library of the target platform, combines keywords and image elements to capture the hot spots of the target platform on that day, and uses this as one of the bases for updating the micro-library of videos exclusively pushed to the designated user on that day, thereby making up for the lack of real-time hot spot capture mechanism in the prior art platform and significantly improving the timeliness and relevance of the content pushed to users.

[0018] (2) The present invention uses the browsing preference of video content theme type in the historical preset period, the video publisher ID preference, and the image element preference as the entry point to carry out the feature recognition of the browsing preference of the specified user, so as to achieve the in-depth understanding and detailed control of the user's browsing preference, and thus ensure the accuracy of the user's browsing preference recognition.

[0019] (3) The present invention determines the updateable video capacity of the micro-library of the exclusive push video for the specified user on that day, takes the hot feature information of the target platform on that day and the browsing preference feature information of the specified user as the basis for selecting the adapted push video, and combines the information flow frame extraction analysis method to realize the update of the micro-library of the exclusive push video for the specified user on that day. This not only effectively reduces the computational complexity and time cost of adapting the push video, but also ensures that the push content always keeps pace with the current popular trends and user interests, and further ensures the high quality of the pushed video.

[0020] (4) The present invention conducts an accuracy evaluation of the update mechanism of the exclusive push video micro-library for designated users of the target platform on the same day by comparing the results of full-frame analysis and frame-by-frame analysis of information flow, as well as the actual browsing behavior of designated users after the exclusive push video micro-library is updated on the same day, objectively understands the optimization space for the video push update of the target platform, and promotes the improvement and development of content push on the target platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0022] Figure 1 It is a schematic diagram of module connection of the present invention.

[0023] Figure 2 This is a logical schematic diagram of the daily hotspot capture module of the present invention.

[0024] Figure 3 This is a logical schematic diagram of the user preference identification module of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] Reference Figure 1 As shown, the present invention provides a real-time information flow management and optimization system based on AI intelligent matching, including: a video update activation module, a daily hot spot capture module, a user preference identification module, a video push update module, a push effect evaluation module and a cloud database.

[0027] The video update activation module is connected to the daily hot spot capture module and the user preference identification module respectively, the daily hot spot capture module and the user preference identification module are both connected to the video push update module, the video push update module is connected to the push effect evaluation module, and the cloud database is connected to the user preference identification module.

[0028] The video update activation module is used to activate the exclusive push video micro-library update mechanism for a registered user on the target platform at a preset time point on the same day, and record the registered user as a designated user.

[0029] It should be noted that the above-mentioned preset time point for the day can be determined by referring to the target platform mechanism to regulate the update time point of the registered users' exclusive push video micro-library, or by considering the regular login time point of the users specified by the target platform, which is specifically determined by the personalized update plan of the target platform.

[0030] Reference Figure 2 As shown, the daily hot spot capture module is used to capture the hot spot feature information of the target platform on that day.

[0031] Specifically, the capturing of the target platform's daily hot feature information includes: collecting the cumulative playback volume, cumulative likes volume and cumulative comments volume of each video in the target platform's video resource library as of a preset time point on the day, as well as the corresponding growth rate of the playback volume, likes volume and comments volume of each video in the target platform's video resource library during a preset period of the day, analyzing the heat coefficient of each video in the target platform's video resource library, screening each hot video therein, and performing keyword extraction on the video title, video tag and video description of each hot video, and performing image element recognition on the video cover of each hot video, thereby collecting the keyword set and image element set of each hot video, and retrieving each keyword and each image element whose occurrence frequency is greater than a preset frequency threshold, and recording them as each common keyword and each common image element.

[0032] It should be noted that the analysis process of the heat coefficient of each video in the target platform video resource library as of the preset time point on the day is as follows: summarizing the cumulative playback volume, cumulative likes volume and cumulative comments volume of each video in the target platform video resource library as of the preset time point on the day, and obtaining the total playback volume, total likes volume and total comments volume of the videos in the target platform video resource library as of the preset time point on the day; using this as a reference, analyzing the cumulative playback volume proportion, cumulative likes volume proportion and cumulative comment volume proportion of each video in the target platform video resource library as of the preset time point on the day, and obtaining the first-order heat coefficient of each video in the target platform video resource library as of the preset time point on the day through cumulative analysis; similarly, according to the corresponding growth rates of the playback volume, likes volume and comments volume of each video in the target platform video resource library within the preset time period on the day, analyzing the second-order heat coefficient of each video in the target platform video resource library within the preset time period on the day, and adding the first-order heat coefficient and the second-order heat coefficient to obtain the heat coefficient of each video in the target platform video resource library.

[0033] It should also be noted that the above-mentioned image elements include character elements, color elements, scene elements and object elements, and the preset time period of the day specifically refers to the time period with the preset time point of the day as the end time and the length before that is the preset time length.

[0034] Get the search keywords and their continuous search heat received by the search bar of the target platform within the preset time period of the day, filter out the search keywords whose continuous search heat is greater than the preset continuous search heat threshold, record them as hot search keywords, summarize the hot search keywords and common keywords, eliminate duplicate keywords, and use the remaining keywords as the hot keywords of the platform on that day.

[0035] It should be noted that the analysis process of the continuous search heat of the above-mentioned search keywords is: collecting the cumulative search times of each search keyword received by the search bar of the target platform during the preset time period of the day, the maximum search times at the same time, and the shortest interval search duration, summarizing the total search times of the search keywords received by the search bar of the target platform during the preset time period of the day and the total maximum search times at the same time, and using this as the reference denominator to obtain the cumulative search times ratio of each search keyword received by the search bar of the target platform during the preset time period of the day and the maximum search times ratio at the same time, performing a ratio analysis of the duration of the preset time period of the day and the shortest interval search duration to obtain the search frequency factor of each search keyword received by the search bar of the target platform during the preset time period of the day, and adding the cumulative search times ratio, the maximum search times ratio at the same time, and the search frequency factor to obtain the continuous search heat of each search keyword.

[0036] Based on the search engine library of the target platform, the associated image elements corresponding to each hot search keyword are searched, and they are summarized with the common image elements, and the duplicate image elements are removed, and the remaining image elements are used as the hot image elements of the platform on that day.

[0037] The hot keywords and hot image elements of the platform on that day are used together as the hot feature information of the target platform on that day.

[0038] The embodiment of the present invention screens the hot videos in the video resource library of the target platform, combines keywords and image elements to capture the hot spots of the target platform on that day, and uses this as one of the bases for updating the exclusive push video micro-library for the designated user on that day, thereby making up for the lack of real-time hot spot capture mechanism of the existing technology platform and significantly improving the timeliness and relevance of the user's push content.

[0039] Reference Figure 3 As shown, the user preference identification module is used to analyze the video browsing behavior of a specified user within a preset historical period and identify the browsing preference feature information of the specified user.

[0040] Specifically, the identification of browsing preference characteristic information of a specified user includes: collecting video browsing data of the specified user within a historical preset period, including the publisher ID, browsing time ratio, like status index and comment status index of each browsed video, analyzing the interaction coefficient of the specified user for each browsed video within the historical preset period, screening each interactive video, extracting keywords from the video title, video tag and video description of each interactive video, comparing the extracted keywords with the standard keyword set of various video content themes of the target platform stored in the cloud database, and analyzing the browsing preference ratio of the specified user for various video content themes within the historical preset period.

[0041] It should be noted that the above-mentioned browsing time ratio refers to the ratio of the user's playback time of the browsed video to the total playback time of the corresponding browsed video. The like status indicator is 1 or 0, 1 means like, and 0 means no like. The comment status indicator is 0 or a positive number, a positive number indicates the number of comments, and 0 means no comment. The interaction coefficient of the user for each browsed video in the historical preset period is thus specified as the cumulative value of the browsing time ratio, the like status indicator and the comment status indicator of each browsed video.

[0042] It should also be noted that the analysis process of the browsing preference ratio for each type of video content theme within the above-mentioned preset historical period for the specified user is as follows: obtaining the number of keywords in the standard keyword set of each type of video content theme on the target platform in each interactive video, counting the total number of keywords in the standard keyword set of each type of video content theme on the target platform in the interactive video, calculating the keyword ratio of the standard keyword set of each type of video content theme on the target platform in the interactive video, and using this as the browsing preference ratio for each type of video content theme within the preset historical period for the specified user.

[0043] The publisher ID corresponding to each interactive video of the specified user within the historical preset period is recorded as the interactive publisher ID, and the interaction coefficients of the interactive videos corresponding to the same interactive publisher ID are accumulated to obtain the comprehensive interaction coefficient of each interactive publisher ID, and each interactive publisher ID whose comprehensive interaction coefficient is greater than or equal to the preset comprehensive interaction coefficient threshold is taken as the biased video publisher ID within the historical preset period of the specified user.

[0044] The key frame images of the interactive videos of the specified user within the historical preset period are extracted and image elements are recognized, and the common image elements therein are retrieved as the biased image elements within the historical preset period of the specified user.

[0045] It should be noted that the key frame images of each interactive video of the specified user within the above-mentioned historical preset period are obtained based on significant content changes and user interaction hotspots. The specific acquisition process is: the frame difference method is used to process each adjacent frame image in the interactive video, the pixel difference between each adjacent frame image is calculated, and the adjacent frame images whose pixel difference is greater than the preset pixel difference threshold are screened, and the frames with the corresponding frames later are recorded as key frame images, and the interaction parameters of each frame image in the interactive video relative to the registered users of the target platform are collected, including the number of likes and the number of comments, and the ratios are analyzed with the total number of likes and the total number of comments of the interactive video to obtain the user like ratio and the user comment ratio of each frame image in the interactive video, and the interaction heat of each frame image in the interactive video is obtained by multiplying them, and the frame images with a value greater than the preset interaction heat threshold are screened as key frame images, which are integrated to obtain the key frame images of each interactive video of the specified user within the historical preset period.

[0046] The browsing preference proportions for various video content topics within a historical preset period, the video publisher IDs of each preference, and the image elements of each preference are collectively used as browsing preference feature information for a specified user.

[0047] The embodiment of the present invention uses the historical preset period video content theme type bias, video publisher ID bias, and image element bias as entry points to carry out designated user browsing preference feature recognition, achieve in-depth understanding and detail control of user browsing preferences, and thus ensure the accuracy of user browsing preference recognition.

[0048] The video push update module is used to determine the updateable video capacity of the designated user's exclusive push video micro-library on the day, combine the target platform's hot spot feature information on the day and the designated user's browsing preference feature information, perform information flow frame analysis on each permitted push video in the target platform's video resource library, screen each adapted push video and add it to the designated user's exclusive push video micro-library on the day to achieve update.

[0049] Specifically, determining the updateable video capacity of the dedicated push video micro-library for the specified user on that day includes: querying the free video capacity of the dedicated push video micro-library for the specified user at a preset time point on that day; If it is greater than or equal to the preset benchmark, update the video capacity , then As the updateable video capacity of the exclusive push video micro library for the specified user on that day, otherwise As the updateable video capacity of the exclusive push video micro library for the specified user on that day, the deviation value between the preset benchmark update video capacity and the free video capacity is used as the excluded video capacity , according to predefined principles, the video micro-library exclusively pushed to the designated user at the preset time point of the day is removed from the video.

[0050] Specifically, the predefined principles include the subject balance principle, the activity principle and the timeliness principle.

[0051] It should be noted that the above-mentioned exclusive push video micro-library for the designated user at the preset time point of the day specifically refers to the exclusive push video micro-library that has not been updated by the designated user at the preset time point of the day. The video elimination process includes: classifying the videos in the exclusive push video micro-library for the designated user at this time by subject, so as to summarize the videos of each type of video content theme, and according to the browsing preference proportion of each type of video content theme within the historical preset period of the designated user , according to the formula Identify the number of video culls for each video content topic.

[0052] The second-order heat coefficient of each video belonging to each type of video content theme within the preset time period of the day is extracted and used as the activity of each video belonging to each type of video content theme.

[0053] Collect the span duration between the release date and the current date of each video belonging to each video content theme, summarize the total span duration between the video release date and the current date of each video content theme, and use this as a reference numerator for ratio analysis to obtain the timeliness of each video belonging to each video content theme.

[0054] Arrange the accumulated values ​​of activity and timeliness in descending order, obtain the video sequence of each type of video content theme, and filter out the videos whose arrangement order in the video sequence is at the latter end of the video elimination number according to the video elimination number of each type of video content theme for elimination. If the actual video elimination capacity does not meet the standard at this time, then according to the remaining video elimination capacity, randomly eliminate the videos of the video content theme type with the smallest browsing preference ratio within the preset historical period of the specified user.

[0055] Specifically, the screening process of each adapted push video includes: recording each video in the target platform video resource library except for each video in the micro-library of push videos exclusively for a designated user at a preset time point on the day and each historically viewed video as each permitted push video.

[0056] Perform frame extraction on each licensed pushed video to obtain its corresponding video frame image, collect the publisher ID of each licensed pushed video, each keyword and each image element in each video frame image, compare and analyze with the hot spot feature information of the target platform on that day and the browsing preference feature information of the specified user, obtain the hot spot adaptation coefficient and user preference coefficient of each licensed pushed video in the frame extraction state, accumulate and obtain the comprehensive adaptation coefficient of each licensed pushed video, arrange each licensed pushed video in order of the comprehensive adaptation coefficient from large to small, and select each licensed push video whose arrangement order is before the updateable video capacity value as each adapted push video.

[0057] Specifically, the frame extraction processing for each licensed pushed video includes: uniformly setting a fixed number of frame extractions for each licensed pushed video, automatically listing the first and last video frames of each licensed pushed video as required frames, and using the duration of each licensed pushed video and the duration that is divisible by the remaining number of frame extractions as the time interval for uniform frame extraction, thereby realizing frame extraction processing for each licensed pushed video.

[0058] Specifically, the hotspot adaptation coefficient of each permitted pushed video in the frame extraction state includes: screening the number of hot keywords on the platform on the day contained in each video frame of each permitted pushed video and the number of hotspot image elements ,in The ID of each licensed video. , is the number of each video frame, , according to the formula Get the hotspot adaptation coefficient of each licensed video push frame extraction state, where is the total number of hot keywords on the platform that day, is the total number of hot image elements on the platform on that day, is the number of video frames.

[0059] Specifically, the user bias coefficient of each permitted pushed video in the frame extraction state is obtained, including: comparing the publisher ID of each permitted pushed video with the publisher ID of each biased video within the preset period of the specified user history, setting the publisher bias weight of each permitted pushed video .

[0060] It should be noted that the above The setting process is as follows: if the publisher ID of a permitted video is a biased video publisher ID within the specified user's historical preset period, then the biased influence weight of the publisher of the permitted video is set to , otherwise set to ,in , thereby obtaining the publishing author bias weight of each licensed push video , .

[0061] Determine the video content theme type of each licensed biased video based on the keywords in each video frame of each licensed biased video, and obtain the browsing preference ratio of the specified user for each licensed biased video by combining the browsing preference ratio of each video content theme within the historical preset period of the specified user. .

[0062] It should be noted that the process of determining the video content theme type of each of the above-mentioned licensed biased videos is as follows: based on the standard keyword sets of various video content themes of the target platform, the total number of keywords belonging to the standard keyword sets of various video content themes in the video frame images of each licensed biased video is collected, and the video content theme type corresponding to the maximum value is selected as the video content theme type of each licensed biased video.

[0063] Screen each licensed video push and check the number of image elements that are biased towards the specified user in each video frame , according to the formula Get the user bias coefficient under each permitted video push frame extraction state, The number of image elements to prefer for a given user.

[0064] The embodiment of the present invention determines the updateable video capacity of the designated user's exclusive push video micro-library on that day, takes the target platform's daily hot spot feature information and the designated user's browsing preference feature information as the basis for adapting the push video screening, and combines the information flow frame extraction analysis method to realize the update of the designated user's exclusive push video micro-library on that day. It not only effectively reduces the calculation complexity and time cost of adapting the push video, but also ensures that the push content is always synchronized with the current popular trends and user interests, and further ensures the high quality of the pushed video.

[0065] The push effect evaluation module is used to randomly extract a preset number of adapted push videos for full-frame information flow analysis, compare them with the corresponding information flow frame analysis results, and combine the actual platform video browsing data of the specified user on that day to evaluate the accuracy of the exclusive push video micro-library update mechanism for the specified user on the target platform on that day, so as to provide feedback for optimization space.

[0066] Specifically, the evaluation of the accuracy of the micro-library update mechanism for pushing videos exclusively to designated users on the target platform on the day includes: randomly selecting a preset number of adapted push videos as sampled push videos, analyzing the hotspot adaptation coefficient and user bias coefficient of each sampled push video in the full-frame state, and calculating the absolute deviation with the hotspot adaptation coefficient and user bias coefficient in the sampled frame state, thereby analyzing the overall frame-sampling state accuracy of the adapted push video. .

[0067] It should be noted that the analysis process of the hotspot adaptation coefficient and the user bias coefficient in the full-frame state of the above-mentioned sampled pushed videos is consistent with the analysis process of the hotspot adaptation coefficient and the user bias coefficient in the sampled frame state of the licensed pushed videos.

[0068] It should also be noted that The analysis process is as follows: the absolute deviation value of the hotspot adaptation coefficient of each sampled pushed video in the full-frame state and the sampled state is ratio analyzed with the hotspot adaptation coefficient in the sampled state to obtain the hotspot adaptation coefficient deviation degree of each sampled pushed video in the sampled frame state, and the inverse of the calculated mean is used as the accuracy of the overall sampled frame state of the adapted pushed video. It should be noted that a preset constant is added to the denominator in the process of calculating the inverse of the mean to avoid the occurrence of a state where the denominator is 0.

[0069] Extract the actual platform video browsing data of the specified user on the day, including the browsing time ratio, like status index and comment status index of each adapted push video, filter the interactive push videos, and use the ratio of the number of interactive push videos to the number of adapted push videos as the overall push status accuracy of the adapted push videos. .

[0070] Depend on , The accumulated value is used as the accuracy of the video micro-library update mechanism that is exclusively pushed to designated users on the target platform on that day.

[0071] The embodiment of the present invention conducts an accuracy evaluation of the update mechanism of the exclusive push video micro-library for designated users of the target platform on the same day through the comparison results of the full-frame analysis and frame-sampling analysis of the information flow, as well as the actual browsing behavior of the designated users after the exclusive push video micro-library is updated on the same day, objectively understands the optimization space for the video push update of the target platform, and promotes the progress and development of content push on the target platform.

[0072] The cloud database is used to store the target platform's various video content theme standard keyword sets, and its data mainly comes from the target platform's video theme tag standard guide and is uploaded externally.

[0073] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A real-time information flow management and optimization system based on AI intelligent matching, characterized in that: The system includes: The video update activation module is used to activate the exclusive push video micro-library update mechanism for a registered user on the target platform at a preset time point on the same day, and record the registered user as the designated user; The hotspot capture module of the day is used to capture the hotspot feature information of the target platform on the day; A user preference identification module is used to analyze the video browsing behavior of a specified user within a preset historical period and identify the browsing preference feature information of the specified user; The video push update module is used to determine the updateable video capacity of the specified user's exclusive push video micro-library on the same day, combine the target platform's hot feature information on the same day and the specified user's browsing preference feature information, perform information flow frame extraction analysis on each permitted push video in the target platform's video resource library, select each adapted push video and add it to the specified user's exclusive push video micro-library on the same day to achieve update; The push effect evaluation module is used to randomly extract a preset number of adapted push videos for full-frame information flow analysis, compare them with the corresponding information flow frame extraction analysis results, and evaluate the accuracy of the exclusive push video micro-library update mechanism for the specified user on the target platform on that day in combination with the actual platform video browsing data of the specified user, so as to provide feedback for optimization space; Cloud database, used to store the target platform's various video content theme specification keyword sets; The screening process of each adapted push video includes: recording each video in the target platform video resource library except each video in the micro library of push videos exclusively for a designated user at a preset time point on the day and each video that has been viewed in history as each permitted push video; Perform frame extraction processing on each licensed pushed video to obtain its corresponding video frame image, collect the publisher ID of each licensed pushed video, each keyword and each image element in each video frame image, compare and analyze with the hot spot feature information of the target platform on that day and the browsing preference feature information of the specified user, obtain the hot spot adaptation coefficient and user preference coefficient of each licensed pushed video in the frame extraction state, accumulate to obtain the comprehensive adaptation coefficient of each licensed pushed video, arrange each licensed pushed video in descending order of the comprehensive adaptation coefficient, and select each licensed pushed video whose arrangement order is before the updateable video capacity value as each adapted pushed video; The hotspot adaptation coefficient of each permitted pushed video in the frame extraction state is obtained, including: screening the number of hot keywords on the platform on the day contained in each video frame of each permitted pushed video and the number of hotspot image elements ,in The ID of each licensed video. , is the number of each video frame, , according to the formula Get the hotspot adaptation coefficient of each licensed video push frame extraction state, where is the total number of hot keywords on the platform that day, is the total number of hot image elements on the platform on that day, is the number of video frames; The method of obtaining the user bias coefficient of each permitted pushed video in the frame extraction state includes: comparing the publisher ID of each permitted pushed video with the publisher ID of each biased video within the preset period of the specified user history, and setting the publisher bias weight of each permitted pushed video ; Determine the video content theme type of each licensed biased video based on the keywords in each video frame of each licensed biased video, and obtain the browsing preference ratio of the specified user for each licensed biased video by combining the browsing preference ratio of each video content theme within the historical preset period of the specified user. ; Screen each licensed video push and check the number of image elements that are biased towards the specified user in each video frame , according to the formula Get the user bias coefficient under each permitted video push frame extraction state, The number of image elements to prefer for a given user.

2. According to claim 1, a real-time information flow management and optimization system based on AI intelligent matching is characterized by: The capturing of the hot feature information of the target platform on the day includes: collecting the cumulative playback volume, cumulative like volume and cumulative comment volume of each video in the target platform video resource library as of a preset time point on the day, as well as the corresponding growth rate of the playback volume, like volume and comment volume of each video in the target platform video resource library during a preset period of the day, analyzing the heat coefficient of each video in the target platform video resource library, screening each hot video therein, and extracting keywords from the video title, video tag and video description of each hot video, and identifying image elements from the video cover of each hot video, collecting the keyword set and image element set of each hot video, retrieving each keyword and each image element whose appearance frequency is greater than a preset frequency threshold, and recording them as each common keyword and each common image element; Obtain the search keywords and their continuous search heat received by the search bar of the target platform during the preset time period of the day, select the search keywords whose continuous search heat is greater than the preset continuous search heat threshold, record them as hot search keywords, summarize the hot search keywords and common keywords, remove the duplicate keywords, and use the remaining keywords as the hot keywords of the platform on the day; Based on the target platform search engine library, search for the image elements corresponding to the hot search keywords, aggregate them with the common image elements, remove the duplicate image elements, and use the remaining image elements as the hot image elements of the platform on that day; The hot keywords and hot image elements of the platform on that day are used together as the hot feature information of the target platform on that day.

3. According to claim 1, a real-time information flow management and optimization system based on AI intelligent matching is characterized in that: The identifying of the browsing preference characteristic information of the designated user includes: collecting the video browsing data of the designated user within a preset historical period, including the publisher ID, browsing time ratio, like status index and comment status index of each browsed video, analyzing the interaction coefficient of the designated user for each browsed video within the preset historical period, screening each interactive video, extracting keywords from the video title, video tag and video description of each interactive video, comparing the extracted keywords with the target platform's various video content theme standard keyword sets stored in the cloud database, and analyzing the browsing preference ratio of the designated user for various video content themes within the preset historical period; Record the publisher ID corresponding to each interactive video of the specified user within the historical preset period as the interactive publisher ID, accumulate the interaction coefficients of each interactive video corresponding to the same interactive publisher ID, thereby obtaining the comprehensive interaction coefficient of each interactive publisher ID, and use each interactive publisher ID whose comprehensive interaction coefficient is greater than or equal to the preset comprehensive interaction coefficient threshold as each biased video publisher ID within the historical preset period of the specified user; Extracting each key frame image of each interactive video of a specified user within a preset historical period and performing image element recognition on the key frame images, and retrieving each common image element therein as each biased image element within a preset historical period of the specified user; The browsing preference proportions for various video content topics within a historical preset period, the video publisher IDs of each preference, and the image elements of each preference are collectively used as the browsing preference feature information of a specified user.

4. The real-time information flow management and optimization system based on AI intelligent matching according to claim 3 is characterized by: Determining the updateable video capacity of the dedicated push video micro-library for the specified user on that day includes: querying the free video capacity of the dedicated push video micro-library for the specified user at a preset time point on that day If it is greater than or equal to the preset benchmark, update the video capacity , then As the updateable video capacity of the exclusive push video micro library for the specified user on that day, otherwise As the updateable video capacity of the exclusive push video micro-library for the designated user on that day, the deviation value between the preset benchmark update video capacity and the free video capacity is used as the elimination video capacity, and the video elimination process is performed on the exclusive push video micro-library for the designated user at the preset time point on that day according to the predefined principle.

5. The real-time information flow management and optimization system based on AI intelligent matching according to claim 4 is characterized by: The predefined principles include the subject balance principle, the activity principle and the timeliness principle.

6. The real-time information flow management and optimization system based on AI intelligent matching according to claim 1 is characterized by: The frame extraction processing of each licensed pushed video includes: uniformly setting a fixed number of frame extractions for each licensed pushed video, automatically listing the first and last video frames of each licensed pushed video as required frames, and using the duration of each licensed pushed video and the duration that is divisible by the remaining number of frame extractions as the time interval for uniform frame extraction, thereby realizing frame extraction processing for each licensed pushed video.

7. The real-time information flow management and optimization system based on AI intelligent matching according to claim 1 is characterized by: The evaluation of the accuracy of the exclusive push video micro-library update mechanism for the designated user of the target platform on the day includes: randomly selecting a preset number of adapted push videos as each sampled push video, analyzing the hotspot adaptation coefficient and user bias coefficient of each sampled push video in the full-frame state, and calculating the absolute deviation of the hotspot adaptation coefficient and user bias coefficient in the sampled frame state, thereby analyzing the overall frame state accuracy of the adapted push video. ; Extract the actual platform video browsing data of the specified user on the day, including the browsing time ratio, like status index and comment status index of each adapted push video, filter the interactive push videos, and use the ratio of the number of interactive push videos to the number of adapted push videos as the overall push status accuracy of the adapted push videos. ; Depend on , The accumulated value is used as the accuracy of the video micro-library update mechanism that is exclusively pushed to designated users on the target platform on that day.

Images