AIGC short video production method and system based on user feedback label

Abstract

The invention discloses an AIGC short video production method and system based on a user feedback tag, and relates to the technical field of short video production, and the method specifically comprises the following steps: obtaining the clip attention information of a user for each video clip in real time, carrying out the analysis after obtaining, evaluating the attention degree of the user for each video clip in a short video, and obtaining a user feedback tag; dividing each video clip into a high-attention clip, a medium-attention clip and a low-attention clip according to an evaluation result; constructing and applying a content cutting adjustment mechanism according to a division result of each video clip; and in the application process of the content clipping adjustment mechanism, clipping information of each video clip is acquired in real time, analysis is performed after acquisition, and whether the clipping degree of each video clip meets a preset content optimization standard is evaluated. According to the invention, the problem that the video cannot be intelligently clipped is solved, and the effects of accurately identifying the watching preference of the user and optimizing the video content are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of short video production, and in particular to an AIGC short video production method and system based on user feedback tags. Background Art

[0002] Short video production refers to the process of editing and processing video materials and combining them with music, subtitles, special effects and other elements to generate multimedia works with short duration, concentrated content and easy dissemination. The traditional short video production process requires creators to proceed step by step from planning, shooting to post-editing, and hand-made content often relies on the personal creativity and editing skills of the creators. However, the AIGC (artificial intelligence generated content) short video production method based on user feedback tags is unique: it uses advanced artificial intelligence technology to extract feedback tags based on users' viewing history and interactive behaviors (such as likes, comments, sharing, etc.), and analyzes these tags through deep learning models to automatically generate short video content that is highly matched with user interests. This method not only greatly improves the efficiency of content generation, but also can dynamically adjust video content in real time to ensure that the videos seen by each user are more in line with their personalized preferences. Compared with traditional short video production, AIGC short video production based on user feedback tags is highly automated and intelligent. It can not only quickly respond to the different needs of a large number of users, but also continuously monitor user feedback after content generation, optimize video effects, provide more accurate content recommendations, meet users' diverse preferences, and improve user participation and satisfaction. This gives AIGC short video production a great advantage in personalized content distribution and user experience optimization.

[0003] The existing AIGC short video production technology based on user feedback tags completes short video production through the following steps: First, the system collects users' viewing history, likes, comments, sharing and other behavioral data, and uses natural language processing and data analysis technology to extract users' interest tags; then, the deep learning model analyzes these tags, intelligently matches related content, and generates personalized short video scripts or picture frames; then, the AI ​​generator automatically creates visual content, or selects suitable video clips from the existing material library for integration; after the video is generated, the system monitors user feedback data in real time through machine learning algorithms, dynamically adjusts and optimizes video content to make it more in line with user preferences, and adapts and distributes it on different platforms to ensure a personalized and highly customized viewing experience.

[0004] The prior art has the following deficiencies:

[0005] When users want to generate short videos, the system can automatically crop the video content according to their long-term repeated viewing behavior of certain clips. However, the system cannot perform intelligent cropping according to the user's viewing behavior. Since the existing technology relies on fixed templates and preset editing rules, the system cannot dynamically adjust the video clips according to the user's actual viewing time and repeated behavior, resulting in the content of interest to the user being diluted by lengthy and irrelevant clips. The existing technology lacks the ability to intelligently identify and process user viewing preferences in real time, so it is impossible to accurately locate the parts that users are highly concerned about and perform corresponding cropping optimization. This will cause the lengthy and unnecessary clips in the video to be unable to be removed, which will cause users to be dissatisfied with the rhythm of the video during the viewing process. Especially when users are extremely concerned about certain clips, the redundant content will weaken their concentration on the key parts, thereby reducing the overall viewing experience.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0007] The purpose of the present invention is to provide an AIGC short video production method and system based on user feedback tags to solve the problems in the above-mentioned background technology.

[0008] In order to achieve the above object, the present invention provides the following technical solution: an AIGC short video production method based on user feedback tags, specifically comprising the following steps:

[0009] When users watch short videos, the viewing behavior information of users is collected in real time, and the viewing patterns of users are analyzed based on the information. When the analysis results show that the short video content needs to be trimmed, the short video to be trimmed is evenly divided into several video clips.

[0010] Acquire the user's attention information on each video segment in real time, analyze it after acquisition, evaluate the user's attention to each video segment in the short video, and classify each video segment into high attention segment, medium attention segment and low attention segment according to the evaluation result;

[0011] According to the results of the division of each video segment, a content cropping and adjustment mechanism is constructed and applied to dynamically adjust the content of each video segment with different attention levels;

[0012] During the application of the content cropping adjustment mechanism, the cropping information of each video clip is obtained in real time, and analyzed after acquisition to evaluate whether the cropping degree of each video clip meets the preset content optimization standard, and the content cropping adjustment mechanism is optimized according to the evaluation result;

[0013] According to the optimized content cropping and adjustment mechanism, continuously monitor users' viewing feedback on the cropped video content, and continuously optimize the content cropping and adjustment mechanism based on these feedback data.

[0014] Preferably, the user's attention information on each video segment is obtained in real time, and analyzed after being obtained to evaluate the user's attention to each video segment in the short video, and the video segments are divided into high attention segments, medium attention segments and low attention segments according to the evaluation results, which specifically includes the following steps:

[0015] Acquire the user's attention information on each video segment in real time and perform preprocessing after acquisition;

[0016] Extracting the pre-processed user's viewing behavior participation information and segment attractiveness behavior information from the segment attention information of each video segment, and analyzing them after extraction to generate the viewing participation coefficient and segment attractiveness index of each video segment respectively;

[0017] The viewing engagement coefficient and segment attractiveness index of each generated video clip are used to construct an attention degree evaluation model to generate an attention coefficient for each video clip. The generated attention coefficient for each video clip is compared with a pre-set attention coefficient threshold range. The user's attention degree to each video clip in the short video is evaluated based on the comparison result, and each video clip is divided into high-attention clips, medium-attention clips and low-attention clips according to the evaluation results.

[0018] Preferably, the logic for obtaining the viewing participation coefficient and the segment attractiveness index of each video segment is as follows:

[0019] Extract the viewing behavior participation information of the pre-processed user attention information on each video clip, including the total duration of each video clip in the short video, the total viewing time of the user on each video clip in the short video, and the frequency of the user's interactive behavior on each video clip at different times within a period of time, and mark them as LSE respectively. i 、YWA i and LSE i Indicates the total duration of the i-th video clip in the short video, YWA i represents the total viewing time of the user on the i-th video segment in the short video, represents the frequency of user interaction on the i-th video clip at time m within a period of time, i = 1, 2, 3, ..., k, m = 1, 2, 3, ..., g, k and g are both positive integers;

[0020] Calculate the viewing engagement coefficient of each video clip. The specific calculation formula is as follows:

[0021]

[0022] Where, EIC i is the viewing engagement coefficient of the i-th video clip;

[0023] Extract the segment attraction behavior information from the pre-processed user segment attention information of each video segment, including the number of repeated views and skips of each video segment in the short video at different times within a period of time, and the total number of views of each video segment in the short video by the user, and mark them as and NTO i , It indicates the number of times the user repeatedly watches the i-th video clip in the short video at time m within a period of time. NTO represents the number of times a user skips the i-th video segment in the short video at time m within a period of time. i Indicates the total number of times the user has watched the i-th video clip in the short video;

[0024] Calculate the segment attractiveness index of each video segment. The specific calculation formula is as follows:

[0025]

[0026] In the formula, SAI i is the segment attractiveness index of the i-th video segment.

[0027] Preferably, the viewing engagement coefficient EIC of each video clip generated is i and Segment Attractiveness Index SAI i Construct an attention evaluation model and generate the attention coefficient GZ of each video clip by weighted summation i , and the generated attention coefficients GZ for each video clip i and the pre-set attention coefficient threshold interval GZ min , GZ max ] to compare, evaluate the user's attention to each video segment in the short video based on the comparison results, and divide each video segment into high-attention segments, medium-attention segments and low-attention segments based on the evaluation results. The specific comparison analysis and division are as follows:

[0028] If GZ i <GZ min , the user's attention to the i-th video segment in the short video is low, then the video segment is classified as a low-attention segment;

[0029] If GZ min ≤GZ i ≤GZ max, the user's attention to the i-th video segment in the short video is medium, then the video segment is classified as a medium attention segment;

[0030] If GZ i >GZ max , if the user's attention to the i-th video segment in the short video is high, then the video segment is classified as a high-attention segment.

[0031] Preferably, a content cropping adjustment mechanism is constructed and applied according to the division results of each video segment, specifically: different content cropping adjustment parameters are set according to the division results of high-attention segments, medium-attention segments and low-attention segments to form a content cropping adjustment mechanism; the content cropping adjustment mechanism automatically determines the length and method of extending, retaining and cropping the content based on the attention coefficient of each video segment and the current segment length through pre-set rules;

[0032] For high-attention segments with a high degree of attention, the extension parameter in the content cutting adjustment mechanism is used to extend the playback time of the high-attention segments;

[0033] For the medium-attention segments with a medium attention level, the retention parameters in the content cropping adjustment mechanism are used to keep the current content unchanged;

[0034] For low-attention segments with a low degree of attention, the reduction parameters in the content cropping adjustment mechanism are used to crop redundant content and shorten the length of the low-attention segments.

[0035] Preferably, during the application of the content cropping adjustment mechanism, the cropping information of each video clip is obtained in real time, and analyzed after acquisition to evaluate whether the cropping degree of each video clip meets the preset content optimization standard, and the content cropping adjustment mechanism is optimized according to the evaluation result, which specifically includes the following steps:

[0036] In the process of applying the content cropping adjustment mechanism, cropping information of each video clip is obtained in real time and preprocessed after being obtained;

[0037] Extracting the cropping integrity information and the clipping smoothness information from the cropping information of each preprocessed video segment, and analyzing them after extraction to generate a cropping integrity coefficient and a clipping smoothness index for each video segment respectively;

[0038] The cropping completeness coefficient and clip smoothness index of each generated video clip are used to construct a cropping degree evaluation model, and the cropping coefficient of each video clip is generated respectively. The generated cropping coefficient of each video clip is compared with the pre-set cropping coefficient threshold of each video clip. According to the comparison result, it is evaluated whether the cropping degree of each video clip meets the preset content optimization standard, and the content cropping adjustment mechanism is optimized according to the evaluation result.

[0039] Preferably, the logic for obtaining the cropping integrity coefficient and the editing smoothness index of each video segment is as follows:

[0040] The cropping integrity information of each preprocessed video clip is extracted, including the proportion of the retained content after cropping of each video clip in the short video, the content change at different moments in the process of cropping each video clip in the short video, and the change in the total length of the clip, and they are marked as CR i , and CR i Indicates the proportion of the content retained after cropping the i-th video clip in the short video. It represents the content change at time n during the process of cropping the i-th video segment in the short video. represents the change in the total length of the i-th video segment in the short video at time n during the cropping process, i=1, 2, 3, ..., k, n=1, 2, 3, ..., h, k and h are both positive integers;

[0041] Calculate the cropping integrity coefficient of each video segment. The specific calculation formula is as follows:

[0042]

[0043] In the formula, CIC i is the cropping completeness coefficient of the i-th video segment;

[0044] Extract the clip smoothness information from the cropping information of each preprocessed video clip, specifically including the ratio of the transition time of each video clip in the total clip length after cropping each video clip in the short video, the change in playback speed at different moments in the process of cropping each video clip in the short video, and the number of cropping, and mark them as TTR respectively. i , and TTR i It represents the ratio of the transition time of the ith video segment in the total segment length after the video segments in the short video are cropped. Indicates the change in playback speed at time n during the process of cropping the i-th video segment in the short video. Indicates the number of times the i-th video segment in the short video is cropped at time n;

[0045] Calculate the clip smoothness index of each video clip. The specific calculation formula is as follows:

[0046]

[0047] Where, CSI i is the clip smoothness index of the i-th video clip.

[0048] Preferably, the cropping integrity coefficient CIC of each generated video segment is i and Clip Smoothness Index CSI i Construct a cropping degree evaluation model and generate the cropping coefficient CJ of each video segment through weighted summation i , and the cropping coefficients CJ of each generated video segment i The cropping factor threshold of each video segment is set Compare and evaluate whether the cropping degree of each video clip meets the preset content optimization standard based on the comparison results, and optimize the content cropping adjustment mechanism based on the evaluation results. The specific comparison and analysis are as follows:

[0049] like The cropping degree of the video clip meets the preset content optimization standard, and there is no need to optimize the content cropping adjustment mechanism;

[0050] like The degree of cropping of the video clip does not meet the preset content optimization standards, and the content cropping adjustment mechanism needs to be optimized, specifically including: optimizing the retention time, transition effect, editing frequency and playback speed of the video clip by dynamically adjusting the weight ratio of the cropping integrity coefficient and the editing smoothness index.

[0051] Preferably, the AIGC short video production system based on user feedback tags includes a real-time viewing behavior collection and analysis module, a segment attention evaluation and classification module, a dynamic content cutting and adjustment module, a cutting effect evaluation and mechanism optimization module, and a feedback-driven cutting mechanism optimization module;

[0052] The real-time viewing behavior collection and analysis module collects the user's viewing behavior information in real time while the user is watching the short video, and analyzes the user's viewing pattern based on this information. When the analysis result shows that the short video content needs to be trimmed, the short video to be trimmed is evenly divided into several video clips;

[0053] The segment attention evaluation and classification module obtains the segment attention information of users on each video segment in real time, analyzes it after acquisition, evaluates the user's attention to each video segment in the short video, and divides each video segment into high-attention segment, medium-attention segment and low-attention segment according to the evaluation results;

[0054] The dynamic content cropping and adjustment module builds and applies a content cropping and adjustment mechanism based on the division results of each video segment, and dynamically adjusts the content of each video segment with different attention levels;

[0055] The cropping effect evaluation and mechanism optimization module obtains the cropping information of each video clip in real time during the application of the content cropping adjustment mechanism, analyzes it after acquisition, evaluates whether the cropping degree of each video clip meets the preset content optimization standard, and optimizes the content cropping adjustment mechanism according to the evaluation results;

[0056] The feedback-driven cropping mechanism optimization module continuously monitors the user's viewing feedback on the cropped video content according to the optimized content cropping adjustment mechanism, and continuously optimizes the content cropping adjustment mechanism based on these feedback data.

[0057] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0058] 1. The present invention realizes dynamic cropping of video content by real-time monitoring of user viewing behavior and intelligent analysis of user viewing patterns. This process enables the system to accurately identify segments that users have watched repeatedly or ignored for a long time, thereby cropping content that meets the user's interests. Compared with traditional editing rules that rely on fixed templates, the present invention can make real-time adjustments based on the user's actual viewing behavior, greatly improving the accuracy and personalization of cropping, avoiding the interference of lengthy and irrelevant content, and thus improving the user's satisfaction with the video rhythm.

[0059] 2. By quantifying the user's viewing engagement and attractiveness for each clip, the system can automatically classify the clips into high, medium, and low attention categories, and dynamically adjust the duration and content retention ratio of each clip accordingly. This mechanism not only enhances the personalized display effect of video content, but also ensures the highlighting and optimization of key content, greatly improving the user's viewing experience. This real-time, data-driven classification and adjustment strategy can effectively solve the problem that existing technologies cannot accurately identify user viewing preferences.

[0060] 3. The present invention provides the ability of continuous improvement through a continuous feedback optimization mechanism after cropping. After generating a short video, the system will perform data analysis based on the user's real-time feedback and dynamically adjust the cropping mechanism to better meet the user's preferences. This closed-loop optimization process ensures that the video content can be continuously improved as user needs change, forming a highly intelligent and adaptive content generation system. This feedback-driven optimization method ensures that the system can quickly respond to users' viewing preferences and continuously improve the quality of video content and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0062] Figure 1 It is a flowchart of the AIGC short video production method and system based on user feedback tags of the present invention.

[0063] Figure 2 It is a module schematic diagram of the AIGC short video production method and system based on user feedback tags of the present invention. DETAILED DESCRIPTION

[0064] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0065] The present invention provides Figure 1 The AIGC short video production method based on user feedback tags shown in the figure specifically includes the following steps:

[0066] When users watch short videos, the viewing behavior information of users is collected in real time, and the viewing patterns of users are analyzed based on the information. When the analysis results show that the short video content needs to be trimmed, the short video to be trimmed is evenly divided into several video clips.

[0067] The user's viewing behavior information can be tracked in real time by embedding a behavior data collection module in the player. The specific implementation method is that when the user watches a short video, the system can record the user's interaction data, such as the viewing time of each video clip, the number of pauses, fast forward and rewind operations, repeatedly watched clips, skipped clips and other behaviors. The behavior data collection module can record the user's operation behavior in real time by triggering event listeners (such as clicks, slides, etc.), and send these data to the back-end server in the form of logs. Then, the system stores and processes these log data according to timestamps and user IDs for subsequent analysis of user viewing patterns. In order to ensure real-time performance, a streaming data processing architecture (such as Kafka, Spark Streaming) can be used to achieve real-time monitoring and collection of viewing behavior.

[0068] Based on the collected information on user viewing behavior, the system can analyze the user's viewing pattern through data analysis and machine learning technology. First, the system will calculate the user's viewing time, viewing frequency, skip rate and other indicators for each clip through statistical analysis, and use clustering algorithms (such as K-means) or rule models in combination with these indicators to determine the user's interest in certain clips. Next, based on these points of interest, the system will evaluate the overall structure of the short video. If the attention or viewing frequency of certain clips is lower than the set threshold, the system determines that the video needs to be cropped. In this case, the system can evenly divide the video to be edited into several clips through a video segmentation algorithm. The division criteria can be based on time (such as one clip every 5 seconds) or scene change detection (such as identifying scene switches through video analysis technology). After the division is completed, the video clips can be used for subsequent cropping optimization processing.

[0069] The reason for collecting users' viewing behavior information in real time and analyzing their viewing patterns is to accurately identify users' interests and concerns in each video clip. This is to solve the problem that the system in current technology cannot perform intelligent cropping based on users' viewing behavior. Existing technologies rely on fixed templates and preset editing rules, which cannot adapt to users' actual preferences, resulting in the retention of lengthy and irrelevant clips. By collecting and analyzing users' viewing behavior in real time, the system can dynamically adjust the content of short videos, so that redundant clips in the video can be removed and the content that users are most interested in can be strengthened. This not only improves the rhythm and fluency of the video, but also ensures that users can have a better experience when watching certain clips repeatedly, ultimately improving the overall viewing satisfaction. This intelligent cropping method is implemented by software, with high flexibility and automation.

[0070] Acquire the user's attention information on each video segment in real time, analyze it after acquisition, evaluate the user's attention to each video segment in the short video, and classify each video segment into high attention segment, medium attention segment and low attention segment according to the evaluation result;

[0071] In this embodiment, the segment attention information of the user on each video segment is obtained in real time, and analyzed after being obtained to evaluate the user's attention to each video segment in the short video, and the video segments are divided into high attention segments, medium attention segments and low attention segments according to the evaluation results, which specifically includes the following steps:

[0072] Acquire the user's attention information on each video segment in real time and perform preprocessing after acquisition;

[0073] By embedding a monitoring mechanism and a user behavior tracking module in the video player, the user's attention information on each video clip can be obtained in real time. The specific implementation method is that when the user plays a short video, the system collects the user's interactive data in real time through the user's click, drag, pause, fast forward, repeat playback and other behaviors. These data can be triggered by event listeners, such as monitoring the playback time, repeated viewing times, skipping behaviors and user interaction frequency of each video clip. Through a real-time connection with the back-end server, the user's viewing behavior data will be transmitted and recorded in the database in real time, and the viewing behavior and corresponding data of each clip will be archived with the timestamp as the index. It is also possible to ensure the delay-free transmission and collection of viewing data and ensure the real-time nature of data acquisition by using WebSocket or other real-time communication protocols.

[0074] Preprocessing is an important step to ensure the integrity and validity of data. It can clean and format the acquired raw data for subsequent analysis and calculation. First of all, the necessity of data preprocessing lies in: 1) the raw data may contain noise (such as short viewing pauses), and invalid behaviors need to be removed by filtering; 2) different types of behavioral data (such as clicks, pauses, and viewing time) need to be standardized to a unified scale for subsequent analysis. Specific preprocessing steps include: abnormal detection of viewing time and removal of outliers (such as short stays that users quickly skip); deduplication of behavioral data such as repeated viewing times to avoid repeated data statistics; finally, standardize the processed data, such as converting data of different dimensions such as viewing time and number of interactions into standardized values ​​of the same dimension. These preprocessing can be achieved through data cleaning algorithms and regularization formulas to ensure the quality and consistency of data input into the model.

[0075] Extracting the pre-processed user's viewing behavior participation information and segment attractiveness behavior information from the segment attention information of each video segment, and analyzing them after extraction to generate the viewing participation coefficient and segment attractiveness index of each video segment respectively;

[0076] The viewing engagement coefficient and segment attractiveness index of each generated video clip are used to construct an attention degree evaluation model to generate an attention coefficient for each video clip. The generated attention coefficient for each video clip is compared with a pre-set attention coefficient threshold range. The user's attention degree to each video clip in the short video is evaluated based on the comparison result, and each video clip is divided into high-attention clips, medium-attention clips and low-attention clips according to the evaluation results.

[0077] The predetermined attention coefficient threshold interval can be determined by analyzing the historical viewing behavior data of a large number of users, and using statistical analysis and machine learning algorithms to dynamically adjust and set the threshold. The specific implementation method is to first collect a large number of users' viewing behavior data on different video clips, including historical data on viewing engagement coefficients and clip attractiveness indexes. These data are classified through cluster analysis (such as K-means algorithm) to divide user behaviors into different attention levels. The system can calculate the mean and standard deviation of each classification based on these classification results, thereby determining the attention coefficient range of high-attention clips, medium-attention clips, and low-attention clips. Further optimization can be performed through machine learning models (such as decision trees or random forests), adjusting the threshold interval based on real-time data, so that the system can more accurately match the user's attention to different clips. In this way, the threshold interval can dynamically respond to the differences between different users and videos, ensuring the accuracy and flexibility of the evaluation results.

[0078] In this embodiment, the logic for obtaining the viewing participation coefficient and the segment attractiveness index of each video segment is as follows:

[0079] Extract the viewing behavior participation information of the pre-processed user attention information on each video clip, including the total duration of each video clip in the short video, the total viewing time of the user on each video clip in the short video, and the frequency of the user's interactive behavior on each video clip at different times within a period of time, and mark them as LSE respectively. i 、YWA i and LSE i Indicates the total duration of the i-th video clip in the short video, YWA i represents the total viewing time of the user on the i-th video segment in the short video, represents the frequency of user interaction on the i-th video clip at time m within a period of time, i = 1, 2, 3, ..., k, m = 1, 2, 3, ..., g, k and g are both positive integers;

[0080] The video player's log system and user behavior analysis module can be used to extract user viewing behavior information for each video clip. The specific implementation method is that the player will record the user's viewing behavior events (such as play, pause, interaction, etc.) during the playback of each video clip, and assign a unique identifier to each clip. By monitoring these events, the system records the user's total viewing time on each video clip, the frequency of interactive behaviors (such as likes, comments, shares, etc.), and the total duration of the clip in the log file. The log file will be uploaded to the back-end server periodically or in real time. Through database query and analysis scripts, the system can automatically extract the viewing data associated with each clip to ensure that the complete behavior record of each user on different clips is obtained.

[0081] The quantitative data extracted include: 1) total duration, which records the duration of each video clip through the player, and can be automatically extracted through the background configuration file or video stream metadata; 2) total viewing time, which calculates the actual viewing time spent by users on a specific clip through the system, and records user operations such as pause, fast forward and repeat viewing; 3) frequency of interactive behavior, which records the number of times users interact with each clip (such as likes or comments). These data are used to calculate the viewing engagement coefficient because they fully reflect the actual level of user engagement with the video clip: the total duration ensures a standardized comparison between different clips, the total viewing time quantifies the user's interest in a certain clip, and the interaction frequency further reveals the user's deep engagement. These data combined can provide a full perspective on user behavior and provide support for more accurate calculation of the engagement coefficient.

[0082] Calculate the viewing engagement coefficient of each video clip. The specific calculation formula is as follows:

[0083]

[0084] Where, EIC i is the viewing engagement coefficient of the i-th video clip;

[0085] This calculation formula is used to quantify the user's engagement with each video clip, using multiple factors for comprehensive calculation. First, Calculate the ratio of the user's viewing time in each video segment to the total length of the video segment to reflect the user's viewing intensity of the segment. The higher this ratio is, the greater the proportion of the user's viewing time in the segment is, indicating that the segment is more attractive to the user. Then, calculate the logarithmic function of the interaction behavior frequency This part measures the frequency of user interaction with the clip at different times. By taking the logarithm, the user's frequent interaction behaviors are compressed to prevent the index from being unstable due to too high number of interaction behaviors. At the same time, the average of these frequencies ensures the balance of interaction behaviors. This calculation method comprehensively considers the weights of both viewing time and interaction behavior, and can more accurately reflect the user's overall engagement with a certain video clip, making the assessment of engagement more accurate and in-depth.

[0086] The viewing engagement coefficient EIC of the i-th video segment i The size of directly reflects the user's attention to the video clip. The larger the engagement coefficient, the more time the user spends watching the clip (relative to the total length of the clip) and the more interactive operations (such as likes, comments, etc.) during the viewing process. Therefore, EIC i The larger the value, the higher the user's interest and engagement in the segment, indicating that the segment plays a more important role in the user's overall viewing experience. By comparing the viewing engagement coefficients of different segments, the system can accurately assess the user's attention to each segment, thereby reasonably dividing the high-attention segments, medium-attention segments, and low-attention segments.

[0087] Extract the segment attraction behavior information from the pre-processed user segment attention information of each video segment, including the number of repeated views and skips of each video segment in the short video at different times within a period of time, and the total number of views of each video segment in the short video by the user, and mark them as and NTO i , It indicates the number of times the user repeatedly watches the i-th video clip in the short video at time m within a period of time. NTO represents the number of times a user skips the i-th video segment in the short video at time m within a period of time. i Indicates the total number of times the user has watched the i-th video clip in the short video;

[0088] The user's segment attraction behavior information for each video clip can be extracted by embedding a behavior tracking module in the video player. The specific implementation method is that the player will record the user's repeated viewing behavior, skipping behavior, and complete viewing behavior when the user watches each video clip. The number of repeated views can be captured by capturing the user's behavioral events of playing the same clip multiple times, and the skipping behavior is recorded by monitoring the user's fast-forward or skipping operation events, while the total number of views is the number of times the user watches a certain clip from beginning to end each time. Each behavior will be timestamped, and the system can identify the user's operations on the same clip at different times. All of this behavioral data will be recorded in log files and extracted in real time through the background data processing service for further analysis.

[0089] Specific quantitative data include: 1) Repeated viewing, which indicates the number of times a user has viewed the same clip multiple times at different times. This is obtained by the player monitoring each operation of playing the clip; 2) Skip count, which indicates the number of times a user has quickly skipped a clip. This is recorded by fast-forwarding or skipping behavior events; 3) Total viewing, which indicates the total number of times a user has watched the clip in its entirety. This is obtained by counting each time a user watches the clip from the beginning to the end. These data are used because they fully reflect the user's attention to the clip: Repeated viewing emphasizes the depth of the user's interest in the clip, skip count indicates that the clip is less attractive, and total viewing measures the user's overall interest in the clip. The combination of these data can more accurately measure the attractiveness of the clip and reflect the user's actual behavior pattern through the clip attractiveness index.

[0090] Calculate the segment attractiveness index of each video segment. The specific calculation formula is as follows:

[0091]

[0092] In the formula, SAI i is the segment attractiveness index of the i-th video segment.

[0093] The calculation formula of the Segment Attractiveness Index (SAI) comprehensively evaluates the attractiveness of a video segment through multiple user behavior data. First, The user's repeated viewing behavior is emphasized. The square process increases the weight of repeated viewing on the attractiveness of the clip, indicating that the user pays high attention to the clip. i and Represents the total number of views and the number of skips, respectively. The number of skips is expressed as The calculation is performed in the form of a loop to ensure that the negative impact of skipping behavior is taken into account and to avoid affecting the calculation results when the number of skips is zero. The purpose of this calculation is to balance the user's repeated viewing behavior and skipping behavior, and to ensure that the segment attractiveness index can reasonably reflect the user's real interest in the segment. Finally, the user's behavior data at multiple times is averaged to eliminate fluctuations in the data, making the calculation results more stable and accurately reflecting the segment attractiveness performance at different time points.

[0094] The segment attractiveness index SAI of the i-th video segment i The size of directly reflects the user's attention to the video clip. The larger the clip attractiveness index, the more times the user repeatedly watches the clip (i.e., has a higher interest in the clip) and the fewer times the user skips, indicating that the user has a stronger overall interest in the clip. In other words, a higher SAI i A value of 0 indicates that the segment is more attractive to users, and users are willing to watch it multiple times and maintain a low skip rate. i The larger the value, the more attention the user pays to the segment. By comparing the SAI of each segment i The value can reasonably evaluate the user's attention to different video clips.

[0095] In this embodiment, the viewing engagement coefficient EIC of each video clip is generated. i and Segment Attractiveness Index SAI i Construct an attention evaluation model and generate the attention coefficient GZ of each video clip by weighted summation i , and the generated attention coefficients GZ for each video clip i and the pre-set attention coefficient threshold interval GZ min , GZ max ] to compare, evaluate the user's attention to each video segment in the short video based on the comparison results, and divide each video segment into high-attention segments, medium-attention segments and low-attention segments based on the evaluation results. The specific comparison analysis and division are as follows:

[0096] If GZ i <GZ min , the user's attention to the i-th video segment in the short video is low, then the video segment is classified as a low-attention segment;

[0097] This means that users are less interested in the clip, neither spending a lot of time watching it nor interacting with it. This usually indicates that the content of the clip is not attractive enough to users, and there may be lengthy, irrelevant or undesirable parts. The impact is that the clip may be automatically cropped or shortened by the system to reduce unnecessary content and optimize the overall video rhythm to improve the user's viewing experience.

[0098] If GZ min ≤GZ i ≤GZ max , the user's attention to the i-th video segment in the short video is medium, then the video segment is classified as a medium attention segment;

[0099] This indicates that the user's attention to the segment is at a moderate level. The user may have watched most of the content, but there was no particularly frequent interaction or repeated viewing. This usually indicates that the segment has some interest in the user, but is not attractive enough. The impact is that the segment may be retained, but it will not be extended or highlighted. The system may make slight adjustments to the segment or optimize the transition to maintain the coherence of the video content.

[0100] If GZ i >GZ max , if the user's attention to the i-th video segment in the short video is high, then the video segment is classified as a high-attention segment.

[0101] This means that the user has a very high interest in the segment, which is usually accompanied by a longer viewing time and more frequent interactions or repeated viewing behaviors. This shows that the segment is the part that the user pays special attention to and has a strong appeal. The impact is that the system will give priority to retaining the segment, and may even extend the playback time or perform special processing on it (such as enhancing clarity or highlighting presentation) to meet the user's high attention needs, thereby improving user satisfaction and experience.

[0102] To build an attention evaluation model, we first need to calculate the viewing engagement coefficient EIC of each video clip. i and Segment Attractiveness Index SAI i Combined, the attention coefficient GZ is generated by weighted summation i The specific implementation method is that the system linearly weights these two coefficients for each video clip, and the formula is as follows: GZ i =ω1*EIC i +ω2*SAI i; Among them, ω1 and ω2 are two weight coefficients, which respectively reflect the relative importance of viewing engagement and segment attractiveness in the user's attention level. Usually, the setting of weight coefficients can be determined through historical data analysis: if users' viewing behavior (such as viewing time and dwell time) can better reflect their attention, then the weight ω1 of the viewing engagement coefficient can be set higher; if interactive behavior (such as repeated viewing and skipping behavior) is more important for evaluating attention, then the weight ω2 of the segment attractiveness index will be higher. By weighted summation, the generated attention coefficient can comprehensively consider users' viewing behavior and interactive behavior, and fully reflect the actual attention level of users to each segment.

[0103] According to the results of the division of each video segment, a content cropping and adjustment mechanism is constructed and applied to dynamically adjust the content of each video segment with different attention levels;

[0104] In this embodiment, according to the division results of each video segment, a content cropping adjustment mechanism is constructed and applied, specifically: according to the division results of high-attention segments, medium-attention segments and low-attention segments, different content cropping adjustment parameters are set respectively to form a content cropping adjustment mechanism; the content cropping adjustment mechanism is based on the attention coefficient of each video segment and the current segment length, and according to pre-set rules, automatically determines the content length and method of extension, retention and cropping;

[0105] The method of realizing "building and applying a content cropping adjustment mechanism based on the results of the division of each video segment" can be accomplished through algorithms and rule engines. The specific implementation method is that the system first generates the attention coefficient of each video segment through the user's viewing behavior data (based on the weighted sum of viewing participation and segment attractiveness index). Subsequently, the system classifies each segment according to the preset high attention, medium attention and low attention standards. For different categories of segments, the system will preset different cropping rules, which can be configured through the rule engine. For example, the rule for high-attention segments is "extend the segment length", the rule for medium-attention segments is "keep the length unchanged", and the rule for low-attention segments is "crop redundant content". Through the rule engine, the system can automatically adjust the length and content of video segments. The specific method is to adjust the timeline and screen cutting logic of the video, and crop or retain the content of different segments according to the preset standards. The reason for doing this is to optimize the video content according to the user's actual viewing preferences, ensure that the segments that the user pays attention to are highlighted, and the redundant segments are cropped, thereby improving the user's viewing experience and maintaining the compactness and rhythm of the video. This dynamic adjustment method can effectively respond to changes in user interests and ensure that the video content is more personalized.

[0106] For high-attention segments with a high degree of attention, the extension parameter in the content cutting adjustment mechanism is used to extend the playback time of the high-attention segments;

[0107] For high-attention clips, the playback time is extended because users show strong interest in these clips and may watch them multiple times or interact frequently. In order to meet the needs of users and enhance their viewing experience, the system can give users more time and content by extending the clips. This can be achieved by dynamically adjusting the timeline of the video. After the system identifies high-attention clips, it can extract more relevant clips from the material library or increase the details of the clip, such as extending the dialogue, adding background information, etc. This process can be set through the rule engine. When the attention coefficient of a clip exceeds the set threshold, the extension logic is automatically triggered to extend the length of the clip to the preset length to enhance user satisfaction with the content.

[0108] For the medium-attention segments with a medium attention level, the retention parameters in the content cropping adjustment mechanism are used to keep the current content unchanged;

[0109] For medium-interest clips, the content is kept unchanged because users have moderate interest in these clips. Although they are interested in watching them, they do not have particularly strong repeated viewing or interactive behaviors. Keeping the content unchanged can ensure the smoothness of the video and not disrupt the overall rhythm. The specific implementation method is that after analyzing the attention coefficient, the system identifies medium-interest clips and automatically triggers the settings for retaining clips in the rule engine. The video editing system does not need to add or delete these clips and directly retains them in the original timeline. This processing method ensures the continuity of the video, while also avoiding unnecessary processing of clips that users are generally interested in, and maintaining the original structure of the video.

[0110] For low-attention segments with a low degree of attention, the reduction parameters in the content cropping adjustment mechanism are used to crop redundant content and shorten the length of the low-attention segments.

[0111] For low-attention segments, redundant content is trimmed and the length is shortened because these segments fail to arouse enough interest from users and may be frequently skipped or watched for a very short time. Trimming these segments can improve the compactness of the video and prevent users from feeling dragged out or lengthy. The specific implementation method is that the system analyzes the attention coefficient of low-attention segments, triggers the trimming mechanism, and uses the video editing algorithm to remove lengthy, repetitive or irrelevant content from the segments. The system can automatically identify invalid images or redundant details in the segments through timeline-based editing tools, and shorten the length of the segments according to preset rules. Doing so can make the video more focused on the content that users are interested in, reduce irrelevant parts, and thus improve the user's overall viewing experience.

[0112] During the application of the content cropping adjustment mechanism, the cropping information of each video clip is obtained in real time, and analyzed after acquisition to evaluate whether the cropping degree of each video clip meets the preset content optimization standard, and the content cropping adjustment mechanism is optimized according to the evaluation result;

[0113] In this embodiment, during the application of the content cropping adjustment mechanism, cropping information of each video segment is obtained in real time, and analyzed after acquisition to evaluate whether the cropping degree of each video segment meets the preset content optimization standard, and the content cropping adjustment mechanism is optimized according to the evaluation result, which specifically includes the following steps:

[0114] In the process of applying the content cropping adjustment mechanism, cropping information of each video clip is obtained in real time and preprocessed after being obtained;

[0115] Real-time acquisition of cropping information of each video segment can be achieved through the real-time monitoring module embedded in the video player, specifically by monitoring and recording user operation behaviors (such as skipping, repeated viewing, pausing, fast forwarding, etc.), as well as the change data generated after the video segment is cropped (such as the cropped duration, content retention ratio, etc.). The system can use event listeners to capture these behaviors and send the data to the server for processing. The significance of real-time acquisition is that it can instantly collect user feedback on each segment and the effect after cropping, providing data support for subsequent optimization. The necessity of preprocessing is that the original data may contain noise or redundant information (such as repeated records, short pauses), so the data needs to be cleaned, deduplicated and standardized. The specific preprocessing steps include: filtering invalid behavior data (such as short stays or invalid clicks), formatting and standardizing the data (such as unifying the duration unit and converting to standardized values), and ensuring that the data input into the model is clean and consistent. The preprocessed data will provide an accurate basis for the analysis and evaluation of the cropping model, ensuring that the system can make optimization judgments.

[0116] Extracting the cropping integrity information and the clipping smoothness information from the cropping information of each preprocessed video segment, and analyzing them after extraction to generate a cropping integrity coefficient and a clipping smoothness index for each video segment respectively;

[0117] The cropping completeness coefficient and clip smoothness index of each generated video clip are used to construct a cropping degree evaluation model, and the cropping coefficient of each video clip is generated respectively. The generated cropping coefficient of each video clip is compared with the pre-set cropping coefficient threshold of each video clip. According to the comparison result, it is evaluated whether the cropping degree of each video clip meets the preset content optimization standard, and the content cropping adjustment mechanism is optimized according to the evaluation result.

[0118] The preset cropping factor thresholds for each video clip can be determined through data analysis and machine learning technology, and the specific method is to set it based on a large amount of user viewing behavior data and cropping feedback results. First, the system can collect the cropping effect data of multiple users when watching different video clips, and analyze the relationship between the cropping integrity coefficient and editing smoothness index of each clip and user feedback (such as viewing time, skip frequency, number of repeated viewings, etc.). Through cluster analysis or regression analysis, the system can identify the cropping factor range corresponding to the efficiently cropped clips. Then, use these historical data to train a machine learning model (such as a random forest or support vector machine), which can dynamically generate reasonable cropping factor thresholds based on the user's behavioral response to different clips. In this data-driven approach, the system can automatically set the cropping thresholds for different types of clips to ensure that the cropping effect of each clip meets the optimization criteria.

[0119] In this embodiment, the logic for obtaining the cropping integrity coefficient and the clipping smoothness index of each video segment is as follows:

[0120] The cropping integrity information of each preprocessed video clip is extracted, including the proportion of the retained content after cropping of each video clip in the short video, the content change at different moments in the process of cropping each video clip in the short video, and the change in the total length of the clip, and they are marked as CR i , and CR i Indicates the proportion of the content retained after cropping the i-th video clip in the short video. It represents the content change at time n during the process of cropping the i-th video segment in the short video. represents the change in the total length of the i-th video segment in the short video at time n during the cropping process, i=1, 2, 3, ..., k, n=1, 2, 3, ..., h, k and h are both positive integers;

[0121] The cropping integrity information can be extracted by embedding a real-time tracking module of the cropping process in the video editing system. The specific implementation method is that when the system is cropping video clips, it automatically records the proportion of retained content, the amount of content change, and the change in the total length of the clip for each clip. The proportion of retained content can be compared by comparing the total length of the clip before and after cropping, and real-time tracking the ratio of the content length retained after cropping to the original length; the amount of content change can be recorded by the system at each moment during the cropping process, and the addition and deletion of content at different time points can be marked through timestamps and video clip log files; the change in the total length of the clip can be automatically calculated by comparing the total length of the clip before and after cropping. All of these data can be recorded and uploaded in real time through the system's log files or backend databases.

[0122] Specific quantitative data include: Retained content ratio: This data reflects the ratio of the content retained after cropping to the content before cropping. The system can automatically calculate the proportion of retained content by comparing the number of key frames or text data of the video before and after cropping. This type of data is used to measure the degree of content loss during the cropping process. Content change: The content change is used to measure the difference between the content before and after cropping. The system calculates the content change value by comparing the video clips at each time point before and after cropping. This data reflects the impact of the cropping operation on the coherence of the content. Total clip length change: The length change is obtained by comparing the total length difference before and after the clip is cropped. The system monitors the cropping operation through the timeline and records the length change, which is used to evaluate the impact of cropping on the overall playback rhythm of the clip. The reason for using these data is that they fully reflect the retention and adjustment of the video content during the cropping process: the retained content ratio can reflect the degree of retention of the core content of the clip after cropping, the content change quantifies the adjustment of the content at each moment in the cropping, and the total length change measures the change trend of the overall content. These quantitative data combined can provide a comprehensive basis for the calculation of the cutting completeness coefficient, ensuring that the matching degree of cutting effect and content optimization standards can be accurately evaluated.

[0123] Calculate the cropping integrity coefficient of each video segment. The specific calculation formula is as follows:

[0124]

[0125] In the formula, CIC i is the cropping completeness coefficient of the i-th video segment;

[0126] The cutting integrity coefficient CIC i The calculation formula of CR comprehensively evaluates the integrity of the video clip after cropping through multiple key parameters to ensure that the cropped content maintains relative quality and consistency. First, the CR in the formula i It indicates the proportion of content retained after cropping, which reflects the degree of retention of the core content of the fragment. The higher the retention ratio, the stronger the integrity of the content. Then, It is used to reflect the amount of change in content, indicating the degree of content reduction during the cropping process. By considering the inverse vector of content change, it ensures that the clips that retain more core content during the cropping process obtain higher integrity scores. It represents the change in the length of the clip, which measures the change in the length of the clip before and after cutting, ensuring that the overall cutting evaluation can reflect the impact of such change when the length changes significantly. By weighted summing and averaging the content retention ratio, content change, and length change at each moment, the system can comprehensively evaluate the cutting integrity of the clip and avoid the excessive impact of abnormal data at a certain moment on the overall result. This calculation method ensures that the cutting process can accurately evaluate the content retention and quality consistency of the clip after cutting, making the results more comprehensive and accurate.

[0127] The cropping completeness coefficient CIC of the i-th video segment i The size of directly reflects the degree of match between the content retention of the clip after cutting and the optimization standard. The larger the cutting integrity coefficient, the more core content is retained in the clip after cutting, the smaller the content change, and the moderate adjustment of the duration, so that the preset content optimization standard can be better met. In other words, CIC i The higher the value, the more consistent the trimmed clips are with the optimization criteria in terms of content quality, coherence, and overall integrity, and the more consistent and visually appealing the content is. i A lower value means that less content is retained after cropping and the changes are too large. It may not meet the preset optimization standards and the cropping method needs to be further adjusted.

[0128] Extract the clip smoothness information from the cropping information of each preprocessed video clip, specifically including the ratio of the transition time of each video clip in the total clip length after cropping each video clip in the short video, the change in playback speed at different moments in the process of cropping each video clip in the short video, and the number of cropping, and mark them as TTR respectively. i , and TTR i It represents the ratio of the transition time of the ith video segment in the total segment length after the video segments in the short video are cropped. Indicates the change in playback speed at time n during the process of cropping the i-th video segment in the short video. Indicates the number of times the i-th video segment in the short video is cropped at time n;

[0129] Clip smoothness information can be extracted by recording and analyzing the operation data of the cropping process in real time in the video editing system. The specific implementation method is that when the system is cropping the video, it automatically tracks the transition time, playback speed change and the number of cropping of each segment. The transition time ratio can be obtained by monitoring the ratio of the connection time between the clips after cropping to the total length of the clips; the system captures the playback timestamp and switching points during the video playback process to calculate the transition time. The playback speed change can be obtained by the system monitoring the change in the playback rate of the video clip, recording the frame rate or playback rate at each moment, and comparing the rate difference before and after cropping. The number of cropping is the number of editing operations automatically recorded by the system, and each editing operation on the video clip will be counted in the log file. These data are summarized and uploaded through real-time logs or background databases to ensure that the impact of cropping operations is accurately recorded.

[0130] Specific quantitative data include: 1) Transition time ratio, which is obtained by calculating the ratio of the transition time between each video segment to the total length of the segment. This data is used to measure the smoothness of segment switching; 2) Playback speed change, which is obtained by changing the playback rate (such as frame rate) of the segment before and after cropping at different times, reflecting the coherence of the video rhythm; 3) Number of cropping, the system automatically records the number of times the video is cropped each time, which is used to measure the impact of frequent adjustments of video segments on smoothness. These quantitative data are used because they can accurately measure the impact of cropping on video rhythm and coherence. The transition time ratio can reflect whether the switching between segments is smooth, the playback speed change can evaluate the stability of the video playback rhythm, and the number of cropping reflects the frequency of video content modification. Combining these data, the system can calculate an accurate editing smoothness index to ensure that the video still maintains a good look and feel and smoothness after cropping.

[0131] Calculate the clip smoothness index of each video clip. The specific calculation formula is as follows:

[0132]

[0133] Where, CSI i is the clip smoothness index of the i-th video clip.

[0134] The clip smoothness index CSI i The calculation formula comprehensively evaluates the smoothness of video clips by considering three core factors in the video editing process: editing frequency, transition time ratio, and playback speed change. First, the inverse of the editing frequency It is used to measure the frequency of editing. Frequent editing will reduce the coherence of the video, so the higher the frequency, the lower the smoothness index. Secondly, the transition time ratio TTR iIt reflects the quality of the connection between clips. The more appropriate the time ratio is, the smoother the video is and the more natural the transition is. It is used to evaluate the change in playback speed before and after editing. The smaller the speed change, the more consistent the rhythm of the video and the higher the smoothness. By weighted averaging these factors (i.e. summing up the smoothness at different moments and taking the average), the overall smoothness index of each video clip can be obtained. The purpose of this is to ensure that the video maintains good fluidity and rhythm consistency after editing, and avoid poor viewing experience due to frequent editing or changes in playback speed.

[0135] The clip smoothness index CSI of the i-th video clip i The size of CSI directly reflects whether the clip after trimming maintains good fluency and rhythm consistency, thereby evaluating whether the trimming meets the preset content optimization standards. i The larger the value, the more reasonable the transition time of the clip, the smaller the change in playback speed, and the moderate number of clips, indicating that the video clip is still coherent and natural after trimming, meeting the smooth viewing experience required by the optimization standard. i A low value indicates that the video clips are not smoothly connected, the playback speed changes too much, or the editing is too frequent, which will affect the viewing experience of the video and require further optimization of the cropping method. Therefore, the size of the clip smoothness index is an important indicator to measure whether the cropping effect meets the optimization standard.

[0136] In this embodiment, the cropping integrity coefficient CIC of each video segment is generated i and Clip Smoothness Index CSI i Construct a cropping degree evaluation model and generate the cropping coefficient CJ of each video segment through weighted summation i , and the cropping coefficients CJ of each generated video segment i The cropping factor threshold of each video segment is set Compare and evaluate whether the cropping degree of each video clip meets the preset content optimization standard based on the comparison results, and optimize the content cropping adjustment mechanism based on the evaluation results. The specific comparison and analysis are as follows:

[0137] like The cropping degree of the video clip meets the preset content optimization standard, and there is no need to optimize the content cropping adjustment mechanism;

[0138] This means that the cropping integrity and editing smoothness of the video clip meet the established optimization standards. The video clip retains enough core content and maintains good playback smoothness, so no further optimization is needed. At this point, the cropping adjustment mechanism has reached the optimal state. The system can automatically confirm the cropping result of the clip and mark it as no need for further processing, and directly generate the final output. In this case, the system can shift its focus to the cropping adjustment of other video clips to ensure that all clips meet the same quality standards.

[0139] like The degree of cropping of the video clip does not meet the preset content optimization standards, and the content cropping adjustment mechanism needs to be optimized, specifically including: optimizing the retention time, transition effect, editing frequency and playback speed of the video clip by dynamically adjusting the weight ratio of the cropping integrity coefficient and the editing smoothness index.

[0140] This situation means that too much core content is lost, the transition between clips is unnatural, or the playback speed is inconsistent, which affects the user's viewing experience. At this time, the software system needs to dynamically adjust the weight ratio of the cropping integrity coefficient and the editing smoothness index. Specifically, the weight of the retained content ratio can be increased, the impact of the editing frequency can be reduced, and the playback speed and transition time can be optimized. By analyzing historical cropping data, the system can identify specific problem points (such as unnatural transitions, choppy playback, etc.), and then automatically recalculate the retention time, transition time and playback rate of each clip according to the preset optimization rules to ensure that the cropped results meet the optimization standards, thereby improving the viewing experience of the video.

[0141] The cropping integrity coefficient CIC of each generated video segment i and Clip Smoothness Index CSI i Combined with the construction of a cropping degree evaluation model, the specific approach is: first, different weight coefficients are assigned to the cropping completeness and editing smoothness of each video clip, reflecting the importance of these two indicators in the overall cropping effect. The weight coefficient can be dynamically adjusted according to the user's viewing preferences or content type. For example, in a video clip that focuses on content preservation, the cropping completeness coefficient CIC can be i It is given a greater weight, and in clips that emphasize playback smoothness, it is the clip smoothness index CSI. i Give a greater weight. Through the weighted summation calculation formula, the clipping coefficient CJ i It can be expressed as: CJ i =ω3*CIC i +ω4*CSI i Among them, ω3 and ω4 are the weight coefficients of clipping completeness and clipping smoothness, respectively, CIC i and CSI iis the cropping completeness coefficient and clip smoothness index of the i-th video clip. Through this weighted summation method, the generated cropping coefficient can fully reflect the cropping quality of the video clip and can be flexibly adjusted according to the characteristics of different clips to better meet the overall content optimization standard.

[0142] According to the optimized content cropping and adjustment mechanism, continuously monitor users' viewing feedback on the cropped video content, and continuously optimize the content cropping and adjustment mechanism based on these feedback data.

[0143] Continuous monitoring of user feedback on cropped video content can be achieved through a data collection module embedded in the video playback platform. Specifically, the system can record user viewing behavior data in real time, including but not limited to video playback duration, number of skipped clips, number of repeated viewings, likes, comments, and sharing and other interactive behaviors. After these data are collected by the front-end player, they are uploaded to the back-end server for aggregation and analysis. The reason for this is that the user's actual viewing behavior can most directly reflect the attractiveness and quality of the video content. Continuous monitoring of these behaviors can help the system understand whether the cropped video meets the user's expectations and provide data support for subsequent optimization.

[0144] Optimizing the cropping mechanism based on user feedback data can be achieved through data analysis algorithms and machine learning models. Specifically, the system will cluster or classify the collected user viewing behavior data to identify which clips have received more attention and which clips are frequently skipped or ignored. Through this data-driven analysis, the system can dynamically adjust the weight ratio of the cropping completeness coefficient and the editing smoothness index to better optimize the retention time and editing effect of video clips in future editing processes. The reason for doing this is that user feedback directly reflects the quality of the video. Continuously optimizing the cropping mechanism based on these feedbacks can improve the content quality of the video, making the subsequently generated videos more in line with the user's viewing habits and preferences.

[0145] Continuous optimization of the cropping mechanism can be achieved in an iterative manner. Specifically, after each video is generated, the system updates the weight coefficients and cropping rules in the cropping model based on the latest user feedback data. The system uses historical data to train the model and continuously improves the cropping algorithm so that it can adapt to different video types and user needs. The goal of continuous optimization is to make the cropping mechanism more and more intelligent, so that it can provide the best cropping effect in different scenarios, thereby improving the attractiveness of the video and user satisfaction. The reason for continuous optimization is that user preferences and the type of video content may change over time, and continuous optimization can ensure that the cropping mechanism can flexibly respond to such changes.

[0146] like Figure 2The AIGC short video production system based on user feedback tags shown in the figure includes a real-time viewing behavior collection and analysis module, a segment attention evaluation and classification module, a dynamic content cropping adjustment module, a cropping effect evaluation and mechanism optimization module, and a feedback-driven cropping mechanism optimization module;

[0147] The real-time viewing behavior collection and analysis module collects the user's viewing behavior information in real time while the user is watching the short video, and analyzes the user's viewing pattern based on this information. When the analysis result shows that the short video content needs to be trimmed, the short video to be trimmed is evenly divided into several video clips;

[0148] The segment attention evaluation and classification module obtains the segment attention information of users on each video segment in real time, analyzes it after acquisition, evaluates the user's attention to each video segment in the short video, and divides each video segment into high-attention segment, medium-attention segment and low-attention segment according to the evaluation results;

[0149] The dynamic content cropping and adjustment module builds and applies a content cropping and adjustment mechanism based on the division results of each video segment, and dynamically adjusts the content of each video segment with different attention levels;

[0150] The cropping effect evaluation and mechanism optimization module obtains the cropping information of each video clip in real time during the application of the content cropping adjustment mechanism, analyzes it after acquisition, evaluates whether the cropping degree of each video clip meets the preset content optimization standard, and optimizes the content cropping adjustment mechanism according to the evaluation results;

[0151] The feedback-driven cropping mechanism optimization module continuously monitors the user's viewing feedback on the cropped video content according to the optimized content cropping adjustment mechanism, and continuously optimizes the content cropping adjustment mechanism based on these feedback data.

[0152] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0153] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0154] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0155] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0156] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0157] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0158] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0159] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. The AIGC short video production method based on user feedback tags is characterized by: The specific steps include: When users watch short videos, the viewing behavior information of users is collected in real time, and the viewing patterns of users are analyzed based on the information. When the analysis results show that the short video content needs to be trimmed, the short video to be trimmed is evenly divided into several video clips. Acquire the user's attention information on each video segment in real time, analyze it after acquisition, evaluate the user's attention to each video segment in the short video, and classify each video segment into high attention segment, medium attention segment and low attention segment according to the evaluation result; According to the results of the division of each video segment, a content cropping and adjustment mechanism is constructed and applied to dynamically adjust the content of each video segment with different attention levels; During the application of the content cropping adjustment mechanism, the cropping information of each video clip is obtained in real time, and analyzed after acquisition to evaluate whether the cropping degree of each video clip meets the preset content optimization standard, and the content cropping adjustment mechanism is optimized according to the evaluation result; According to the optimized content cropping and adjustment mechanism, continuously monitor users' viewing feedback on the cropped video content, and continuously optimize the content cropping and adjustment mechanism based on these feedback data.

2. The AIGC short video production method based on user feedback tags according to claim 1 is characterized in that: The user's attention information on each video segment is obtained in real time, and analyzed after being obtained to evaluate the user's attention to each video segment in the short video, and each video segment is divided into a high-attention segment, a medium-attention segment, and a low-attention segment according to the evaluation result, specifically including the following steps: Acquire the user's attention information on each video segment in real time and perform preprocessing after acquisition; Extracting the pre-processed user's viewing behavior participation information and segment attractiveness behavior information from the segment attention information of each video segment, and analyzing them after extraction to generate the viewing participation coefficient and segment attractiveness index of each video segment respectively; The viewing engagement coefficient and segment attractiveness index of each generated video clip are used to construct an attention degree evaluation model to generate an attention coefficient for each video clip. The generated attention coefficient for each video clip is compared with a pre-set attention coefficient threshold range. The user's attention degree to each video clip in the short video is evaluated based on the comparison result, and each video clip is divided into high-attention clips, medium-attention clips and low-attention clips according to the evaluation results.

3. The AIGC short video production method based on user feedback tags according to claim 2 is characterized in that: The logic for obtaining the viewing participation coefficient and segment attractiveness index of each video segment is as follows: Extract the viewing behavior participation information of the pre-processed user attention information on each video clip, including the total duration of each video clip in the short video, the total viewing time of the user on each video clip in the short video, and the frequency of the user's interactive behavior on each video clip at different times within a period of time, and mark them as LSE respectively. i 、YWA i and LSE i Indicates the total duration of the i-th video clip in the short video, YWA i represents the total viewing time of the user on the i-th video segment in the short video, represents the frequency of user interaction on the i-th video clip at time m within a period of time, i = 1, 2, 3, ..., k, m = 1, 2, 3, ..., g, k and g are both positive integers; Calculate the viewing engagement coefficient of each video clip. The specific calculation formula is as follows: Where, EIC i is the viewing engagement coefficient of the i-th video clip; Extract the segment attraction behavior information from the pre-processed user segment attention information of each video segment, including the number of repeated views and skips of each video segment in the short video at different times within a period of time, and the total number of views of each video segment in the short video by the user, and mark them as and NTO i , It indicates the number of times the user repeatedly watches the i-th video clip in the short video at time m within a period of time. NTO represents the number of times a user skips the i-th video segment in the short video at time m within a period of time. i Indicates the total number of times the user has watched the i-th video clip in the short video; Calculate the segment attractiveness index of each video segment. The specific calculation formula is as follows: Where SAI i is the segment attractiveness index of the i-th video segment.

4. The AIGC short video production method based on user feedback tags according to claim 3 is characterized in that: The viewing engagement coefficient EIC of each video clip generated i and Segment Attractiveness Index SAI i Construct an attention evaluation model and generate the attention coefficient GZ of each video clip by weighted summation i , and the generated attention coefficients GZ for each video clip i and the pre-set attention coefficient threshold interval GZ min , GZ max ] to compare, evaluate the user's attention to each video segment in the short video based on the comparison results, and divide each video segment into high-attention segments, medium-attention segments and low-attention segments based on the evaluation results. The specific comparison analysis and division are as follows: If GZ i <GZ min , the user's attention to the i-th video segment in the short video is low, then the video segment is classified as a low-attention segment; If GZ min ≤GZ i ≤GZ max , the user's attention to the i-th video segment in the short video is medium, then the video segment is classified as a medium attention segment; If GZ i >GZ max , if the user's attention to the i-th video segment in the short video is high, then the video segment is classified as a high-attention segment.

5. The AIGC short video production method based on user feedback tags according to claim 4 is characterized in that: According to the results of the division of each video segment, a content cropping adjustment mechanism is constructed and applied, specifically: according to the division results of high-attention segments, medium-attention segments and low-attention segments, different content cropping adjustment parameters are set respectively to form a content cropping adjustment mechanism; the content cropping adjustment mechanism is based on the attention coefficient of each video segment and the current segment length, and through pre-set rules, automatically determines the content length and method of extension, retention and cropping; For high-attention segments with a high degree of attention, the extension parameter in the content cutting adjustment mechanism is used to extend the playback time of the high-attention segments; For the medium-attention segments with a medium attention level, the retention parameters in the content cropping adjustment mechanism are used to keep the current content unchanged; For low-attention segments with a low degree of attention, the reduction parameters in the content cropping adjustment mechanism are used to crop redundant content and shorten the length of the low-attention segments.

6. The AIGC short video production method based on user feedback tags according to claim 5 is characterized in that: In the process of applying the content cropping adjustment mechanism, the cropping information of each video clip is obtained in real time, and analyzed after acquisition to evaluate whether the cropping degree of each video clip meets the preset content optimization standard, and the content cropping adjustment mechanism is optimized according to the evaluation result, which specifically includes the following steps: In the process of applying the content cropping adjustment mechanism, cropping information of each video clip is obtained in real time and preprocessed after being obtained; Extracting the cropping integrity information and the clipping smoothness information from the cropping information of each preprocessed video segment, and analyzing them after extraction to generate a cropping integrity coefficient and a clipping smoothness index for each video segment respectively; The cropping completeness coefficient and clip smoothness index of each generated video clip are used to construct a cropping degree evaluation model, and the cropping coefficient of each video clip is generated respectively. The generated cropping coefficient of each video clip is compared with the pre-set cropping coefficient threshold of each video clip. According to the comparison result, it is evaluated whether the cropping degree of each video clip meets the preset content optimization standard, and the content cropping adjustment mechanism is optimized according to the evaluation result.

7. The AIGC short video production method based on user feedback tags according to claim 6 is characterized in that: The logic for obtaining the cropping integrity coefficient and the clipping smoothness index of each video segment is as follows: The cropping integrity information of each preprocessed video clip is extracted, including the proportion of the retained content after cropping of each video clip in the short video, the content change at different moments in the process of cropping each video clip in the short video, and the change in the total length of the clip, and they are marked as CR i , and CR i Indicates the proportion of the content retained after cropping the i-th video clip in the short video. It represents the content change at time n during the process of cropping the i-th video segment in the short video. represents the change in the total length of the i-th video segment in the short video at time n during the cropping process, i=1, 2, 3, ..., k, n=1, 2, 3, ..., h, k and h are both positive integers; Calculate the cropping integrity coefficient of each video segment. The specific calculation formula is as follows: In the formula, CIC i is the cropping completeness coefficient of the i-th video segment; Extract the clip smoothness information from the cropping information of each preprocessed video clip, specifically including the ratio of the transition time of each video clip in the total clip length after cropping each video clip in the short video, the change in playback speed at different moments in the process of cropping each video clip in the short video, and the number of cropping, and mark them as TTR respectively. i , and TTR i It represents the ratio of the transition time of the ith video segment in the total segment length after the video segments in the short video are cropped. Indicates the change in playback speed at time n during the process of cropping the i-th video segment in the short video. Indicates the number of times the i-th video segment in the short video is cropped at time n; Calculate the clip smoothness index of each video clip. The specific calculation formula is as follows: Where, CSI i is the clip smoothness index of the i-th video clip.

8. The AIGC short video production method based on user feedback tags according to claim 7 is characterized in that: The cropping integrity coefficient CIC of each generated video segment i and Clip Smoothness Index CSI i Construct a cropping degree evaluation model and generate the cropping coefficient CJ of each video segment through weighted summation i , and the cropping coefficients CJ of each generated video segment i The cropping factor threshold of each video segment is set Compare and evaluate whether the cropping degree of each video clip meets the preset content optimization standard based on the comparison results, and optimize the content cropping adjustment mechanism based on the evaluation results. The specific comparison and analysis are as follows: like The cropping degree of the video clip meets the preset content optimization standard, and there is no need to optimize the content cropping adjustment mechanism; like The degree of cropping of the video clip does not meet the preset content optimization standards, and the content cropping adjustment mechanism needs to be optimized, specifically including: optimizing the retention time, transition effect, editing frequency and playback speed of the video clip by dynamically adjusting the weight ratio of the cropping integrity coefficient and the editing smoothness index.

9. An AIGC short video production system based on user feedback tags, used to implement the AIGC short video production method based on user feedback tags described in any one of claims 1 to 8, characterized in that: It includes real-time viewing behavior collection and analysis module, segment attention evaluation and classification module, dynamic content cutting and adjustment module, cutting effect evaluation and mechanism optimization module and feedback-driven cutting mechanism optimization module; The real-time viewing behavior collection and analysis module collects the user's viewing behavior information in real time while the user is watching the short video, and analyzes the user's viewing pattern based on this information. When the analysis result shows that the short video content needs to be trimmed, the short video to be trimmed is evenly divided into several video clips; The segment attention evaluation and classification module obtains the segment attention information of users on each video segment in real time, analyzes it after acquisition, evaluates the user's attention to each video segment in the short video, and divides each video segment into high-attention segment, medium-attention segment and low-attention segment according to the evaluation results; The dynamic content cropping and adjustment module builds and applies a content cropping and adjustment mechanism based on the division results of each video segment, and dynamically adjusts the content of each video segment with different attention levels; The cropping effect evaluation and mechanism optimization module obtains the cropping information of each video clip in real time during the application of the content cropping adjustment mechanism, analyzes it after acquisition, evaluates whether the cropping degree of each video clip meets the preset content optimization standard, and optimizes the content cropping adjustment mechanism according to the evaluation results; The feedback-driven cropping mechanism optimization module continuously monitors the user's viewing feedback on the cropped video content according to the optimized content cropping adjustment mechanism, and continuously optimizes the content cropping adjustment mechanism based on these feedback data.

Images