We-media streaming pushing method and system based on big data analysis

Abstract

The invention provides a self-media streaming pushing method and system based on big data analysis. According to the method, real-time behavior data (including timestamps, staying durations and interaction types) of a user during cross-device switching are collected, and user scene intention features are generated through a dynamic aggregation module; then, a cross-platform collaborative analysis module is used for calculating content format conversion priorities and push time window decision parameters; dynamically disassembling the content material based on the equipment screen resolution, and generating an adaptive content block conforming to a target platform typesetting rule; when it is detected that a user continuously switches more than three platforms, loading delay data according to each platform to correct a push time window, re-optimizing a transmission queue sequence of content blocks in combination with equipment resolution difference, and finally generating an optimal content combination considering both loading efficiency and display quality; according to the technical scheme provided by the invention, content real-time intelligent adaptation under a user multi-device switching scene is realized, and the cross-platform pushing precision and loading fluency are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet content distribution, and in particular to a self-media stream pushing method and system based on big data analysis. BACKGROUND

[0002] With the development of multi-terminal self-media platforms and the significant increase in user cross-device operation frequency, the demand for users to frequently switch between mobile phones, tablets, PCs and other devices to browse content in the same physical scene is increasingly urgent.

[0003] Existing solutions mostly use collaborative filtering recommendation algorithms based on user historical behavior tags to generate fixed pushing strategies by analyzing user single-device behavior logs. For example, after identifying the user identity through device fingerprints, multiple platform adaptive versions of the same content are pushed within a preset time window; or the content is processed by a static block rule engine, and a pre-stored resolution template is selected according to the device type for format conversion.

[0004] However, the existing solutions have significant defects. First, the dynamic aggregation capability of cross-device real-time behavior data is insufficient, which cannot capture the changes in the scene intent of the user when switching devices, resulting in mismatch between the pushed content and the current operation scene; second, static block rules are difficult to adapt to dynamic differences in layout rules of different platforms, such as layout conflicts between Douyin vertical screen videos and WeChat public number text and image mixed layout, resulting in element occlusion or proportion imbalance after content reorganization; third, in the scenario of continuous switching between multiple platforms, the fixed pushing time window is decoupled from the device screen characteristics, and the high-resolution content loading delay exceeds the user tolerance threshold, causing content flow interruption. SUMMARY

[0005] The present application provides a self-media stream pushing method and system based on big data analysis to solve the problems of insufficient dynamic aggregation capability of cross-device real-time behavior data leading to scene intent mismatch, static block rules failing to adapt to dynamic layout differences of multiple platforms causing content reorganization distortion, and decoupling of fixed pushing time window and device screen characteristics resulting in high-resolution content loading delay exceeding the limit.

[0006] In a first aspect, the present application provides a self-media stream pushing method based on big data analysis, comprising:

[0007] Collecting a set of real-time behavior data of a user on different self-media platforms, the set of real-time behavior data including cross-platform operation timestamps, page dwell time and content interaction types of the user when switching devices in the same physical scene;

[0008] Inputting the set of real-time behavior data into a dynamic aggregation module to generate a preference feature sequence reflecting the current scene intent of the user;

[0009] inputting the preference feature sequence into a cross-platform collaborative analysis module to generate a decision parameter group including a content format conversion priority and a push time window;

[0010] triggering a content reorganization engine according to the push time window in the decision parameter group to dynamically disassemble original content materials based on a screen resolution of a current active device end of the user to generate an adaptive content block set matching a layout rule of the target platform;

[0011] When it is detected that there are more than three continuous switching events of the self-media platforms within a preset time threshold in the real-time behavior data set, the push time window in the decision parameter group is corrected according to the content loading delay data of each self-media platform, and based on the corrected push time window, the transmission queue order of the adaptive content block set is re-allocated according to the screen resolution difference of different device ends to obtain an optimal content block set.

[0012] Optionally, triggering a content reorganization engine according to the push time window in the decision parameter group to dynamically disassemble original content materials based on a screen resolution of a current active device end of the user to generate an adaptive content block set matching a layout rule of the target platform, comprising:

[0013] According to the remaining length of the push time window and the content loading rate limit of the target platform, a dynamic blocking threshold of the screen resolution is calculated, and the dynamic blocking threshold is an inverse proportional function of the screen horizontal pixel value and the remaining length of the push time window;

[0014] Based on the dynamic blocking threshold, a non-uniform grid division is performed on a picture-text mixed area of the original content materials, and a candidate block library including priority markers is generated according to the semantic correlation degree of picture-text elements in the divided grid;

[0015] Performing layout compatibility evaluation on each candidate block in the candidate block library, and generating a layout compatibility matrix including a block size correction coefficient and a position offset amount by matching picture-text spacing, font scaling ratio and multi-column layout rules in the historical layout template of the target platform;

[0016] According to the correction coefficient in the layout compatibility matrix, combined with a screen refresh rate dynamic weight coefficient of the current active device end, cross-column adaptive reorganization is performed on the candidate block library to generate an adaptive content block set matching the layout rule of the target platform.

[0017] Optionally, according to the correction coefficient in the layout compatibility matrix, combined with a screen refresh rate dynamic weight coefficient of the current active device end, cross-column adaptive reorganization is performed on the candidate block library to generate an adaptive content block set matching the layout rule of the target platform, comprising:

[0018] Extracting a patch size correction coefficient and a position offset of each patch from the layout compatibility matrix, fusing the patch size correction coefficient and the dynamic patch threshold value by a linear weighting algorithm to generate a priority sorting value of each patch;

[0019] Multiplying the priority sorting value and a screen refresh rate dynamic weight coefficient to obtain a dynamic scheduling parameter group containing a patch column mapping relationship, a longitudinal compression ratio coefficient, and a loading timing weight;

[0020] According to the priority sorting value in the dynamic scheduling parameter group, arranging the candidate patches in the candidate patch library in descending order, filling the candidate patches into the target platform first screen core column according to the arrangement result, and compressing the distance of the graphic-text mixed area according to the longitudinal compression ratio coefficient to obtain compressed patch vertical compression parameters and column layout data;

[0021] Based on the column layout data, the total vertical pixel value of the patch in the first screen core column of the current device end is calculated in real time, and when it is detected that the total vertical pixel value exceeds the preset range value of the longitudinal pixels of the device end screen, the patch that causes vertical space overrun is marked as a vertical overrun patch. The vertical overrun patch is split into an upper and lower stacked sub-block combination according to the graphic-text semantic correlation degree, and a patch splitting record containing the sub-block size and the identification of the parent block is generated;

[0022] According to the patch splitting record, the position offset in the layout compatibility matrix is updated, and the loading timing weight in the dynamic scheduling parameter group is combined to generate an adaptive content block set.

[0023] Optionally, when it is detected that there are more than three continuous switching events of the self-media platform within a preset time threshold in the real-time behavior data set, the push time window in the decision parameter group is corrected according to the content loading delay data of each self-media platform, including:

[0024] Extracting the self-media platform identifier triggering the self-media platform switching operation and the time interval of the adjacent switching operation in the continuous switching event to generate an original delay data set containing the time interval of the self-media platform switching sequence;

[0025] According to the content loading delay data of each self-media platform in the original delay data set, the delay fluctuation rate and the peak delay value are calculated;

[0026] Based on the time interval of the self-media platform switching sequence and the delay fluctuation rate and the peak delay value, a multi-platform delay feature set containing delay mean offset and variance is generated by a weighted fusion algorithm;

[0027] Inputting the multi-platform delay feature set into a timing compensation model to generate a corrected push time window.

[0028] Optionally, based on the modified push time window, re-allocate the transmission queue order of the set of adaptive content blocks according to the screen resolution difference of different device ends, to obtain an optimal set of content blocks, comprising:

[0029] According to the time margin of the modified push time window, in combination with the ratio between the current device end screen resolution and the preset reference resolution, generate a resolution adaptation parameter;

[0030] Weighted fusion of the time margin and the resolution adaptation parameter generates the transmission priority weight of each block, and the set of adaptive content blocks is arranged in descending order based on the transmission priority weight of each block to generate an initial transmission queue order;

[0031] Based on the product of the current device end screen resolution and the modified time margin, dynamically determine the capacity range of the first screen core queue, select blocks with weight greater than a preset threshold from the initial transmission queue order as the first screen core queue, and compress blocks with resolution lower than the device end screen resolution to generate a compression instruction set containing a compression algorithm identifier;

[0032] Real-time monitoring of the cumulative loading duration of the first screen core queue, when detecting that the cumulative loading duration exceeds the preset proportion threshold of the modified push time window, the block causing the overrun is marked as a timing overrun block, and the timing overrun block is split into multiple sub-block sequences according to the compression instruction set and the resolution of the timing overrun block, to obtain a split record containing a sub-block loading timing mark;

[0033] According to the split record, insert the sub-blocks in the multiple sub-block sequences into the non-overrun position of the initial transmission queue order, and re-calculate the transmission priority weight of all blocks in the initial transmission queue order, based on the re-calculated transmission priority weight, to generate an optimal set of content blocks.

[0034] Optionally, input the set of real-time behavior data into a dynamic aggregation module to generate a preference feature sequence reflecting the user's current scene intention, comprising:

[0035] Time axis alignment and outlier filtering of the cross-platform operation time stamp in the set of real-time behavior data to generate a cross-platform behavior log containing user cross-platform switching behavior;

[0036] Based on the page dwell time and the interaction behavior type between the user and the content, a dynamic weight is given to the user behavior in the set of real-time behavior data through a time decay function to generate a weighted behavior data set containing a behavior time effectiveness mark;

[0037] extract the device switching frequency and interaction type distribution characteristics of the user in the same physical scene in the cross-platform behavior log, and construct a multi-dimensional behavior association matrix in combination with the dynamic weight in the weighted behavior data set;

[0038] input the multi-dimensional behavior association matrix into a sequence modeling network based on a multi-head attention mechanism, generate a scene intention vector by capturing the spatio-temporal dependence relationship of cross-platform behavior;

[0039] perform feature decoupling and dimensionality reduction processing on the scene intention vector, and output a preference feature sequence reflecting the current scene intention of the user.

[0040] Optionally, input the preference feature sequence into a cross-platform collaborative analysis module to generate a decision parameter group containing content format conversion priority and push time window, including:

[0041] perform multi-platform relevance analysis on the preference feature sequence, extract the association weight between the content format constraints of each media platform and the user behavior preferences through a multi-head attention mechanism, and generate a platform feature association matrix;

[0042] based on the platform feature association matrix, in combination with the layout rule library of the target platform and the historical content distribution efficiency data, calculate the conversion priority of different content formats through a dynamic programming algorithm, and generate a priority parameter set containing format compatibility score and conversion path;

[0043] According to the user behavior timeliness label and cross-platform switching frequency in the preference feature sequence, generate an initial push time window through a time series prediction model, and combine the content loading delay data of each media platform to perform time margin compensation correction to generate a modified push time window;

[0044] fuse the priority parameter set and the modified push time window in multiple dimensions to output a decision parameter group containing content format conversion priority, push time window and dynamic compensation threshold.

[0045] In a second aspect, the present application provides a self-media streaming push system based on big data analysis, comprising:

[0046] a collection module that gracefully collects a real-time behavior data set of a user on different self-media platforms, the real-time behavior data set containing cross-platform operation timestamps, page dwell time and content interaction types when the user switches devices in the same physical scene;

[0047] a first input module for inputting the real-time behavior data set into a dynamic aggregation module to generate a preference feature sequence reflecting the current scene intention of the user;

[0048] The second input module is configured to input the preference feature sequence into the cross-platform collaborative analysis module to generate a decision parameter group including a content format conversion priority and a push time window.

[0049] The disassembly module is configured to trigger the content reorganization engine according to the push time window in the decision parameter group, disassemble the original content material based on a screen resolution of a current active device of the user, and generate an adaptive content block set matching the layout rules of the target platform.

[0050] The correction module is configured to correct the push time window in the decision parameter group according to content loading delay data of each self-media platform when detecting that there are more than three continuous switching events of the self-media platforms within a preset time threshold in the real-time behavior data set, and reassign a transmission queue order of the adaptive content block set according to a screen resolution difference of different device ends based on the corrected push time window, to obtain an optimal content block set.

[0051] In a third aspect, an embodiment of the present application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the self-media streaming push method based on big data analysis as described in the first aspect.

[0052] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program, and the computer program is executed by a computer to implement the self-media streaming push method based on big data analysis as described in the first aspect.

[0053] The embodiments of the present application can accurately capture the user intention change in the multi-device switching scenario by collecting the user cross-platform behavior data in real time and constructing the dynamic preference feature sequence, and realize scenario-based content adaptation. Based on the decision parameter group generated by the cross-platform collaborative analysis, the content material is disassembled dynamically combined with the screen resolution of the device end, the problem that the traditional static block rule does not match the multi-platform layout rule is solved, and the content distortion is effectively avoided. Especially in the multi-platform high-frequency switching scenario, the push time window is corrected in real time and the transmission queue is optimized through the loading delay data, the dynamic balance of content loading efficiency and display quality is realized, the probability of streaming interruption caused by device difference is significantly reduced, and the coherence experience of the user cross-platform content consumption is improved.

[0054] Further, by pushing time window and dynamic association calculation of screen resolution, intelligent optimization of image-text content block threshold is realized, ensuring the maximum efficiency of the first screen core content display within the limited loading time. Non-uniform grid division combined with semantic correlation analysis can accurately identify the core elements of the image-text mixed area, forming a priority block strategy. Based on the compatibility evaluation and dynamic weight fusion of historical layout templates, the content reorganization meets the multi-column layout rules and device refresh rate characteristics, effectively solving the problems of image-text spacing disorder and font scaling abnormalities in traditional methods. The cross-column adaptive reorganization mechanism ensures the visual integrity of complex pages through vertical compression optimization and sub-block splitting strategy, realizing seamless content adaptation between devices with different resolutions.

[0055] These aspects or other aspects of the present application will be more apparent in the following description of the embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor under the premise of the drawings.

[0057] Figure 1 A flow chart of a self-media streaming push method based on big data analysis provided by the present application is shown;

[0058] Figure 2 A structural schematic diagram of a self-media streaming push system based on big data analysis provided by the present application is shown;

[0059] Figure 3 A structural schematic diagram of a computing device provided by the present application is shown. DETAILED DESCRIPTION

[0060] In order to make the person skilled in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0061] In some of the processes described in this specification and in the claims and in the accompanying drawings, various operations are described in a particular, sequential order. However, it should be understood that, in some implementations, the order of operations can be changed, or omitted, or that some operations can be performed in parallel, or with precision not required by the specification. Moreover, some operations can be performed by different entities in different implementations. The order in which operations are described is not intended to be a limitation unless specifically stated.

[0062] The researchers found that the existing self-media push method has the defects of scene intention recognition lag and content adaptation static in the multi-device cross-platform operation of users. The traditional technology cannot capture the behavior association when the user switches the device in real time, resulting in the disconnection of the pushed content and the current scene. At the same time, the fixed block rule conflicts with the dynamic layout of multiple platforms, causing content loading delay or display distortion. Based on this, a self-media streaming push method based on big data analysis is provided. The method can generate scene intention features through cross-platform behavior data dynamic aggregation, realize content intelligent disassembly and reorganization combined with device resolution and time window constraints, and dynamically correct the transmission strategy in the multi-platform high-frequency switching scene, effectively improving the accuracy and fluency of cross-device content push.

[0063] The technical solution of the present application can be applied to the scene where the user frequently switches mobile phones, tablets, PCs and other multi-terminal devices in the same physical scene to browse different self-media platforms (such as WeChat, Douyin, and some books, etc.), and needs to adapt to the layout rules of each platform in real time and optimize the content loading efficiency. In the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0064] Figure 1 A flowchart of a self-media streaming push method based on big data analysis is provided for the embodiments of the present application, as shown in Figure 1 The method comprises:

[0065] Step 101, collect a set of real-time behavior data of the user on different self-media platforms, wherein the set of real-time behavior data includes the cross-platform operation timestamp, page stay duration and content interaction type of the user when switching devices in the same physical scene;

[0066] In this step, cross-platform operation timestamp refers to recording the precise time information of the user when switching between different devices to access the media platform, including device identifier, platform type and switching action triggering time; page dwell time refers to the duration of the user from entering a certain platform page to leaving, which is dynamically calculated through device-side page visibility monitoring technology; content interaction type refers to the specific interaction behavior classification data of the user with the content, such as likes, collections, comments, swipes, etc., which is captured and associated with the content identifier through front-end interaction embedding technology.

[0067] In the embodiments of the present application, first, through device sensors (such as gyroscopes, touch event detection modules) and cross-platform identity binding technology, the device switching behavior of the user in the same physical scene is tracked in real time, and the original operation log containing the device identifier, platform type and timestamp is generated. Secondly, the time nodes of the user entering and exiting the page are recorded by using the page life cycle monitoring module (such as browser or application programming interface), the dwell time is calculated by combining the time difference value algorithm, and the invalid records caused by network interruption or device hibernation are eliminated by the abnormal data filtering mechanism (such as outlier detection). Then, the front-end embedding technology is used to capture the interaction behavior of the user in the page, and the touch operation is mapped to the standardized interaction type (such as clicking a specific area to mark as “like”) according to the preset rules. Finally, the multi-source data is aggregated according to the user identifier and time window by using the distributed stream processing system, forming a real-time behavior data set containing cross-platform operation timestamp, dwell time and interaction type.

[0068] For example, user A uses a mobile phone to browse a short video platform in a coffee shop, then switches to a tablet computer to read a social media article, and finally views a note on a content sharing platform on a notebook computer. The system detects through the device sensor that the mobile phone is placed on the table and the tablet is picked up, records the cross-platform switching timestamp, and triggers the visibility monitoring of the social media page to calculate the dwell time in the page. When user A performs a like operation on an article, the front-end embedding technology captures the behavior and associates the content identifier. After all the data is aggregated by the stream processing system, a real-time behavior data set containing the device switching sequence (mobile phone→tablet→notebook), the dwell time of each platform and the interaction event (such as liking an article) is formed. This data will be used in subsequent modules to analyze the user's current scene intention, for example, the dynamic aggregation module infers that the user may be in a “fragmented learning” scene, and preferentially pushes lightweight content that adapts to multi-device switching.

[0069] Step 102, inputting the real-time behavior data set into a dynamic aggregation module to generate a preference feature sequence reflecting the user's current scene intention;

[0070] In this step, the dynamic aggregation module is used to fuse the spatio-temporal correlation of cross-platform behavior data and extract the model components of user scene intention; the preference feature sequence represents the multi-dimensional vector sequence of user content preference in the current scene, including interest labels, interaction intensity and spatio-temporal distribution characteristics.

[0071] In the embodiments of the present application, first, the cross-platform operation timestamps in the real-time behavior data set are time axis aligned (such as based on a device clock synchronization algorithm) to generate cross-platform behavior logs. Second, a time decay function (such as an exponential decay model) is used to give dynamic weights to page stay duration and interaction behavior, and the recent behavior has a higher weight, forming a weighted behavior data set. Then, the device switching frequency (such as the number of switching times per unit time) and the interaction type distribution characteristics (such as the proportion of likes / comments) are extracted, and a multi-dimensional behavior correlation matrix is constructed in combination with the dynamic weights. Then, the correlation matrix is input into a sequence modeling network based on a multi-head attention mechanism to capture the spatio-temporal dependence relationship of cross-platform behavior (such as the correlation between device switching and content interaction), and output a scene intention vector. Finally, the preference feature sequence reflecting the user's current scene intention is generated through feature decoupling (such as independent component analysis) and dimensionality reduction processing (such as the t-SNE algorithm).

[0072] For example, based on the real-time behavior data of user A (mobile phone to tablet switching, medium and long stay, article liking), the dynamic aggregation module first aligns the time axis and finds that the stay duration on the tablet is significantly higher than that on the mobile phone. Through the time decay model, the recent behavior (liking on the tablet) is given a higher weight, and the behavior correlation matrix is constructed in combination with the device switching frequency (low frequency switching). The multi-head attention network identifies the “cross-device deep reading” intention, and the preference feature sequence is generated after feature decoupling, including “long text reading preference” and “cross-device continuity”. This feature sequence will be input into the cross-platform collaborative analysis module to generate a push strategy that adapts to the multi-column layout of the tablet.

[0073] In step 103, the preference feature sequence is input into the cross-platform collaborative analysis module to generate a decision parameter group including content format conversion priority and push time window;

[0074] In this step, the preference feature sequence is a multi-dimensional vector sequence output by the dynamic aggregation module, representing the interest labels (such as “video preference” and “long text reading”) in the current scene of the user, the interaction intensity (such as high-frequency likes and low-frequency swiping), and the spatio-temporal distribution characteristics (such as device switching continuity); the cross-platform collaborative analysis module is used to fuse multi-platform constraints and user preferences to generate intelligent decision components of content distribution strategies; the content format conversion priority refers to the scoring and sorting of different content formats (such as picture-text mixed layout, vertical screen video, and multi-column long text) adapting to the layout rules of the target platform; and the push time window refers to the best time range for content push, which is dynamically calculated in combination with the time effectiveness of user behavior and platform loading delay.

[0075] In the embodiments of the present application, firstly, the preference feature sequence is analyzed for multi-platform relevance: the association weight of the content format constraints of each platform (such as the text-image spacing rules of WeChat public number, the vertical screen proportion limit of Douyin) and the user preference features (such as “long text reading”) is extracted through a multi-head attention mechanism, and a platform feature association matrix (such as the weight value of WeChat platform and long text reading is higher) is generated. Secondly, based on the historical layout template library of the target platform (storing text-image layout, video size, etc.) and the historical distribution efficiency data (such as click rate, loading completion rate), the conversion path and compatibility score of different content formats (such as splitting long text into multi-column layout scores are better than direct scaling) are calculated through a dynamic programming algorithm. At the same time, according to the behavior timeliness label (such as the user frequently switches devices recently) and the cross-platform switching frequency in the preference feature sequence, an initial push time window is generated using a time series prediction model (such as an LSTM network), and then the time margin compensation correction is performed by superimposing the content loading delay data (such as the average loading time consumption of WeChat articles) to finally output the decision parameter group containing the format conversion priority (such as “multi-column long text > vertical screen video”), the push time window (such as within the next three minutes), and the dynamic compensation threshold (such as allowing a floating range of loading delay).

[0076] For example, based on the preference feature sequence of user A (containing “cross-device deep reading” and “long text preference” labels), the cross-platform collaborative analysis module discovers through the attention mechanism that its current active device is a tablet computer, and the layout rules of the target platform WeChat support multi-column layout. The dynamic programming algorithm calculates that the format compatibility score of splitting the original long text into double-column text-image layout is the highest, and the priority is set to first. At the same time, the time series prediction model combines the characteristics that user A stays on the tablet computer for a stable time (about switching once every minute) and the average loading delay of WeChat platform is low, and generates a push time window of “within 30 seconds after device switching”. Finally, the decision parameter group is output: prefer to convert long text into double-column format, the push time window is 30 seconds after switching, and the loading delay is allowed to float by 10%. This parameter group will trigger the content reorganization engine to split the content according to the tablet resolution to ensure that the adaptation and push are completed within the time window.

[0077] Step 104, triggering the content reorganization engine according to the push time window in the decision parameter group, dynamically splitting the original content material based on the screen resolution of the user's current active device terminal to generate an adapted content block set matching the layout rules of the target platform;

[0078] In this step, the content reorganization engine is a module for dynamically disassembling content according to device resolution and typesetting rules, the core parameters of which include a screen resolution dynamic blocking threshold (a value of an inverse proportional function of screen horizontal pixels and a remaining time window length), a typesetting compatibility matrix (a data structure for storing block size correction coefficients and position offsets), and an adaptive content block set, which is a content unit set after dynamic blocking, priority sorting and typesetting optimization, and meets the target platform typesetting rules and adapts to the device resolution.

[0079] In the embodiments of the present application, first, the dynamic blocking threshold of the screen resolution (such as the horizontal pixel value / remaining time length) is determined by calculating the inverse proportional function based on the remaining time window length and the target platform content loading rate limit, to determine the content disassembly granularity. Second, the non-uniform grid division is performed on the graphic-text mixed area of the original material, the semantic correlation algorithm (such as OCR text extraction and image area clustering) is used to identify the graphic-text logical relationship, and a candidate block library with priority labels is generated. Third, the candidate blocks are matched with the historical typesetting template library of the target platform, the rule engine is used to evaluate the graphic-text spacing, font scaling ratio and multi-column layout compatibility, and a typesetting compatibility matrix containing block size correction coefficients (such as width scaling ratio) and position offsets (such as column left alignment offset value) is generated. Finally, the adaptive content block set is generated by using the linear weighting algorithm to fuse the correction coefficients and the block threshold, in combination with the dynamic weight of the device screen refresh rate (such as high refresh rate device priority loading dynamic content), for cross-column adaptive reorganization (such as block column mapping and vertical compression) of the candidate blocks.

[0080] For example, the decision parameter group of user A indicates that the double-column graphic-text content needs to be pushed to the tablet end WeChat platform within 30 seconds. The content reorganization engine first calculates the inverse proportional threshold of the current screen resolution (such as horizontal pixels) and the remaining time window, to determine the maximum block width of each column. The original long text is divided into non-uniform grids, the semantic correlation areas of the title, text paragraphs and pictures are identified, and priority-sorted candidate blocks (such as the title with the highest priority) are generated. The typesetting compatibility evaluation finds that the picture needs to be scaled to 80% of the column width and left aligned, and the line spacing of the text paragraphs needs to be compressed by 10%. In combination with the high refresh rate characteristics of the tablet, the title and the first paragraph of the text are preferentially loaded into the double-column core position of the first screen, and the picture is compressed vertically to fill the right column. When it is detected that the total vertical pixels of the first screen exceed the limit, the exceeding paragraphs are semantically split into upper and lower sub-blocks, the typesetting matrix offset is updated, and finally the content block set that adapts to the WeChat double-column rules is generated. This set will be dynamically adjusted in the subsequent steps according to the device switching to transmit the queue, for example, when user A switches to the mobile phone end, the double-column content is automatically converted into a single-column flow layout.

[0081] Step 105, when detecting that there are at least three continuous switching events of the self-media platforms in the real-time behavior data set within the preset time threshold, correcting the push time window in the decision parameter set according to the content loading delay data of each self-media platform, and based on the corrected push time window, re-distributing the transmission queue order of the adaptive content block set according to the screen resolution difference of different device ends to obtain the optimal content block set;

[0082] In this step, the continuous switching event refers to the operation behavior of the user continuously switching at least three self-media platforms within a preset time threshold (such as two minutes); the content loading delay data refers to the average time consumption from request to completion of content rendering of each platform, including network transmission, resource analysis and layout calculation time consumption; the transmission queue order refers to the loading priority order of each block in the adaptive content block set, which is dynamically adjusted based on resolution difference and time margin.

[0083] In the embodiments of the present application, firstly, whether there are three or more continuous switching events of the platforms within the preset time threshold in the real-time behavior data set is detected. If so, the platform identifier and adjacent switching time interval of the switching sequence are extracted to generate an original delay data set. The loading delay fluctuation rate (standard deviation / mean) and peak delay value of each platform are calculated, and a multi-platform delay feature set (including mean offset and variance) is generated by a weighted fusion algorithm combined with the switching time interval. The delay feature set is input into a time compensation model (such as a delay prediction network based on LSTM), and a delay compensation coefficient and a corrected time margin are output to update the push time window. Subsequently, the adaptive parameters are generated according to the ratio of the screen resolution of the device end to the reference resolution, and are weighted and fused with the corrected time margin to calculate the block transmission priority weight. The product of the resolution and the time margin is used to dynamically determine the first-screen core queue capacity, and high-weight blocks are preferentially loaded. Compression or sub-block splitting is performed on low-resolution blocks. The first-screen loading time is monitored in real time, and if it exceeds the limit, the time-overrun block is split into a sub-block sequence, inserted into a non-overrun position and the weight is recalculated, and finally the optimal content block set is generated.

[0084] For example, user A continuously switches WeChat, Douyin and Xiaohongshu within two minutes in a coffee shop, triggering step 105. The system extracts the switching sequence and interval, calculates the loading delay fluctuation rate of each platform (the peak delay of Xiaohongshu is the highest), and uses a time compensation model to predict that the time margin needs to be increased. The original push time window is extended from 30 seconds to 40 seconds. According to the screen resolution of the mobile phone (which is lower than the tablet reference value), adaptive parameters are generated, and the block transmission priority is recalculated: the title and the first paragraph have the highest weight, and the image resolution is compressed. The first-screen core queue is dynamically determined to only accommodate the title and the first paragraph, and it is found that the first paragraph is overrun during real-time loading. It is split into two sub-blocks, inserted at the end of the queue and reordered to generate the optimal content block set for the adaptive mobile phone screen, ensuring that the user switches to the mobile phone to preferentially render the core content.

[0085] In summary, the application embodiment realizes intelligent content disassembly and layout adaptation driven by device resolution by collecting cross-device behavior data in real time and dynamically aggregating user scenario intentions, and generating content conversion and pushing strategies in combination with multi-platform collaborative decision-making. Further, in a multi-platform high-frequency switching scenario, through loading delay compensation and resolution difference optimization, the content transmission queue is dynamically adjusted to form a closed-loop cross-device streaming system. This method breaks through the defects of traditional pushing strategies such as static blocking and fixed time window, so that the content stream maintains semantic coherence and visual integrity when switching between multiple devices, significantly reduces the risk of loading interruption in cross-platform, and improves the content consumption fluency and core information reach efficiency of users in a multi-device scenario.

[0086] To solve the problem that the traditional content blocking method cannot dynamically adapt to device loading time and layout rules, in some embodiments, according to step 104, the content reorganization engine is triggered according to the push time window in the decision parameter group, and the original content material is dynamically disassembled based on the screen resolution of the user's current active device end, to generate an adaptive content block set matching the layout rules of the target platform, including:

[0087] Step 201, according to the remaining time length of the push time window and the content loading rate limit of the target platform, calculate the dynamic blocking threshold of the screen resolution, the dynamic blocking threshold is the inverse proportional function of the screen horizontal pixel value and the remaining time length of the push time window;

[0088] In this step, the dynamic blocking threshold refers to the dynamic constraint value based on the remaining time length of the push time window and the content loading rate limit of the target platform, which is used to control the upper limit of the horizontal size of content disassembly, to ensure that content blocking is completed within the remaining time.

[0089] In the application embodiment, first, the content loading rate limit (such as the amount of pixels that can be loaded per second on the platform) of the target platform and the remaining time length of the push time window are obtained. Through the dynamic blocking threshold calculation model, the screen horizontal pixel value is inversely proportional to the remaining time length and the loading rate to generate a dynamic constraint value of the blocking width. For example, the higher the screen horizontal pixel value and the shorter the remaining time length, the smaller the blocking width threshold, to ensure that the content disassembly can be completed within a limited time.

[0090] Step 202, based on the dynamic blocking threshold, non-uniform grid division is performed on the graphic-text mixed area of the original content material, and according to the semantic correlation degree of the graphic-text elements in the divided grid, a candidate block library containing priority labels is generated;

[0091] In this step, the non-uniform grid division refers to non-equal-width segmentation of the graphic-text mixed area according to a dynamic block threshold, and semantic association graphic-text units are retained; the semantic association degree is used to identify the association strength of graphic-text elements (such as the corresponding relationship between the picture and the text below) through visual layout analysis and semantic logical reasoning.

[0092] In the embodiments of the present application, the original content is divided into non-uniform grids based on a dynamic block threshold: the edge detection algorithm is used to identify the boundary of the graphic-text mixed area, and the content is divided into blocks of different widths in combination with the threshold. Subsequently, the text recognition and image clustering technology is used to analyze the semantic association of graphic-text elements, such as the strong association between the title and the first picture, and the logical continuity between text paragraphs. The blocks are marked with priorities according to the association degree, and a candidate block library is generated.

[0093] In step 203, the typesetting compatibility evaluation is performed on each candidate block in the candidate block library, and the typesetting compatibility matrix containing the block size correction coefficient and the position offset is generated by matching the graphic-text spacing, font scaling ratio and multi-column layout rules in the historical typesetting template of the target platform.

[0094] In this step, the typesetting compatibility matrix is a matrix used to record the adaptation parameters of the candidate block and the typesetting rules of the target platform, including the block size correction coefficient (such as the scaling ratio) and the position offset (such as the alignment parameter).

[0095] In the embodiments of the present application, the candidate block is matched with the historical typesetting template library of the target platform, and the rules engine is used to compare the graphic-text spacing, font scaling ratio and multi-column layout rules. For example, if the width of a block exceeds the maximum limit of the single column of the platform, the width scaling coefficient is generated to adapt to the column width, and the offset of left alignment or center alignment is calculated. Finally, the matrix is output to record the correction parameters of each block.

[0096] In step 204, the candidate block library is cross-column adaptively reorganized according to the correction coefficient in the typesetting compatibility matrix and in combination with the screen refresh rate dynamic weight coefficient of the current active device terminal, and a set of adaptation content blocks matching the typesetting rules of the target platform is generated.

[0097] In this step, the cross-column adaptive reorganization refers to dynamically adjusting the block layout in combination with the correction parameters of the typesetting compatibility matrix and the screen characteristics of the device, so as to ensure that the content adapts to the rules of the target platform.

[0098] In the embodiments of the present application, first, the correction coefficient and the offset in the layout compatibility matrix are extracted, and the dynamic weight of the device screen refresh rate (such as high refresh rate device preferentially loading dynamic content) is combined to generate a block priority sorting value. The blocks are filled into the target platform first screen core field according to the priority, and the line spacing is adjusted according to the vertical compression ratio. If the total height of the blocks exceeds the vertical range of the screen, the blocks are split into upper and lower sub-blocks according to the semantic correlation degree, and the final adaptive content block set is generated after updating the offset.

[0099] In order to solve the layout overflow problem caused by the vertical space overflow after the block filling, in some embodiments, according to step 204, the candidate block library is adaptively reorganized across the column according to the correction coefficient in the layout compatibility matrix, combined with the dynamic weight coefficient of the screen refresh rate of the current active device end, to generate an adaptive content block set matching the layout rules of the target platform, including:

[0100] Step 301: Extract the block size correction coefficient and the position offset of each block from the layout compatibility matrix, fuse the block size correction coefficient and the dynamic block threshold value according to the linear weighting algorithm to generate the priority sorting value of each block;

[0101] In this step, the linear weighting algorithm is a calculation method for fusing the block size correction coefficient and the dynamic block threshold value according to the preset weight ratio, which is used to quantify the loading priority of the block; the priority sorting value represents the comprehensive score of the loading priority of the block in the reorganization process, and the higher the value is, the higher the priority is.

[0102] In the embodiments of the present application, first, the size correction coefficient (such as width scaling ratio) and the position offset (such as left alignment offset value) of each block are extracted from the layout compatibility matrix. Then, the block size correction coefficient and the dynamic block threshold value are fused by the linear weighting algorithm, for example, the correction coefficient weight is set to be a preset proportion (such as sixty-four allocation) of the dynamic block threshold weight. After fusion, the priority sorting value is generated, which is used for global sorting of the subsequent block loading order.

[0103] Step 302: Multiply the priority sorting value and the screen refresh rate dynamic weight coefficient to obtain a dynamic scheduling parameter group containing block field mapping relationship, vertical compression ratio coefficient and loading timing weight;

[0104] In this step, the screen refresh rate dynamic weight coefficient represents a dynamic adjustment factor set according to the device screen refresh rate (such as 60Hz / 120Hz), and a higher refresh rate device is given a higher weight to optimize the smoothness of dynamic content loading; the block column mapping relationship represents the position allocation rule of the block in the target platform layout (such as the top of the left column or the middle of the right column); the vertical compression ratio coefficient represents the row spacing or height compression ratio of the block in the vertical direction, which is used to adapt to the vertical space of the screen; and the loading timing weight represents the loading order priority of the block in the transmission queue, and the higher the value, the earlier the loading.

[0105] In the embodiment of the application, first, the priority sorting value generated in step 301 is multiplied by the screen refresh rate dynamic weight coefficient (such as high refresh rate device weight is high and low refresh rate device weight is low) to obtain the comprehensive priority score of the block. Then, the block column mapping relationship is determined according to the score (such as high blocks are filled into the first screen core column), and the vertical compression ratio coefficient is generated in combination with the multi-column layout rule of the target platform (such as text paragraph row spacing compression). At the same time, the loading timing weight is allocated based on the score to ensure that high-priority blocks are loaded first. Finally, the dynamic scheduling parameter group containing the column mapping, compression ratio and loading timing is output.

[0106] Step 303: According to the priority sorting value in the dynamic scheduling parameter group, the candidate blocks in the candidate block library are arranged in descending order, and the candidate blocks are filled into the target platform first screen core column according to the arrangement result, and the spacing compression is performed on the graphic-text mixed area according to the vertical compression ratio coefficient, to obtain the vertical compression parameters of the compressed blocks and the column layout data;

[0107] In this step, the block vertical compression parameter refers to the row spacing or height ratio parameter of the graphic-text mixed area after vertical compression, which is used to adapt to the vertical space limitation of the screen. The column layout data refers to the structured information recording the position, size and compressed state of the block in the target platform first screen core column, including block coordinates, size and compression ratio.

[0108] In the embodiment of the application, first, the candidate blocks are arranged in descending order according to the priority sorting value in the dynamic scheduling parameter group to generate a block loading order list. The blocks are filled into the target platform first screen core column (such as the top of the left column or the middle of the right column) in order, and the row spacing or height of the graphic-text mixed area is adjusted according to the vertical compression ratio coefficient. If it is detected during the filling process that the total vertical pixels of the block exceed the vertical range of the screen, the block splitting mechanism is triggered, the over-limit block is split into upper and lower sub-blocks according to the semantic correlation degree, the column layout data is updated and the compression parameters are recorded.

[0109] At step 304, based on the column layout data, the total vertical pixel value of the sub-blocks in the first screen core column of the current device is calculated in real time. When it is detected that the total vertical pixel value exceeds the preset range of the vertical pixel of the device screen, the sub-blocks that cause the vertical space to exceed the limit are marked as vertical over-limit sub-blocks. The vertical over-limit sub-blocks are split into an upper and lower stacked sub-block combination according to the graphic-text semantic correlation degree, and a sub-block splitting record containing the size of the sub-block and the identification of the parent block is generated.

[0110] In this step, the vertical over-limit sub-block refers to a sub-block whose total vertical pixel value exceeds the preset range of the vertical pixel of the device screen after being filled into the first screen core column. The sub-block splitting record is structured data recording that the vertical over-limit sub-block is split into sub-blocks, containing the size of the sub-block, the identification of the parent block, and the stacking order between the sub-blocks.

[0111] In the embodiments of the present application, first, based on the column layout data, the total vertical pixel value of all sub-blocks in the first screen core column of the current device is calculated in real time. When it is detected that the total vertical pixel value exceeds the preset range of the vertical pixel of the device screen, the sub-blocks that cause the vertical space to exceed the limit are located and marked as vertical over-limit sub-blocks. Subsequently, semantic correlation degree analysis is performed on the vertical over-limit sub-blocks, for example, logical segmentation points (such as the first sentence of a paragraph, the position of a transition word) in a paragraph are identified through natural language processing technology, and the sub-blocks are split into an upper and lower stacked sub-block combination according to the semantics. After splitting, a sub-block splitting record is generated, recording the size of the sub-block, the identification of the parent block, and the stacking order, and the position and size information of the sub-blocks in the column layout data is updated.

[0112] At step 305, the position offset in the typesetting compatibility matrix is updated according to the sub-block splitting record, and the adaptive content block set is generated by combining the loading timing weight in the dynamic scheduling parameter group.

[0113] In the embodiments of the present application, first, the position offset of the corresponding parent block in the typesetting compatibility matrix is updated according to the size of the sub-block and the identification of the parent block in the sub-block splitting record. For example, if the parent block is split into two upper and lower sub-blocks, the upper sub-block retains the original offset of the parent block, and the offset of the lower sub-block is adjusted to the original offset of the parent block plus the height of the upper sub-block. Subsequently, the global loading order of the sub-blocks and the non-split sub-blocks is sorted by combining the loading timing weight (such as the upper sub-block needs to be loaded first due to high priority) in the dynamic scheduling parameter group. Finally, the adaptive content block set is generated, containing the position, size, compression parameter, and loading timing label of all sub-blocks.

[0114] In order to improve the accuracy of time window prediction in the multi-platform continuous switching scenario, in some embodiments, according to step 105, when it is detected that there are at least three continuous switching events of the self-media platforms within the preset time threshold in the real-time behavior data set, the push time window in the decision parameter group is corrected according to the content loading delay data of each self-media platform, including:

[0115] Step 401, extract the self-media platform identifier triggering the self-media platform switching operation and the time interval of the adjacent switching operation in the continuous switching event, and generate an original delay dataset containing the time interval of the self-media platform switching sequence;

[0116] In this step, the self-media platform identifier is a symbol or code uniquely identifying the self-media platform switched by the user, such as a platform name abbreviation or a hash value, used to distinguish different platforms. The time interval is the time difference between two consecutive platform switching operations, reflecting the speed of the user switching platforms. The original delay dataset is a structured dataset containing the platform switching sequence and its corresponding time interval, used for subsequent analysis of the correlation between platform switching behavior and loading delay.

[0117] In the embodiments of the present application, first, the continuous switching events (such as operation records of switching at least three platforms within a preset time threshold) are extracted from the real-time behavior data set, and the self-media platform identifiers triggering the switching (such as platform A, platform B, platform C) are obtained. Subsequently, the time interval of adjacent platform switching operations is calculated by the time stamp difference (such as the interval of platform A switching to platform B is T1, and the interval of platform B switching to platform C is T2). Finally, the platform identifiers are arranged in switching order, and the corresponding time intervals are associated to generate the original delay dataset.

[0118] Step 402, according to the content loading delay data of each self-media platform in the original delay dataset, calculate the delay fluctuation rate and the peak delay value;

[0119] In this step, the delay fluctuation rate is an index measuring the dispersion degree of the content loading delay data of each platform, reflecting the stability of the platform loading performance, which is calculated by statistical methods (such as the ratio of standard deviation to mean). The peak delay value is the maximum loading delay time of each platform in the original delay dataset, which is used to quantify the loading performance bottleneck in extreme cases.

[0120] In the embodiments of the present application, first, the loading delay data of each self-media platform is extracted from the original delay dataset (such as the delay sequence of platform A is L1, L2, L3). Subsequently, the ratio of the standard deviation to the mean of the delay data of each platform is calculated to obtain the delay fluctuation rate (such as the fluctuation rate of platform A is the ratio of the standard deviation to the mean). At the same time, the delay data of each platform is traversed, and the maximum value is extracted as the peak delay value. For example, the peak delay value of platform A is the maximum value in its delay sequence.

[0121] Step 403, based on the time interval of the self-media platform switching sequence and the delay fluctuation rate and the peak delay value, generate a multi-platform delay feature set containing delay mean offset and variance through a weighted fusion algorithm;

[0122] In this step, the weighted fusion algorithm is a calculation model for fusing the self-media platform switching time interval, delay fluctuation rate and peak delay value according to a preset weight ratio, for quantifying the delay characteristics in the multi-platform collaborative scenario; the delay mean offset is the difference between the actual loading delay of each platform and the benchmark delay mean, reflecting the deviation of the platform loading performance relative to the benchmark state; and the multi-platform delay feature set is a structured data set integrating the delay mean offset, variance and switching sequence, for describing the comprehensive delay characteristics in the multi-platform switching scenario.

[0123] In the embodiments of the present application, first, the time interval of the self-media platform switching sequence is standardized (for example, normalized to the interval of 0-1), the delay fluctuation rate and the peak delay value are linearly weighted according to the preset weight (for example, the fluctuation rate weight and the peak weight), and the delay comprehensive score of each platform is generated. Subsequently, the offset of the loading delay data of each platform relative to the benchmark mean (for example, the historical average delay) is calculated, and the variance is calculated in combination with the delay fluctuation rate. Finally, the standardized switching time interval, the delay mean offset and the variance are integrated according to the multi-dimensional feature fusion rule to generate the multi-platform delay feature set.

[0124] Step 404, inputting the multi-platform delay feature set into a time sequence compensation model to generate a corrected push time window;

[0125] In this step, the time sequence compensation model is a time sequence prediction model trained based on historical delay data and real-time features, for dynamically adjusting the push time window to adapt to the delay fluctuation in the multi-platform switching scenario. The corrected push time window is the optimized time range output by the model, which comprehensively considers the platform loading delay characteristics and the timeliness of user behavior to ensure efficient loading of content on the device side.

[0126] In the embodiments of the present application, first, the multi-platform delay feature set is standardized (for example, normalizing the platform switching time interval, the delay mean offset and the variance), and is input into the pre-trained time sequence compensation model. The model captures the correlation between historical delay data and real-time features (for example, the platform with high peak delay needs to extend the push time window), and predicts the delay compensation coefficient and the time margin correction amount. For example, if the model detects that the delay fluctuation rate of a certain platform is high and the switching frequency is high, a larger time margin compensation value is output, and the original push time window is extended. Finally, the corrected push time window is generated, including the dynamically adjusted time range and the floating threshold.

[0127] In order to solve the problem of decreased first-screen loading efficiency caused by device resolution difference and improve the fluency of content loading on devices with different resolutions, in some embodiments, according to step 105, based on the corrected push time window, the transmission queue order of the adaptive content block set is redistributed according to the screen resolution difference of different device sides to obtain an optimal content block set, including:

[0128] Step 501, according to the time margin of the modified push time window, combining the ratio between the current device end screen resolution and the preset reference resolution, generate resolution adaptation parameter;

[0129] In this step, the resolution adaptation parameter is a dynamic generated adjustment factor based on the modified push time window time margin and the device end screen resolution difference, used to quantify the adaptation degree of block transmission priority and device display characteristics, its value is positively related to the time margin and negatively related to the resolution difference.

[0130] In the embodiments of the present application, first, the ratio of the current device end screen resolution (such as mobile phone screen resolution) and the preset reference resolution (such as tablet reference resolution) is obtained, and the resolution difference coefficient (such as mobile phone resolution / tablet resolution) is calculated. Then, the time margin of the modified push time window (such as the extended time window) and the resolution difference coefficient are fused through a preset function model (such as product or weighted sum) to generate the resolution adaptation parameter. For example, when the time margin is long and the resolution difference is small, the adaptation parameter is high, indicating that high-resolution blocks can be transmitted preferentially; otherwise, the adaptation parameter is low, and the blocks need to be compressed or split preferentially.

[0131] Step 502, the time margin and the resolution adaptation parameter are weighted and fused to generate the transmission priority weight of each block, and the adaptive content block set is arranged in descending order based on the transmission priority weight of each block to generate the initial transmission queue order;

[0132] In this step, the transmission priority weight is a block loading priority quantitative value generated by fusing the time margin and the resolution adaptation parameter, and the higher the weight, the higher the priority of the block in the transmission queue. The initial transmission queue order is a block loading order list arranged in descending order according to the transmission priority weight, used to guide the content blocks to be transmitted to the device end according to the priority.

[0133] In the embodiments of the present application, first, the time margin of the modified push time window (such as the extended time window) and the resolution adaptation parameter (such as the medium compression ratio) are weighted and fused according to the preset weight ratio (such as the time margin weight accounts for a high proportion), to generate the transmission priority weight of each block. For example, the block with sufficient time margin and high resolution adaptation parameter (such as the title block) has a higher weight. Then, all blocks are arranged in descending order according to the weight to generate the initial transmission queue order, ensuring that high-weight blocks are loaded to the first-screen core position preferentially.

[0134] Step 503, based on the product of the current device end screen resolution and the modified time margin, dynamically determine the capacity range of the first screen core queue, select the block with weight greater than the preset threshold from the initial transmission queue order as the first screen core queue, and compress the block with resolution lower than the device end screen resolution to generate a compression instruction set containing a compression algorithm identifier;

[0135] In this step, the first screen core queue refers to the high-weight block set selected from the initial transmission queue, and its capacity range is dynamically limited by the product of the device screen resolution and the modified time margin, ensuring that the priority loading content adapts to the first screen display space; the compression algorithm identifier is a marker identifying different compression algorithm types (such as lossless compression, lossy compression), used to guide the block to perform resolution compression according to the specified algorithm.

[0136] In the embodiment of the present application, first, the capacity upper limit of the first screen core queue is calculated based on the product of the current device end screen resolution and the modified time margin (e.g., the capacity upper limit is large when the resolution is high and the time margin is long). The block with weight greater than the preset threshold (e.g., the top 30% of the weight) is selected from the initial transmission queue order as the first screen core queue. At the same time, for the block with resolution lower than the device end screen resolution (e.g., the resolution is insufficient when the large image block on the tablet is displayed on the mobile phone), a compression instruction set is generated according to the preset compression algorithm library (e.g., WebP lossy compression, PNG lossless compression), which identifies the compression algorithm type and parameters to be executed.

[0137] Step 504, real-time monitoring of the cumulative loading time length of the first screen core queue, when detecting that the cumulative loading time length exceeds the preset proportion threshold of the modified push time window, marking the block causing overrun as a time sequence overrun block, and according to the compression instruction set and the resolution of the time sequence overrun block, the time sequence overrun block is split into multiple sub-block sequences to obtain a split record containing a sub-block loading time sequence marker;

[0138] In this step, the cumulative loading time length is the total time from the start of loading to the current time of all blocks in the first screen core queue, used to monitor the matching of the actual loading progress and the time window. The preset proportion threshold is the allowed cumulative loading time length proportion in the modified push time window (e.g., 80% of the total time length of the time window), and the threshold is exceeded, which is determined as overrun. The time sequence overrun block is the block that causes the cumulative loading time length to exceed the preset proportion threshold, which needs to be dynamically split and optimized. The sub-block loading time sequence marker is the loading order identifier of the sub-block in the transmission queue after splitting, used to guide the sub-block to be loaded in batches according to the priority.

[0139] In the embodiments of the present application, first, the cumulative loading duration of the first screen core queue (such as the sum of the title block loading time and the text block loading time) is monitored in real time. When it is detected that the cumulative duration exceeds the preset proportion threshold (such as 80% of the total duration) of the modified push time window, the last loaded block is located as the time sequence overrun block. Subsequently, according to the algorithm (such as lossy compression) specified by the compression instruction set and the resolution of the time sequence overrun block, it is split into multiple sub-block sequences, wherein the semantic association splitting identifies the logical segmentation points (such as the first sentence of a paragraph and the title of a chart) in the block through natural language processing technology, and splits the sub-blocks into upper and lower superimposed sub-blocks according to the semantic association degree; and the resolution adaptation splitting is to cut the large-size block into sub-blocks according to the device screen resolution in the horizontal or vertical direction, and finally the sub-block size, the parent block identifier and the loading time sequence mark (such as sub-block A immediate loading and sub-block B delayed loading) are generated, forming a splitting record, and updating to the transmission queue.

[0140] Step 505, according to the splitting record, inserting the sub-blocks in the multiple sub-block sequences into the non-overrun position of the initial transmission queue order, and recalculating the transmission priority weight of all blocks in the initial transmission queue order, and generating the optimal content block set based on the recalculated transmission priority weight;

[0141] In order to solve the problem of user intention recognition deviation caused by the dispersion of cross-platform behavior data, in some embodiments, according to step 102, the real-time behavior data set is input into the dynamic aggregation module to generate a preference feature sequence reflecting the user's current scene intention, including:

[0142] Step 601, time axis alignment and outlier filtering are performed on the cross-platform operation time stamps in the real-time behavior data set to generate a cross-platform behavior log containing user cross-platform switching behavior;

[0143] In this step, time axis alignment refers to unifying the cross-platform operation time stamps recorded by different devices or platforms to the same time reference (such as Coordinated Universal Time UTC), eliminating the problem of time inconsistency caused by device clock deviation. Outlier filtering is to identify and eliminate time stamp data (such as future time stamps and continuous switching in a very short time) that do not conform to user behavior logic through a rule engine or a statistical model

[0144] In an embodiment of the present application, the timestamp of each device is first calibrated through a device clock synchronization protocol (such as the Network Time Protocol NTP), and the switching operation time recorded on different platforms is unified to the same time axis. For example, the switching timestamp recorded on the user's mobile phone is the local time of the device, which is converted to UTC time after synchronization. Subsequently, an outlier detection algorithm (such as sliding window statistics based on time intervals) is used to identify outliers: if the time interval between adjacent switching operations is lower than a preset threshold (such as multiple switches within seconds), or the timestamp exceeds a reasonable range (such as earlier than the account registration time), it is marked as abnormal and removed. Finally, a cross-platform behavior log is generated, which records the platform type, device identification and calibrated timestamp of the user's switching device in chronological order.

[0145] Step 602: Based on the page dwell time and the type of interaction between the user and the content, dynamic weights are assigned to user behaviors in the real-time behavior data set using a time decay function to generate a weighted behavior data set containing behavior timeliness markers.

[0146] In this step, the weighted behavior dataset is a structured data set that integrates the timeliness of user behavior and the value of content interaction. It uses a time decay model to strengthen the decision weight of recent behavior. The time decay function exponentially reduces the influence of historical behavior, while the interaction type weighting rule dynamically adjusts the weight coefficient based on the value level of the behavior (for example, liking is more important than swiping).

[0147] In an embodiment of the present application, the length of time users stay on pages of each platform and the type of interaction (such as favorites and comments) are first extracted, and the time decay weight of the behavior is calculated through an exponential decay model. For example, the behavior on the same day retains its full weight, and the weight of the behavior three days ago decays to a negligible value. Subsequently, the original behavior data is weighted according to a preset interaction value mapping table (such as high weights for likes and low weights for swipes), and a weighted data set with timeliness tags (such as "high timeliness" and "low timeliness") is generated. This data set highlights the user's recent high-value behaviors and provides high-confidence input for intent recognition.

[0148] Step 603: Extract the device switching frequency and interaction type distribution characteristics of the user in the same physical scene from the cross-platform behavior log, and combine them with the dynamic weights in the weighted behavior dataset to construct a multi-dimensional behavior association matrix;

[0149] In this step, the multi-dimensional behavior association matrix is ​​a feature space mapping model that quantifies the correlation between user cross-device behavior and content preferences. It constructs a three-dimensional device-platform-behavior association network by integrating device switching frequency, interaction type distribution, and timeliness weight.

[0150] In the embodiments of the present application, first, the device switching density of the user in the same physical scene (such as the number of switching times per unit time) is extracted from the cross-platform behavior log, and the interaction type distribution of each platform (such as WeChat mainly likes and Douyin mainly slides) is counted. Subsequently, the dynamic weight in the weighted behavior data set and the interaction type distribution are matrixed and fused according to the platform-device dimension, for example, the platform corresponding to the high-frequency switching device is given a basic weight bonus, and the high-time-efficiency interaction behavior further superimposes a weight coefficient. Finally, the dimension is reduced through the non-negative matrix factorization algorithm to generate an interpretable device-behavior association matrix, which is used to identify the cross-device scene intent.

[0151] Step 604, input the multi-dimensional behavior association matrix into a sequence modeling network based on a multi-head attention mechanism to generate a scene intent vector by capturing the spatio-temporal dependence of cross-platform behavior;

[0152] In this step, the m scene intent vectors are hidden space feature vectors generated after modeling the spatio-temporal dependence of cross-platform behavior by the multi-head attention mechanism, and the dimension encodes the behavior association strength and intent tendency of the user between different devices and platforms. The multi-head attention mechanism captures the time sequence mode of device switching, the spatial correlation of interaction type distribution, and the dynamic influence of behavior time efficiency through parallel multi-group self-attention calculation.

[0153] In the embodiments of the present application, first, the multi-dimensional behavior association matrix is divided into multiple sub-matrices according to the time sequence, and each sub-matrix represents the cross-platform behavior characteristics of the user in a time segment. Subsequently, the sub-matrix is input into the multi-head attention network, and three groups of attention mechanisms are used to model different dimensions of behavior association through parallel calculation: the first group focuses on the change mode of device switching frequency over time to identify the cross-device operation rhythm of the user in different time periods; the second group captures the spatial correlation of interaction type in the cross-platform distribution, such as the mutual exclusivity feature of long text reading and short video browsing; the third group fuses the time efficiency label in the weighted behavior data set to strengthen the decision-making influence of recent high-weight behavior. The weighted feature vectors output by the three groups of attention are fused through a fully connected layer to generate a scene intent vector, and the hidden layer dimension represents the comprehensive intent of the user in the current scene, such as the tendency feature of cross-device continuous reading or multi-platform fragmented browsing.

[0154] Step 605, feature decoupling and dimension reduction processing are performed on the scene intent vector to output a preference feature sequence reflecting the current scene intent of the user;

[0155] In this step, the preference feature sequence is an interpretable feature sequence generated after feature decoupling and dimension reduction of the scene intent vector, and the dimension is mapped to the intent label (such as "long text preference" and "cross-device collaboration demand") that the user can perceive. Feature decoupling separates the independent components of the intent vector through orthogonal constraint, and dimension reduction processing retains the core intent dimension and eliminates noise.

[0156] In the embodiments of the present application, first, the scene intention vector is subjected to orthogonal decoupling constraint, forcing different dimensions of features to be independent of each other (such as “reading depth” and “device switching frequency” are decoupled into independent components). Subsequently, the high-dimensional vector is mapped to a low-dimensional space by a nonlinear dimension reduction algorithm (such as t-SNE or UMAP) to generate a preference feature sequence. Each feature dimension is associated with a preset semantic label library (such as “long text reading”, “video preference”, “cross-device collaboration”), and the final output preference label sequence is determined by maximum weight matching.

[0157] In order to achieve the balance between content conversion priority and push timeliness, and improve the overall efficiency of cross-platform content distribution, in some embodiments, according to step 103, the preference feature sequence is input into a cross-platform collaboration analysis module to generate a decision parameter group including content format conversion priority and push time window.

[0158] Step 701, performing multi-platform relevance analysis on the preference feature sequence, extracting the correlation weight between content format constraints and user behavior preferences of each media platform through multi-head attention mechanism, and generating a platform feature correlation matrix;

[0159] In this step, the platform feature correlation matrix is a multi-dimensional matrix quantifying the correlation weight between user behavior preferences and self-media platform content format constraints, the row dimension is mapped to platform content format features (such as long text formatting rules, video resolution limits), and the column dimension is mapped to user preference features (such as long text reading inclination, short video interaction preference). Multi-head attention mechanism is used to capture the nonlinear correlation between user behavior sequence and platform format constraint, for example, strong correlation between long text preference and graphic platform or high weight matching between video preference and short video platform.

[0160] In the embodiments of the present application, first, the preference feature sequence is tensor-spliced with a preset platform content format constraint library (such as WeChat supporting multi-column long text and Douyin limiting vertical screen video) to form a behavior-format joint input. Subsequently, through multiple parallel attention heads of the multi-head attention network, different dimensional correlation features are extracted: the first group focuses on the adaptability of text preference to graphic platform format (such as the matching degree of long text reading depth to WeChat multi-column layout); the second group analyzes the compatibility of video preference to short video platform rules (such as the correlation strength of video interaction frequency to Douyin vertical screen template); the third group identifies the coupling of cross-platform collaboration demand and multi-device adaptation strategy (such as the weight relationship between cross-device reading preference and multi-end synchronization function). The outputs of the attention heads of each group are weighted and fused to generate a platform feature correlation matrix, and the matrix element value represents the correlation strength of the user's specific preference to the platform format constraint, which is used for subsequent dynamic decision-making.

[0161] Step 702, based on the platform feature correlation matrix, combining the layout rule library of the target platform and the historical content distribution efficiency data, the conversion priority of different content formats is calculated by dynamic programming algorithm, and the priority parameter set containing format compatibility score and conversion path is generated;

[0162] In this step, the format compatibility score is an index to quantify the adaptation degree of content format to the layout rules of the target platform, and the higher the score, the lower the format conversion cost; the conversion path is the technical link of content conversion from the original format to the adapted format of the target platform, including compression, blocking, layout reorganization and other operation steps; the priority parameter set is a structured decision parameter integrating the format compatibility score and the conversion path, which is used to guide the conversion order and resource allocation of content format.

[0163] In the embodiment of the application, first, the correlation weight of user preference features and platform format constraints is extracted based on the platform feature correlation matrix, and a multi-objective optimization function is constructed by combining the historical layout rule library of the target platform (such as the minimum column width limit of WeChat multi-column layout, the vertical screen video resolution threshold of Douyin) and the historical content distribution efficiency data (such as the loading time and user retention rate of a certain format). By dynamic programming algorithm, all possible format conversion paths are traversed, the format compatibility score of each path is calculated (such as the score of converting long text blocking to WeChat multi-column layout is higher than that of converting Douyin vertical screen summary), and the path with the highest score and the lowest conversion cost is selected to generate the priority parameter set. For example, long text blocking is preferentially converted into WeChat multi-column layout, and short video summary is converted into Douyin vertical screen template, and the compression ratio and reorganization rules of each path are recorded.

[0164] Step 703, according to the user behavior timeliness label and the cross-platform switching frequency in the preference feature sequence, an initial push time window is generated by a time series prediction model, and a time margin compensation correction is made by combining the content loading delay data of each media platform to generate a corrected push time window;

[0165] In this step, the initial push time window is the content loading time range predicted based on the user behavior timeliness label and the cross-platform switching frequency; the time margin compensation correction is a mechanism to dynamically extend the push time window according to the platform loading delay data, which is used to offset the delay risk caused by network fluctuations or resource competition.

[0166] In the embodiments of the present application, firstly, the user behavior timeliness label (such as "high timeliness" behavior needs quick response) in the preference feature sequence is marked and input into a time sequence prediction model (such as an LSTM network) across platforms to predict the platform active period of the user in the future time segment, and an initial push time window is generated. Subsequently, the content loading delay data (such as the delay fluctuation rate of Douyin in the peak period) of each media platform is extracted, and the time margin compensation value of each platform is calculated through a weighted compensation algorithm, for example, the time margin of the high delay fluctuation platform needs to be increased. Finally, the compensation value is superimposed on the initial time window to generate a corrected push time window, ensuring stable loading of the content on the device.

[0167] In step 704, the priority parameter set is fused with the corrected push time window in a multi-dimensional parameter, and a decision parameter group containing content format conversion priority, push time window and dynamic compensation threshold is output.

[0168] In this step, the decision parameter group is a multi-dimensional decision parameter set integrating content format conversion priority, corrected push time window and dynamic compensation threshold, which is used to guide the global strategy of content cross-platform distribution. The content format conversion priority determines the adaptation order of different formats, the push time window limits the time margin of content loading, and the dynamic compensation threshold sets the time window adjustment threshold value in different delay fluctuation scenarios.

[0169] In the embodiments of the present application, firstly, the format conversion path in the priority parameter set is mapped with the corrected push time window in the parameter space, and the format compatibility score, time window remaining time and platform delay compensation coefficient are associated to a unified decision space through a nonlinear fusion algorithm. For example, the content with high format compatibility score is allocated to a short time window to improve the loading efficiency, and the content with low score is relaxed in time window and superimposed with dynamic compensation threshold. Subsequently, based on the dynamic compensation threshold rule library (such as triggering time window extension when the delay fluctuation rate exceeds the threshold), adaptive adjustment strategies in different scenarios are generated. Finally, the output decision parameter group contains block conversion order, time window constraint and delay compensation trigger condition, forming a push strategy that can dynamically respond to network environment and user behavior.

[0170] Figure 2 A structural schematic diagram of a self-media streaming push system based on big data analysis is provided in the embodiments of the present application, as shown in FIG. 1, the system comprises: Figure 2

[0171] The collection module 21 collects a real-time behavior data set of the user on different self-media platforms, and the real-time behavior data set contains the cross-platform operation timestamp, page stay duration and content interaction type of the user when switching devices in the same physical scene.

[0172] ​The first input module 22 is configured to input the real-time behavior data set into the dynamic aggregation module to generate a preference feature sequence reflecting the current scene intention of the user.

[0173] The second input module 23 is configured to input the preference feature sequence into the cross-platform collaborative analysis module to generate a decision parameter group including a content format conversion priority and a push time window.

[0174] The disassembly module 24 is configured to trigger the content reorganization engine according to the push time window in the decision parameter group, and dynamically disassemble original content materials based on the screen resolution of the current active device end of the user to generate an adaptive content block set matching the layout rules of the target platform.

[0175] The correction module 25 is configured to correct the push time window in the decision parameter group according to the content loading delay data of each self-media platform when detecting that there are at least three continuous switching events of the self-media platforms within a preset time threshold in the real-time behavior data set, and reassign the transmission queue order of the adaptive content block set according to the screen resolution difference of different device ends based on the corrected push time window to obtain an optimal content block set.

[0176] Figure 2 The self-media stream pushing system based on big data analysis can perform Figure 1 The self-media stream pushing method based on big data analysis of the embodiment has the same implementation principles and technical effects as the self-media stream pushing system based on big data analysis, and will not be described here.

[0177] In one possible design, Figure 2 The self-media stream pushing system based on big data analysis of the embodiment can be implemented as a computing device, such as a server. Figure 3 As shown in the figure, the computing device can include a storage component 31 and a processing component 32.

[0178] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32.

[0179] The processing component 32 is configured to perform the above Figure 1 The self-media stream pushing method based on big data analysis of the embodiment.

[0180] The processing component 32 can include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component can also be one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components, configured to perform the methods described above.

[0181] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage devices or their combinations, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0182] Of course, the computing device can also include other components, such as an input / output interface, a display component, a communication component, etc.

[0183] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0184] The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, etc.

[0185] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform, and the computing device can be a cloud server, and the processing component, the storage component, etc. can be a basic server resource rented or purchased from the cloud computing platform.

[0186] The embodiment of the application further provides a computer storage medium, which stores a computer program, and the computer program can implement the above Figure 1 The embodiment shown in the figure is a self-media flow pushing method based on big data analysis.

[0187] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0188] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0189] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary general hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, and the computer software products can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and include a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0190] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A self-media streaming push method based on big data analysis, characterized in that: include: Collect real-time behavioral data sets of users on different self-media platforms. The real-time behavioral data sets include cross-platform operation timestamps, page dwell time, and content interaction types when users switch devices in the same physical scenario. Inputting the real-time behavior data set into a dynamic aggregation module to generate a preference feature sequence reflecting the user's current scenario intention; Inputting the preference feature sequence into a cross-platform collaborative analysis module to generate a decision parameter group including content format conversion priority and push time window; The content reorganization engine is triggered according to the push time window in the decision parameter group, and the original content material is dynamically disassembled based on the screen resolution of the user's currently active device to generate a set of adapted content blocks that match the layout rules of the target platform; When continuous switching events of at least three or more self-media platforms within a preset time threshold are detected in the real-time behavior data set, the push time window in the decision parameter group is corrected according to the content loading delay data of each self-media platform, and based on the corrected push time window, the transmission queue order of the adapted content block set is reallocated according to the screen resolution differences of different device ends to obtain the optimal content block set.

2. The method according to claim 1, characterized in that The content reorganization engine is triggered according to the push time window in the decision parameter group, and the original content material is dynamically disassembled based on the screen resolution of the user's currently active device to generate a set of adapted content blocks that match the target platform's layout rules, including: Calculate a dynamic block threshold for screen resolution based on the remaining duration of the push window and the content loading rate limit of the target platform. The dynamic block threshold is an inversely proportional function of the horizontal pixel value of the screen and the remaining duration of the push window. Based on the dynamic block threshold, the image and text mixed area of ​​the original content material is divided into non-uniform grids, and according to the semantic relevance of the image and text elements in the divided grids, a candidate block library containing priority tags is generated; Performing a typesetting compatibility assessment on each candidate block in the candidate block library, and generating a typesetting compatibility matrix including block size correction coefficients and position offsets by matching the image-text spacing, font scaling ratio, and multi-column layout rules in the historical typesetting template of the target platform; According to the correction coefficient in the layout compatibility matrix and the dynamic weight coefficient of the screen refresh rate of the currently active device, the candidate block library is adaptively reorganized across hurdles to generate a set of adapted content blocks that match the layout rules of the target platform.

3. The method according to claim 2, characterized in that Based on the correction coefficient in the layout compatibility matrix and the dynamic weight coefficient of the screen refresh rate of the currently active device, the candidate block library is adaptively reorganized across hurdles to generate a set of adapted content blocks that match the layout rules of the target platform, including: Extracting a block size correction coefficient and a position offset of each block from the typesetting compatibility matrix, fusing the block size correction coefficient with the dynamic block threshold according to a linear weighted algorithm, and generating a priority ranking value for each block; Multiplying the priority ranking value by the dynamic weight coefficient of the screen refresh rate to obtain a dynamic scheduling parameter group including a block column mapping relationship, a vertical compression ratio coefficient, and a loading timing weight; Arrange the candidate blocks in the candidate block library in descending order according to the priority ranking value in the dynamic scheduling parameter group, fill the candidate blocks into the core columns of the first screen of the target platform according to the arrangement result, and compress the spacing of the mixed image and text area according to the vertical compression ratio coefficient to obtain the compressed block vertical compression parameters and column layout data; Based on the column layout data, the total vertical pixel value of the blocks within the core column of the first screen of the current device is calculated in real time. When it is detected that the total vertical pixel value exceeds the preset range of vertical pixels of the device screen, the block that causes the vertical space to exceed the limit is marked as a vertically exceeded block. The vertically exceeded block is split into a combination of sub-blocks stacked up and down according to the semantic relevance of the image and text, and a block split record is generated, which includes the sub-block size and the parent block identifier; The position offset in the typesetting compatibility matrix is ​​updated according to the block splitting record, and a set of adapted content blocks is generated in combination with the loading timing weight in the dynamic scheduling parameter group.

4. The method according to claim 1, wherein When it is detected that at least three or more self-media platforms continuously switch events within a preset time threshold in the real-time behavior data set, the push time window in the decision parameter group is modified according to the content loading delay data of each self-media platform, including: Extracting the self-media platform identifier that triggers the self-media platform switching operation and the time interval between adjacent switching operations in the continuous switching events, and generating an original delay dataset containing the time interval of the self-media platform switching sequence; Calculate the delay fluctuation rate and peak delay value based on the content loading delay data of each self-media platform in the original delay dataset; Based on the time interval of the self-media platform switching sequence, the delay fluctuation rate, and the peak delay value, a multi-platform delay feature set including delay mean offset and variance is generated through a weighted fusion algorithm; The multi-platform delay feature set is input into a timing compensation model to generate a corrected push time window.

5. The method according to claim 1, wherein Based on the revised push time window, the transmission queue order of the adapted content block set is reallocated according to the screen resolution differences of different device terminals to obtain the optimal content block set, including: Generate a resolution adaptation parameter based on the time margin of the corrected push time window and the ratio between the current device screen resolution and the preset reference resolution; Performing weighted fusion of the time margin and the resolution adaptation parameter to generate a transmission priority weight for each block, and arranging the adapted content block set in descending order based on the transmission priority weight for each block to generate an initial transmission queue order; Based on the product of the current device screen resolution and the corrected time margin, dynamically determine the capacity range of the first-screen core queue, select blocks with weights greater than a preset threshold from the initial transmission queue sequence as the first-screen core queue, and compress blocks with resolutions lower than the device screen resolution to generate a compression instruction set including a compression algorithm identifier; Real-time monitoring of the cumulative loading time of the first screen core queue. When it is detected that the cumulative loading time exceeds a preset ratio threshold of the revised push time window, the block that exceeds the limit is marked as a timing-exceeded block, and the timing-exceeded block is split into multiple sub-block sequences according to the compression instruction set and the resolution of the timing-exceeded block, to obtain a split record containing the sub-block loading timing mark; According to the split records, the sub-blocks in the multiple sub-block sequences are inserted into the unexceeded positions of the initial transmission queue sequence, and the transmission priority weights of all the sub-blocks in the initial transmission queue sequence are recalculated. Based on the recalculated transmission priority weights, an optimal content block set is generated.

6. The method according to claim 1, characterized in that The real-time behavior data set is input into a dynamic aggregation module to generate a preference feature sequence reflecting the user's current scenario intention, including: Performing time axis alignment and outlier filtering on the cross-platform operation timestamps in the real-time behavior data set to generate a cross-platform behavior log containing the user's cross-platform switching behavior; Based on the page dwell time and the type of interaction between users and content, a time decay function is used to assign dynamic weights to user behaviors in the real-time behavior data set, generating a weighted behavior data set containing behavior timeliness markers. Extracting the device switching frequency and interaction type distribution characteristics of users in the same physical scene from the cross-platform behavior logs, and combining them with the dynamic weights in the weighted behavior dataset to construct a multi-dimensional behavior association matrix; Input the multi-dimensional behavior association matrix into a sequence modeling network based on a multi-head attention mechanism to generate a scene intention vector by capturing the spatiotemporal dependencies of cross-platform behaviors; The scene intention vector is subjected to feature decoupling and dimensionality reduction processing to output a preference feature sequence reflecting the user's current scene intention.

7. The method according to claim 1, characterized in that The preference feature sequence is input into the cross-platform collaborative analysis module to generate a decision parameter group including content format conversion priority and push time window, including: Perform multi-platform correlation analysis on the preference feature sequence, extract the correlation weights between the content format constraints and user behavior preferences of each self-media platform through a multi-head attention mechanism, and generate a platform feature correlation matrix; Based on the platform feature association matrix, combined with the target platform's typesetting rule library and historical content distribution efficiency data, the conversion priorities of different content formats are calculated through a dynamic programming algorithm to generate a priority parameter set including format compatibility scores and conversion paths; Based on the user behavior timeliness markers and cross-platform switching frequency in the preference feature sequence, an initial push time window is generated through a time series prediction model. Time margin compensation correction is performed based on the content loading delay data of each self-media platform to generate a corrected push time window; The priority parameter set is subjected to multi-dimensional parameter fusion with the revised push time window, and a decision parameter group including content format conversion priority, push time window and dynamic compensation threshold is output.

8. A self-media streaming push system based on big data analysis, characterized in that: include: The collection module elegantly collects real-time behavioral data sets of users on different self-media platforms. The real-time behavioral data sets include cross-platform operation timestamps, page dwell time, and content interaction types when users switch devices in the same physical scene; A first input module is used to input the real-time behavior data set into a dynamic aggregation module to generate a preference feature sequence reflecting the user's current scenario intention; A second input module is configured to input the preference feature sequence into a cross-platform collaborative analysis module to generate a decision parameter group including a content format conversion priority and a push time window; A disassembly module is configured to trigger a content reassembly engine according to the push time window in the decision parameter group, dynamically disassemble the original content material based on the screen resolution of the user's currently active device, and generate a set of adapted content blocks that match the target platform's layout rules; The correction module is used to correct the push time window in the decision parameter group according to the content loading delay data of each self-media platform when detecting the occurrence of continuous switching events of at least three or more self-media platforms within a preset time threshold in the real-time behavior data set, and based on the corrected push time window, reallocate the transmission queue order of the adapted content block set according to the screen resolution differences of different device ends to obtain the optimal content block set.

9. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a self-media streaming push method based on big data analysis as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, a self-media streaming push method based on big data analysis as described in any one of claims 1 to 7 is implemented.

Images