AI-based trend analysis method and system for popular video

Abstract

The invention relates to the field of data analysis, in particular to an AI-based trend analysis method and system for an explosive video, and the method comprises the steps: carrying out the collection and preprocessing of potential trend cluster video popularity data, and obtaining a structured time series data set; comparing the comprehensive strength of the trend cluster maximum contribution video group and the comprehensive strength of the trend cluster non-maximum contribution video group to obtain a contribution structure counterbalance factor; accumulating and comparing the contribution differences of all videos in the trend cluster in pairs to obtain an internal contribution difference factor; correcting by sequentially combining the original propagation score with a contribution structure counterbalance factor and an internal contribution difference factor to obtain a final trend score; and the final trend scores of all the potential trend clusters are sorted to obtain a trend cluster identification result with an implosion potential, so that the problem of trend propagation potential misjudgment caused by total amount aggregation in the prior art is solved.

Description

A trend analysis method and system for viral videos based on AI Technical Field

[0001] This invention relates to the field of data analysis, and in particular to an AI-based method and system for trend analysis of viral videos. Background Technology

[0002] In internet platforms where short videos are the primary content format, accurately identifying and predicting trends with the potential to become viral hits from massive amounts of video data is a core technology supporting content creation, platform operation, and commercial promotion. To this end, existing technologies typically employ a data-driven analysis method, which extracts multimodal features from videos (visual, audio, and textual) and uses clustering algorithms to group videos with similar attributes into potential trend clusters. These potential trend clusters generally consist of multiple videos using the same background music, participating in the same challenges, or imitating the same actions, and can reflect, to some extent, the potential direction of content dissemination.

[0003] In trend assessment, existing technologies generally use the propagation coefficient as the core indicator. This propagation coefficient is typically calculated by comparing the total new views of a trend cluster within a preset time window with the cumulative views at the start of the time window, thus obtaining the relative growth rate of the trend. In existing methods, trend clusters with higher propagation coefficients are often considered to have greater potential and are more likely to develop into viral trends. However, because this method only relies on the overall new and cumulative views of the trend cluster, the contribution distribution of each video within the cluster is completely aggregated into a single scalar. This aggregation method leads to the loss of key structural information. Specifically, if the new views of a trend cluster mainly come from a single video, its growth pattern is often centralized, and the trend relies on the video creator's fan base or platform recommendations, lacking replicability and sustainability; while another trend cluster, even with the same total new views, has its popularity increase distributed among a large number of videos, indicating that the trend content has low barriers to entry and high diffusion potential, and is more self-sustaining. Under existing methods, these two distinct trends are assigned similar propagation potential scores, leading to misjudgments and failing to meet the requirements of practical applications for trend identification accuracy and reliability. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide an AI-based trend analysis method for viral videos to solve the problem of misjudging the potential for trend propagation caused by aggregated total data in the prior art.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0006] Firstly, this application provides an AI-based trend analysis method for viral videos, which includes the following steps:

[0007] Step S1: Obtain a structured time series dataset by collecting and preprocessing video popularity data of potential trend clusters;

[0008] Step S2: By comparing the overall strength of the trend cluster's largest contributing video with that of the non-largest contributing video group, obtain the contribution structure balancing factor;

[0009] Step S3: Obtain the internal contribution difference factor by summing and comparing the pairwise contribution differences of all videos within the trend cluster;

[0010] Step S4: The final trend score is obtained by successively correcting the original propagation score by combining the contribution structure check factor and the internal contribution difference factor.

[0011] Step S5: By sorting the final trend scores of all potential trend clusters, the identification results of trend clusters with blockbuster potential are obtained.

[0012] Preferably, the step of acquiring a structured time-series dataset by collecting and preprocessing video popularity data of potential trend clusters includes:

[0013] Clustering methods are used to cluster videos based on their multimodal features to obtain potential trend clusters.

[0014] Set the time window length; collect the identification information and corresponding cumulative playback data of each video in the potential trend cluster at the start and end times within the time window to obtain a video popularity dataset containing video unique identifiers, start cumulative playback data and end cumulative playback data; format and organize the collected video popularity dataset, establish an index with the video unique identifier as the primary key, and obtain a structured time series dataset storing the playback changes of each video within the time window.

[0015] Preferably, the step of obtaining the contribution structure balancing factor by comparing the comprehensive strength of the trend cluster's largest contributing video group with that of the non-largest contributing video group includes:

[0016] By statistically processing the new play count data of each video within the potential trend cluster, the new play count of the video with the largest contribution is obtained. By summarizing and analyzing the new play count data of the videos with the smaller contribution within the potential trend cluster, the comprehensive strength benchmark of the group with the smaller contribution is obtained. By calculating the ratio of the new play count of the video with the largest contribution to the comprehensive strength benchmark of the group with the smaller contribution and processing it with a nonlinear function, the contribution structure balancing factor is obtained.

[0017] Preferably, the step of statistically processing the new play count data of each video within the trend cluster to obtain the new play count of the video with the largest contribution, and summarizing and analyzing the new play count data of the videos with the smaller contribution within the trend cluster to obtain the comprehensive strength benchmark of the group with the smaller contribution, includes:

[0018] Set a time window, and iterate through and count the new play count data of all videos in the potential trend cluster within the time window to obtain the target video with the largest new play count data, and take the new play count of the target video as the new play count of the video with the largest contribution.

[0019] The summation of new play counts for all non-maximum contributing videos (excluding the maximum contributing video) in the potential trend cluster is used to obtain the total new play count for the non-maximum contributing group. The average new play count for the non-maximum contributing group is calculated by dividing the total new play count of the non-maximum contributing group by the number of non-maximum contributing videos.

[0020] The comprehensive strength benchmark of the non-maximum contributing group is obtained by summing the total new play count of the non-maximum contributing group with the average new play count of the non-maximum contributing group.

[0021] Preferably, the step of obtaining the contribution structure balancing factor by calculating the ratio of the new views of the video with the largest contribution to the comprehensive strength benchmark of the non-largest contribution group and processing it with a nonlinear function includes:

[0022] The numerator is the number of new views of the video with the largest contribution, and the denominator is the comprehensive strength benchmark of the non-largest contributing group. The resulting fraction is used as the ratio of the largest contribution. The ratio of the largest contribution is mapped by a hyperbolic tangent function to obtain the corresponding centralization penalty intensity. The result of subtracting the centralization penalty intensity from the constant 1 is used as the contribution structure balancing factor.

[0023] Preferably, the step of obtaining the internal contribution difference factor by summing and comparing the pairwise contribution differences of all videos within the trend cluster includes:

[0024] By performing pairwise difference calculations on the new play count data of all videos within the trend cluster, the sum of pairwise contribution differences is obtained. Furthermore, by deriving the theoretical extreme values ​​of the number of videos within the trend cluster and the total new play count data, the maximum possible difference value is obtained.

[0025] The internal contribution difference factor is obtained by normalizing and comparing the sum of paired contribution differences with the maximum possible difference value.

[0026] Preferably, the step of obtaining the sum of pairwise contribution differences by performing pairwise difference calculations on the new play count data of all videos within the trend cluster, and obtaining the maximum possible difference value by theoretically deriving the extreme values ​​of the number of videos within the trend cluster and the total new play count data, includes:

[0027] The contribution difference value of any two different videos in the potential trend cluster is obtained by performing a difference operation on the new play count data and taking the absolute value; the contribution difference value of all non-repeating video pairs is accumulated to obtain the sum of pairwise contribution differences; the maximum possible difference value is obtained by multiplying the result of subtracting a constant 1 from the number of videos in the potential trend cluster with the total new play count data in the potential trend cluster.

[0028] Preferably, the step of obtaining the internal contribution difference factor by normalizing and comparing the sum of paired contribution differences with the maximum possible difference value includes:

[0029] For any potential trend cluster, the sum of paired contribution differences is used as the numerator, the maximum possible difference value is used as the denominator, and the resulting fraction is used as the imbalance index of the potential trend cluster; the result of subtracting the imbalance index of the potential trend cluster from the constant 1 is used as the internal contribution difference factor of the potential trend cluster.

[0030] Preferably, the step of correcting the original propagation score by sequentially combining the contribution structure check factor and the internal contribution difference factor to obtain the final trend score includes:

[0031] Set a time window. For any potential trend cluster, use the cumulative total number of plays in the potential trend cluster at the beginning of the time window as the denominator, and use the total number of new plays in the potential trend cluster within the time window as the numerator. Use the resulting fraction as the original propagation score of the potential trend cluster.

[0032] The first propagation score after structural correction is obtained by multiplying the original propagation score with the contribution structure balancing factor; the final trend score is obtained by multiplying the first propagation score with the internal contribution difference factor.

[0033] Secondly, this application provides an AI-based trend analysis system for viral videos, comprising: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an AI-based trend analysis method for viral videos is implemented.

[0034] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0035] This invention introduces a contribution structure balancing factor and an internal contribution difference factor into the trend cluster propagation potential assessment process. This ensures that the assessment results not only rely on the absolute increase in views but also reflect the contribution distribution characteristics among different videos within the cluster. In practical short video platform scenarios, this invention can effectively identify centralized trends driven by only a single or a few videos, thus preventing their rapid decline due to top creators ceasing updates or the platform's traffic recommendation ending. Simultaneously, for distributed trends formed by a large number of ordinary creators, this invention can accurately capture their broad audience base and sustained diffusion potential, significantly improving the ability to identify truly viral trends. Technically, this invention overcomes the limitations of existing methods that lose structural information due to total aggregation by quantitatively modeling the group strength, balance, and difference of contributions within a cluster. Through a dual-factor joint correction mechanism, this invention not only improves the interpretability of trend scores but also enhances the adaptability of the assessment model under different scales and diffusion patterns. Ultimately, this invention can provide more reliable technical support for short video platforms in content distribution, hot topic recommendation, and business operation decisions, reducing resource waste caused by misjudgments and improving the health and sustainability of the overall propagation ecosystem. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 is a flowchart of a trend analysis method for trending viral videos based on AI, provided in Embodiment 1 of the present invention. Detailed Implementation

[0038] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0039] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0040] Referring to Figure 1, which is a flowchart of a trend analysis method for trending viral videos based on AI according to Embodiment 1 of the present invention, as shown in Figure 1, the method may include:

[0041] Step S1: Obtain a structured time series dataset by collecting and preprocessing the video popularity data of potential trend clusters.

[0042] This step aims to provide the necessary basic data for subsequent propagation score calculation and correction. Specifically, it first uses existing clustering methods to cluster the video based on its multimodal features to obtain the video's potential trend clusters.

[0043] The time window length is set. In this embodiment of the invention, the time window length is set to 6 hours. The length of the time window can be adjusted according to the actual scenario and is not required. By collecting the identification information and corresponding cumulative playback data of each video in the potential trend cluster at the start and end times respectively within the time window, a video popularity dataset containing video unique identifiers, start cumulative playback data and end cumulative playback data is obtained. By formatting and organizing the collected video popularity dataset, an index is established with the video unique identifier as the primary key to obtain a structured time series dataset that stores the playback changes of each video within the time window.

[0044] This completes the process of collecting and preprocessing video popularity data from potential trend clusters to obtain a structured time series dataset.

[0045] Step S2: By comparing the overall strength of the trend cluster's largest contributing video with that of the non-largest contributing video group, the contribution structure balancing factor is obtained.

[0046] In existing technologies, information loss makes it impossible to identify centralized growth patterns dominated by a single video. When the majority of the increased popularity of a potential trend cluster originates from a single video within the cluster, the strength of the trend depends entirely on this single video, and its subsequent spread and fermentation capabilities are uncertain. In this case, the high propagation coefficient actually reflects the influence of the video publisher or accidental traffic recommendations from the platform, rather than the widespread potential for the trend content itself to be widely replicated. Therefore, to address this issue, the first step in correction is to quantify this degree of centralization. An effective quantification method should not only consider the proportion of the largest contributing video but also analyze the power balance between the largest contributor and all other participants (i.e., the non-largest contributor group). A trend with distributed growth potential should have a group of non-largest contributors possessing considerable collective strength. This strength can be considered from two dimensions: first, the total collective contribution of the group, i.e., the sum of views contributed by all non-largest contributors, which represents the breadth and base of participation in the trend; second, the average contribution level of the group, i.e., the ratio of the total collective contribution to the number of group members, which represents the average quality of participation of ordinary participants. Based on this logic, this step constructs a benchmark that can simultaneously measure the overall strength of the non-maximum contributor group across both dimensions, and compares the strength of the maximum contributor with this benchmark to generate a correction factor. When the strength of the maximum contributor far exceeds the overall strength of the non-maximum contributor group, it indicates that the trend exhibits a strong centralization characteristic, and the correction factor will approach 0, significantly suppressing the original propagation score. Conversely, if the non-maximum contributor group demonstrates strong collective strength and a high average participation quality, capable of balancing the maximum contributor, it indicates that the trend structure is relatively healthy, and the correction factor will approach 1, preserving or only slightly adjusting the original propagation score.

[0047] In summary, this step first statistically processes the new play count data of each video within the potential trend cluster to obtain the new play count of the video with the largest contribution. Then, it summarizes and analyzes the new play count data of the non-largest contributing videos within the potential trend cluster to obtain the comprehensive strength benchmark of the non-largest contributing group. Specifically, by traversing and statistically analyzing the new play count data of all videos in the potential trend cluster within the time window, the target video with the largest new play count is identified and its new play count is taken as the new play count of the largest contributing video. Next, the new play count data of all non-largest contributing videos in the potential trend cluster (excluding the largest contributing video) are summed to obtain the total new play count of the non-largest contributing group. The result of dividing the total new play count of the non-largest contributing group by the number of non-largest contributing videos is taken as the average new play count of the non-largest contributing group. Finally, the comprehensive strength benchmark of the non-largest contributing group is obtained by summing the total new play count of the non-largest contributing group and the average new play count of the non-largest contributing group.

[0048] After obtaining the comprehensive strength benchmark of the non-maximum contributing group, the contribution structure balancing factor is obtained by calculating the ratio of the new views of the maximum contributing video to the comprehensive strength benchmark of the non-maximum contributing group and processing it with a nonlinear function. Specifically, the new views of the maximum contributing video are used as the numerator, the comprehensive strength benchmark of the non-maximum contributing group is used as the denominator, and the corresponding fraction is used as the maximum contribution ratio. The maximum contribution ratio is mapped by a hyperbolic tangent function to obtain the corresponding centralization penalty intensity. The result of subtracting the centralization penalty intensity from the constant 1 is used as the contribution structure balancing factor.

[0049] In one implementation, assume the first The maximum number of new views for a single video within a set time window for each potential trend cluster is: ;No. The number of new views for each video is ;No. The total number of videos that generated new views within the set time window for each potential trend cluster is: Then the first The formula for calculating the contribution structure check factor of each potential trend cluster is as follows:

[0050]

[0051] in, Indicates the first Contribution structure balancing factors of each potential trend cluster; Indicates the first The maximum number of new views for a single video within a set time window for a potential trend cluster; Indicates the first The number of new views for each video; Indicates the first The total number of videos that generate new views within a set time window for each potential trend cluster; This represents the hyperbolic tangent function.

[0052] It should be noted that, firstly, the denominator of the formula... To comprehensively assess the overall strength of the non-maximum contributing video group, since a single dimension is insufficient, the denominator is the sum of two parts: Part 1 The average contribution level of the non-maximum contributor group was calculated, reflecting the average quality of ordinary participants; Part Two The total collective contribution of the group was calculated, reflecting the scale of participation in the trend. Adding these two figures together constitutes a benchmark for the group's overall strength. This benchmark considers both the breadth and the average depth of participation, thus enabling a more robust assessment of the strength of the follower group. Secondly, the numerator of the formula... The absolute strength representing the greatest contributor. The ratio is calculated to represent the ratio of the strength of the largest contributor to the overall strength of the non-largest contributor group. This ratio describes the degree of dominance of a single point. If the ratio is much greater than 1, it indicates that the largest contributor is far stronger than the sum and average of all other participants, indicating a very high degree of centralization; if the ratio is much less than 1, it indicates that the other participants constitute the main body of the trend development, and the trend structure is healthy. The hyperbolic tangent function is used to map this core ratio, thus... The ratio value is mapped to This range forms a saturated penalty intensity. This means that when the degree of centralization exceeds a certain threshold, the system will impose the maximum penalty, rather than amplifying it indefinitely. Finally, through... The structure transforms the penalty intensity into a multiplicative correction factor. When the degree of centralization is high, the hyperbolic tangent function mapping value approaches 1. When the centering degree is low, the hyperbolic tangent function mapping value approaches 0, resulting in maximum suppression of the original score. If the score approaches 1, the original score is retained.

[0053] Thus, the contribution structure balancing factor was obtained by comparing the overall strength of the trend cluster's largest contributing video with that of the non-largest contributing video group.

[0054] Step S3: Obtain the internal contribution difference factor by summing and comparing the pairwise contribution differences of all videos within the trend cluster.

[0055] After the initial correction in step S2, the system has been able to effectively identify and suppress centralized trends dominated by a single video. However, in the actual video dissemination environment, there exists a more complex and hidden growth pattern. Specifically, a trend cluster may not have an absolute leader, but its popularity increment is highly concentrated in a few small group videos, while the contributions of a large number of other participants are negligible. Under this secondary concentration structure, the spread of the trend also lacks a broad mass base and self-sustainability, and its vitality remains questionable. The contribution structure balancing factor in step S2, because it mainly focuses on the relationship between the largest contributor and the rest of the group, has limited ability to identify such structural imbalances within the group. Therefore, to solve this secondary concentration problem, the goal of the second correction step is to directly quantify the overall balance of contributions of all participants within the trend cluster. To achieve this goal, a measurement method that does not depend on any specific center point (average) or specific individuals (such as the largest contributor) is needed. Directly examine the degree of difference between pairs of all members within the group. In a balanced distributed participation pattern, the contribution levels of any two participants should be similar, so the difference in their contributions will be small. The sum of the contribution differences between all possible participant pairs will inevitably be small. Conversely, in a secondary concentration model, the contribution differences between high-contribution small groups and low-contribution long-tail members will be large, resulting in a large sum of all differences. Based on this logic, this step calculates the sum of contribution differences between all pairs of participants within the trend cluster and compares it with a theoretical maximum difference value derived from the current data conditions, thereby generating a correction factor that accurately reflects the degree of internal contribution balance. When the internal contribution differences are large, approaching the theoretical maximum difference, it indicates a severely unbalanced trend structure, and the correction factor will approach 0, strongly suppressing the propagation score; conversely, if the internal contribution differences are small, it indicates that the contribution levels of all participants are similar and the trend structure is healthy, and the correction factor will approach 1, preserving the propagation score.

[0056] In summary, firstly, the sum of pairwise contribution differences is obtained by performing pairwise difference calculations on the new play count data of all videos within the trend cluster. Secondly, the maximum possible difference value is obtained by theoretically deriving the extreme values ​​of the number of videos within the trend cluster and the total new play count data. Specifically, the contribution difference value of any two different videos within the potential trend cluster is obtained by performing difference calculations on the new play count data and taking the absolute value. The sum of pairwise contribution differences is obtained by accumulating the contribution difference values ​​of all non-repeating video pairs. Finally, the maximum possible difference value is obtained by multiplying the result of subtracting a constant 1 from the number of videos within the potential trend cluster with the total new play count data within the potential trend cluster.

[0057] After obtaining the sum of pairwise contribution differences and the maximum possible difference value between videos, the internal contribution difference factor is obtained by normalizing and comparing the sum of pairwise contribution differences and the maximum possible difference value. Specifically,

[0058] For any potential trend cluster, the sum of paired contribution differences is used as the numerator, the maximum possible difference value is used as the denominator, and the resulting fraction is used as the imbalance index of the potential trend cluster; the result of subtracting the imbalance index of the potential trend cluster from the constant 1 is used as the internal contribution difference factor of the potential trend cluster.

[0059] In one implementation, assume the first The total number of new plays for each potential trend cluster within the set time window is Then the first The formula for calculating the internal contribution difference factor of each potential trend cluster is:

[0060]

[0061] in, Indicates the first Internal contribution difference factor of each potential trend cluster; Indicates the first The number of new views for each video; Indicates the first The number of new views for each video; Indicates the first The total number of videos that generate new views within a set time window for each potential trend cluster; Indicates the first The total number of new plays for each potential trend cluster within a set time window.

[0062] It should be noted that, firstly, the numerator of the formula... The sum of pairwise contribution differences is not a measure of equal contribution, but rather of whether there are structural discontinuities in the contribution distribution. In real-world video distribution scenarios, a healthy distributed trend, while uneven in its contribution distribution, typically exhibits a continuous, smoothly decreasing hierarchical structure. That is, there are a few high-contribution videos, a group of medium-contribution videos, and a large number of low-contribution videos, with transitions between different levels. In contrast, a secondary concentrated trend exhibits a significant gap in contribution distribution: for example, a few videos contribute hundreds of thousands of views, while hundreds or thousands of other videos contribute only a few hundred. These two groups lack intermediate-level videos with tens or thousands of views as a connection. The sum of pairwise contribution differences in this invention aims to capture this structural discontinuity. In secondary concentrated scenarios, the pairwise contribution difference between a few top-performing videos and a large number of bottom-performing videos can be very large. Due to the lack of an intermediate layer between the top and bottom, this large difference dominates the entire summation result, causing an abnormal inflation of the numerator value. Conversely, in a hierarchical, continuous healthy trend, although there are significant differences between high-contribution and low-contribution videos, the presence of numerous mid-tier videos broadens the distribution of pairwise differences across the entire group, resulting in a relatively small proportion of extremely large differences. Consequently, the cumulative sum of the numerator is much smaller than its theoretical maximum. Therefore, the numerator can very effectively quantify the internal polarization of the data distribution, directly addressing the secondary concentration problem that needs to be solved. Secondly, the numerator alone is insufficient for comparisons between trends because its absolute value varies with total views and number of participants. Therefore, this invention addresses this by using the denominator... Given the number of participating videos Total new views Under certain constraints, this represents the theoretical maximum degree of dissimilarity that molecules can achieve. The physical scenario corresponding to this theoretical maximum value is the most extreme case of imbalance, where a single video receives all the new views. And the rest The contribution of each video is 0. Then, the core ratio, obtained by dividing the numerator by the denominator, represents the ratio of the actually observed internal variability to the theoretically maximum possible variability. This ratio is normalized to... The imbalance index of the interval is no longer affected by the absolute heat or scale of the trend, but only reflects its inherent structural problems. Finally, through The structure transforms the imbalance index into a multiplicative correction factor. When there are severe discontinuities in the internal structure, the imbalance index approaches 1. When the value approaches 0, the propagation score is strongly suppressed; when the internal structure hierarchy is continuous, the imbalance index approaches 0. A value close to 1 indicates that the trend has passed the test of deep correction.

[0063] This completes the process of obtaining the internal contribution difference factor by summing and comparing the pairwise contribution differences of all videos within the trend cluster.

[0064] Step S4: The original propagation score is corrected by combining the contribution structure check factor and the internal contribution difference factor in turn to obtain the final trend score.

[0065] After calculations in steps S2 and S3, the system has obtained a contribution structure balancing factor for assessing the degree of trend centralization and an internal contribution difference factor for assessing the internal equilibrium of the trend. These two factors, from different but complementary dimensions, quantify the structural information missing from the original propagation score. Therefore, to obtain a comprehensive assessment result that fully reflects the trend's growth rate and quality, these two correction factors need to be applied together to the original propagation score. This is achieved by comparing the original score with these two factors... The correction factors of the intervals are multiplied to filter the original scores, thereby obtaining the final trend score of the structural information correction.

[0066] For any potential trend cluster, the cumulative total number of plays in the potential trend cluster at the beginning of the time window is used as the denominator, and the new total number of plays in the potential trend cluster within the time window is used as the numerator. The resulting fraction is used as the original propagation score of the potential trend cluster.

[0067] After obtaining the initial propagation score of the potential trend, the initial propagation score is multiplied by the contribution structure balancing factor to obtain the first propagation score after structure correction; the first propagation score is multiplied by the internal contribution difference factor to obtain the final trend score.

[0068] At this point, the final trend score is obtained by sequentially correcting the original propagation score by combining the contribution structure check factor and the internal contribution difference factor.

[0069] Step S5: By sorting the final trend scores of all potential trend clusters, the identification results of trend clusters with blockbuster potential are obtained.

[0070] This step is the final output of the method of this invention, aiming to solve the misjudgment problem existing in the prior art based on the calculation results of the preceding steps. The system sorts all potential trend clusters in descending order according to their final trend scores calculated in step S4. Potential trend clusters ranked higher in this sorting result have higher final scores, indicating that the trend not only has a high growth rate, but more importantly, its growth pattern has passed the dual tests of contribution structure balancing factors and internal contribution difference factors. This proves that the trend does not have a serious single-point dominance problem, and that there are no structural breaks in its internal participation structure. Therefore, the system identifies these top-ranked trend clusters as truly explosive trends with high vitality and broad dissemination potential. Conversely, for trend clusters with high initial dissemination scores but low final trend scores due to low contribution structure balancing factors or internal contribution difference factors, the system identifies them as short-lived trends driven by centralization or secondary concentration effects.

[0071] Thus, by introducing a correction factor that can quantify the distribution structure of heat contribution within a cluster, this invention compensates for the loss of key information caused by total aggregation operations, and achieves accurate differentiation between centralized and distributed growth modes, thereby significantly improving the accuracy of trend prediction for viral videos.

[0072] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A trend analysis method for viral videos based on AI, characterized in that, The AI-based trend analysis method for viral videos includes: Step S1: Collecting and preprocessing the popularity data of potential trend cluster videos to obtain a structured time series dataset; Step S2: Comparing the comprehensive strength of the videos with the largest contribution and those with the smallest contribution within a trend cluster to obtain a contribution structure balancing factor; Step S3: Accumulating and comparing the pairwise contribution differences of all videos within a trend cluster to obtain an internal contribution difference factor; Step S4: Correcting the original propagation score by sequentially combining the contribution structure balancing factor and the internal contribution difference factor to obtain the final trend score; Step S5: Sorting the final trend scores of all potential trend clusters to obtain the identification results of trend clusters with viral potential.

2. The trend analysis method for viral videos based on AI according to claim 1, characterized in that, The process of collecting and preprocessing video popularity data from potential trend clusters to obtain a structured time-series dataset includes: clustering videos based on their multimodal features using a clustering method to obtain potential trend clusters; setting a time window length; collecting the identifier information and corresponding cumulative play count data of each video in the potential trend cluster at the start and end times within the time window to obtain a video popularity dataset containing unique video identifiers, start cumulative play count data, and end cumulative play count data; formatting and organizing the collected video popularity dataset, establishing an index with the unique video identifier as the primary key, and obtaining a structured time-series dataset storing the play count changes of each video within the time window.

3. The trend analysis method for viral videos based on AI according to claim 1, characterized in that, The step of obtaining a contribution structure balancing factor by comparing the comprehensive strength of the videos with the largest and smallest contributions in a trend cluster includes: statistically processing the new play count data of each video within a potential trend cluster to obtain the new play count of the largest contributing video; summarizing and analyzing the new play count data of the videos with the smallest contributions within a potential trend cluster to obtain the comprehensive strength benchmark of the group with the smallest contributions; and obtaining the contribution structure balancing factor by calculating the ratio of the new play count of the largest contributing video to the comprehensive strength benchmark of the group with the smallest contributions and processing it using a nonlinear function.

4. The trend analysis method for viral videos based on AI according to claim 3, characterized in that, The process involves statistically processing the new play count data of each video within a trend cluster to obtain the new play count of the video with the largest contribution, and summarizing and analyzing the new play count data of non-maximum contributing videos within the trend cluster to obtain a comprehensive strength benchmark for the non-maximum contributing group. This includes: setting a time window; statistically analyzing the new play count data of all videos in the potential trend cluster within the time window to obtain the target video with the largest new play count, and using this target video's new play count as the new play count of the maximum contributing video; summing the new play count data of all non-maximum contributing videos in the potential trend cluster (excluding the maximum contributing video) to obtain the total new play count of the non-maximum contributing group, and dividing the total new play count of the non-maximum contributing group by the number of non-maximum contributing videos to obtain the average new play count of the non-maximum contributing group; and summing the total new play count of the non-maximum contributing group with the average new play count of the non-maximum contributing group to obtain a comprehensive strength benchmark for the non-maximum contributing group.

5. The trend analysis method for viral videos based on AI according to claim 3, characterized in that, The method of obtaining a contribution structure balancing factor by calculating the ratio of the new views of the video with the largest contribution to the comprehensive strength benchmark of the non-largest contribution group and processing it with a nonlinear function includes: taking the new views of the video with the largest contribution as the numerator, taking the comprehensive strength benchmark of the non-largest contribution group as the denominator, and taking the corresponding fraction as the ratio of the largest contribution; mapping the ratio of the largest contribution to the hyperbolic tangent function to obtain the corresponding centralization penalty intensity; and taking the result of subtracting the centralization penalty intensity from the constant 1 as the contribution structure balancing factor.

6. The trend analysis method for viral videos based on AI according to claim 1, characterized in that, The method of obtaining the internal contribution difference factor by summing and comparing the pairwise contribution differences of all videos within the trend cluster includes: performing pairwise difference calculations on the new play count data of all videos within the trend cluster to obtain the sum of paired contribution differences; and deriving the maximum possible difference value by theoretically extreme value derivation of the number of videos within the trend cluster and the total new play count data; and obtaining the internal contribution difference factor by normalizing and comparing the sum of paired contribution differences and the maximum possible difference value.

7. The trend analysis method for viral videos based on AI according to claim 2, characterized in that, The process involves performing pairwise difference calculations on the new play count data of all videos within a trend cluster to obtain the sum of paired contribution differences, and deriving the maximum possible difference value by theoretically extreme value derivation of the number of videos within the trend cluster and the total new play count data. This includes: performing difference calculations on the new play count data of any two different videos in the potential trend cluster and taking the absolute value to obtain the contribution difference value of any two different videos; accumulating the contribution difference values ​​of all non-repeating video pairs to obtain the sum of paired contribution differences; and multiplying the result of subtracting a constant 1 from the number of videos within the potential trend cluster with the total new play count data within the potential trend cluster to obtain the maximum possible difference value.

8. The trend analysis method for viral videos based on AI according to claim 1, characterized in that, The method of obtaining the internal contribution difference factor by normalizing and comparing the sum of paired contribution differences with the maximum possible difference value includes: for any potential trend cluster, taking the sum of paired contribution differences as the numerator, taking the maximum possible difference value as the denominator, and taking the corresponding fraction as the imbalance index of the potential trend cluster; and taking the result of subtracting the imbalance index of the potential trend cluster from the constant 1 as the internal contribution difference factor of the potential trend cluster.

9. The trend analysis method for viral videos based on AI according to claim 1, characterized in that, The process of obtaining the final trend score by sequentially correcting the original propagation score with the contribution structure balancing factor and the internal contribution difference factor includes: setting a time window; for any potential trend cluster, using the cumulative total number of plays in the potential trend cluster at the beginning of the time window as the denominator, and the total number of new plays in the potential trend cluster within the time window as the numerator, and using the resulting fraction as the original propagation score of the potential trend cluster; obtaining the first propagation score after structure correction by multiplying the original propagation score with the contribution structure balancing factor; and obtaining the final trend score by multiplying the first propagation score with the internal contribution difference factor.

10. A trend analysis system for viral videos based on AI, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement an AI-based trend analysis method for viral videos according to any one of claims 1-9.

Images