Short video traffic dynamic transaction and multi-dimensional value evaluation intelligent system

Abstract

The application discloses a short video traffic dynamic transaction and multi-dimensional value evaluation intelligent system and relates to the field of electronic commerce.The system is characterized in that, in the put-in window optimization module, a fixed put-in window length twindow is determined, so that the put-in period can be adaptively adjusted according to the fractal robustness of short videos at multiple time scales such as minute, hour and day levels; when the multi-time dimension fractal continuous robustness index FCSms is high, the fixed put-in window length twindow is lengthened to utilize the long tail effect of short videos; and when the multi-time dimension fractal continuous robustness index FCSms is low, the fixed put-in window length twindow is shortened to avoid excessive put-in. Compared with the existing fixed-length put-in mode, the system makes the related short-time put-in window of the short video platform more suitable for the video traffic life cycle, and improves the accuracy and timeliness of the advertisement put-in.

Description

Technical Field

[0001] This invention relates to the field of e-commerce technology, specifically to an intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic. Background Technology

[0002] With the development of e-commerce, especially the dynamic management and value assessment technology of short video content traffic, this paper specifically involves an intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic based on multi-timescale fractal feature analysis. This system treats the changes in short video views at multiple timescales, such as minutes, hours, and days, as multi-scale fractal curves. Combining fractal dimension, porosity, slope variance, and other multi-dimensional feature indicators, it establishes a quantitative model for traffic sustainability and stability. The analysis results are directly applied to the optimization of advertising resource placement windows and transaction priority ranking, achieving refined mining and utilization of the commercial value of short videos.

[0003] Currently, most short video traffic evaluation and advertising systems use fixed-time-window statistics, such as total views in 24 hours or incremental views over 7 days, as the basis for judging video value. This single-time-scale statistical method ignores the self-similarity of short video views across different time scales and fails to reflect the extensibility and multi-stage growth patterns of traffic. Furthermore, existing methods generally employ static priorities or single-metric rankings when allocating advertising resources, lacking cross-time-scale analysis of traffic structure stability. This leads to the premature elimination of videos with long-term growth potential or a severe misalignment between resource allocation rhythm and video popularity curves, thereby reducing the overall return on advertising and user reach.

[0004] The main reason for these shortcomings lies in the fact that existing traffic assessment systems fail to capture the fractal and gap structure characteristics of video views across multiple time scales, and lack the ability to model dynamic factors such as the delay in the surge of popularity and cross-scale stability. When this deficiency occurs, it leads to the following abnormal effects: First, videos that experience a surge in popularity later in the campaign cannot receive sufficient advertising support in the early stages, missing the optimal period for viral spread; second, videos that experience a short-term surge followed by a rapid decline consume a large amount of high-value resources in the early stages of advertising, resulting in a decrease in the advertiser's return on investment; third, there is a severe mismatch between the advertising window and the video traffic lifecycle, leading to significant waste of exposure during some campaign periods, while other periods suffer from insufficient resource supply. These problems not only reduce the efficiency of advertising resource utilization but may also affect the platform's overall commercial revenue and the user's viewing experience. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic, solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic, comprising:

[0007] Multi-dimensional traffic acquisition module: Collects video playback data from short video platforms in real time using acquisition tools, and transmits the video playback data to the cross-scale fractal analysis module;

[0008] Cross-scale fractal analysis module: By preprocessing video playback data, a standardized dataset is obtained, and based on the standardized dataset, feature extraction is performed to obtain a fractal dataset;

[0009] Fractal robustness assessment module: Calculates and outputs the multi-time-dimensional fractal robustness index FCSms based on the fractal dataset, and presets the multi-time-dimensional fractal robustness index threshold Fth for preliminary comparative assessment;

[0010] Priority ranking module: Triggers trading priority strategies through preliminary comparison and evaluation, calculates and outputs fractal survival deployment window score FSWms based on the multi-time dimension fractal robustness index FCSms, and prioritizes video trading based on the fractal survival deployment window score FSWms.

[0011] The ad delivery window optimization module determines the ad delivery window length (twindow) based on the multi-time dimension fractal robustness index (FCSms) and allocates ad resources and delivery time based on the ranking results of the fractal survival ad delivery window score (FSWms).

[0012] Preferably, the multi-dimensional traffic acquisition module includes an acquisition unit and a transmission unit;

[0013] The acquisition unit collects video playback data from the short video platform at minute, hour, and day time scales by setting up acquisition tools on the short video platform.

[0014] The data acquisition tool includes a data acquisition terminal and an acquisition control program;

[0015] The data acquisition terminal is deployed on the backend data interface node of the short video platform, located between the platform's data exchange layer and business processing layer, to acquire raw playback data that has not been aggregated or compressed. The data acquisition terminal has data buffering, real-time data packaging, and network communication functions.

[0016] The acquisition control program runs within the data acquisition terminal, calls the platform's open API, collects log streams from the short video platform, and extracts video playback data and additional parameters from the log streams.

[0017] The video play count data includes a unique video identifier, a corresponding timestamp, scale consistency slope variance Xslope, popularity burst delay time TB, reference delay time TR, future exposure availability EA, burst skew SB, minute-level play count sequence Pmin, hour-level play count sequence Phour, and day-level play count sequence Pday.

[0018] The minute-level playback sequence Pmin has a sampling step of 1 minute and a box counting scale r∈{1,2,4,8,16} minutes;

[0019] The hourly playback sequence Phour has a sampling step of 1 hour and a box counting scale r∈{1,2,4,8} hours;

[0020] The daily play count sequence Pday has a sampling step of 1 day and a box counting scale r∈{1,2,4} days.

[0021] Preferably, the transmission unit packages the video playback data into data packets with unique video identifiers and corresponding timestamps according to the time scale, and sends them to the cross-scale fractal analysis module in an encrypted manner via the high-speed message queue transmission protocol MQTT. After receiving the data packets, the cross-scale fractal analysis module reconstructs the order of the data packets and generates a receipt confirmation receipt.

[0022] Preferably, the cross-scale fractal analysis module includes a data preprocessing unit for preprocessing video playback data to generate a standardized dataset;

[0023] The preprocessing includes time synchronization processing, unit unification processing, and data structuring and storage.

[0024] The time synchronization process aligns the minute-level playback sequence Pmin, hour-level playback sequence Phour, and day-level playback sequence Pday in the video playback data with timestamps according to a global time base, and fills in missing time data points using an interpolation algorithm, wherein the interpolation method is linear interpolation.

[0025] The dimensionless processing involves applying the Z-Score standardization method to all parameters in the video playback data to convert parameters with different physical dimensions into dimensionless values ​​with a mean of 0 and a standard deviation of 1.

[0026] The data structuring and storage involves grouping and storing the video playback data, after unifying the units of measurement, into a standardized dataset according to time scale and parameter type, and adding a unique video identifier and corresponding timestamp to each record.

[0027] Preferably, the cross-scale fractal analysis module further includes a fractal feature extraction unit, used to extract features from the standardized dataset to obtain a fractal dataset;

[0028] The fractal dataset includes minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, day-level fractal dimension Dday, geometric mean porosity LG, minute-level porosity Amin, hour-level porosity Ahour, day-level porosity Aday, scale consistency slope variance Xslope, heat burst delay time TB, reference delay time TR, future exposure availability EA, and burst skewness SB.

[0029] Preferably, the fractal robustness assessment module includes a fractal robustness analysis unit;

[0030] The fractal robustness analysis unit extracts the heat burst delay time TB and reference delay time TR from the fractal dataset to form a slow-heat factor, constructs a stability factor by scaling the slope variance Xslope, and constructs a complexity and spatial normalization factor by combining the minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, and day-level fractal dimension Dday with the geometric mean porosity LG. By multiplying and summing the slow-heat factor, stability factor, and complexity and spatial normalization factor, the multi-time-dimensional fractal robustness index FCSms is calculated and output. This index is used to measure the complexity, stability, void structure characteristics, and heat growth pattern of a short video at different time scales, and reflects its long-term sustainability potential through a single numerical value.

[0031] Preferably, the fractal robustness assessment module includes an assessment unit;

[0032] The evaluation unit calculates the multi-time-dimensional fractal robustness index FCSms in batches on the historical sample set of the same product category, obtains the empirical distribution, takes the 80th percentile as the initial multi-time-dimensional fractal robustness index threshold Fth, and re-evaluates it on a monthly rolling basis.

[0033] Next, a preliminary comparative evaluation is conducted between the real-time acquired multi-time-dimensional fractal robustness index FCSms and the multi-time-dimensional fractal robustness index threshold Fth to determine the robustness of the short video's traffic structure across multiple time scales, and triggering based on the preliminary comparative evaluation results; the specific evaluation content is as follows:

[0034] When the multi-time dimension fractal continuous robustness index FCSms is greater than or equal to the multi-time dimension fractal continuous robustness index threshold Fth, the current short video traffic structure is determined to be robust, and the transaction priority strategy is triggered at this time.

[0035] When the multi-time dimension fractal robustness index FCSms is less than the multi-time dimension fractal robustness index threshold Fth, the current short video traffic structure is determined to be unstable, and the conventional strategy should be applied.

[0036] Preferably, the priority ranking module includes a fractal survival deployment window analysis unit;

[0037] The fractal survival deployment window analysis unit obtains the effective fractal dimension Deff by averaging the minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, and day-level fractal dimension Dday after triggering the transaction priority strategy.

[0038] Simultaneously, the multi-time-dimensional fractal robustness index FCSms, output from the fractal robustness assessment module, is combined with the effective fractal dimension Deff, as well as the scale consistency slope variance Xslope, future exposure availability EA, and burst skewness SB to calculate and output the fractal survival deployment window score FSWms.

[0039] Preferably, the priority sorting module further includes a sorting unit;

[0040] The sorting unit sorts all videos that trigger the transaction priority strategy by generating a priority queue in descending order of the Fractal Survival Window Score (FSWms). The specific sorting steps are as follows:

[0041] The 85th percentile P85 and 95th percentile P95 of the fractal survival delivery window score (FSWms) distribution based on historical short videos are used as step thresholds. The real-time fractal survival delivery window score (FSWms) is compared and evaluated with the step thresholds, and a priority queue is generated based on the secondary comparison and evaluation.

[0042] When the fractal survival delivery window score FSWms ≥ 95 quantile P95, it is classified as a secondary priced item and is loaded preferentially.

[0043] When 85 percentile P85 ≤ Fractal Survival Delivery Window Score FSWms < 95 percentile P95, it is classified as Level 1 premium loading, and regular loading is performed.

[0044] When the fractal survival delivery window score FSWms < 85 quantile P85, it is classified as a three-level priced system and classified as a basic layer load.

[0045] Preferably, the delivery window optimization module includes a delivery optimization unit;

[0046] The delivery optimization unit determines a fixed delivery window length twindow based solely on the multi-time dimension fractal robustness index FCSms when the target short video undergoes preliminary comparative evaluation. Specifically, twindow = t0 × (1 + FCSms), where t0 represents the basic delivery duration set by the platform.

[0047] And register to form a fixed delivery window [ts, te], where ts represents the start time of the delivery window and te represents the end time of the delivery window;

[0048] The ad delivery optimization unit does not accept any additions or subtractions to the fixed ad delivery window length twindow by the Fractal Survival Ad Delivery Window Score (FSWms). When the fixed ad delivery windows [ts,te] of different videos overlap and platform resources are limited, the loading intensity and starting order of ad resources are allocated only within their respective fixed ad delivery windows [ts,te] based on the ranking results of the Fractal Survival Ad Delivery Window Score (FSWms), keeping the start and end boundaries of each fixed ad delivery window [ts,te] unchanged.

[0049] This invention provides an intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic. It has the following beneficial effects:

[0050] (1) This system optimizes the delivery window by setting a fixed delivery window length (twindow) to allow the delivery cycle to be adaptively adjusted based on the fractal robustness of short videos across multiple time scales, including minutes, hours, and days. When the multi-time-dimensional fractal robustness index (FCSms) is high, the fixed delivery window length (twindow) is extended to utilize the long-tail effect of the video; when the FCSms is low, the fixed delivery window length (twindow) is shortened to avoid over-delivery. Compared with existing fixed-duration delivery methods, this system makes the delivery window more aligned with the video traffic lifecycle, improving the accuracy and timeliness of ad delivery.

[0051] (2) In the system's delivery window optimization module, the start and end boundaries of the fixed delivery window [ts,te] are not affected by the adjustment of the fractal survival delivery window score (FSWms). Only within the fixed window, the system prioritizes the loading intensity and delivery order of advertising resources based on the FSWms score. When platform resources are limited and different video delivery windows overlap, this strategy ensures fairness in window length while allocating high-value resources to higher-scoring videos through a priority queue, preventing resources from being occupied by low-potential videos and thus improving resource allocation efficiency.

[0052] (3) This system determines the delivery time boundary by using the multi-time dimension fractal robustness index FCSms and optimizes the resource allocation order by using the fractal survival delivery window score FSWms, thus realizing the division of labor and cooperation between the time dimension and the resource dimension. This cooperation mechanism effectively prevents low robustness videos from occupying high-value exposure resources during the window period, while ensuring that high-potential videos receive sufficient delivery support during key time periods, thereby improving the conversion rate of advertising and overall business revenue, and reducing resource waste. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the intelligent system structure for dynamic trading and multi-dimensional value assessment of short video traffic according to the present invention.

[0054] Figure 2 This is a schematic diagram of the playback volume sequence over a time scale. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1: This invention provides an intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic. Please refer to [link / reference]. Figure 1 and Figure 2 ,include:

[0057] Multi-dimensional traffic acquisition module: Collects video playback data from short video platforms in real time using acquisition tools, and transmits the video playback data to the cross-scale fractal analysis module;

[0058] Cross-scale fractal analysis module: By preprocessing video playback data, a standardized dataset is obtained, and based on the standardized dataset, feature extraction is performed to obtain a fractal dataset;

[0059] Fractal robustness assessment module: Calculates and outputs the multi-time-dimensional fractal robustness index FCSms based on the fractal dataset, and presets the multi-time-dimensional fractal robustness index threshold Fth for preliminary comparative assessment;

[0060] Priority ranking module: Triggers trading priority strategies through preliminary comparison and evaluation, calculates and outputs fractal survival deployment window score FSWms based on the multi-time dimension fractal robustness index FCSms, and prioritizes video trading based on the fractal survival deployment window score FSWms.

[0061] The ad delivery window optimization module determines the ad delivery window length (twindow) based on the multi-time dimension fractal robustness index (FCSms) and allocates ad resources and delivery time based on the ranking results of the fractal survival ad delivery window score (FSWms).

[0062] In this embodiment, the system deploys a data acquisition terminal and acquisition control program on the short video platform through a multi-dimensional traffic acquisition module. It collects video playback data in real time at minute, hourly, and daily timescales, including video unique identifiers, timestamps, and parameters such as scale consistency slope variance (Xslope), popularity burst delay time (TB), reference delay time (TR), future exposure availability (EA), and burst skewness (SB). This data is then sent to the cross-scale fractal analysis module via an encrypted transmission protocol. Next, the cross-scale fractal analysis module performs time synchronization, dimension unification, and data structuring preprocessing on the received video playback data to generate a standardized dataset. Based on this standardized dataset, fractal feature parameters are extracted, including minute-level fractal dimension (Dmin), hourly fractal dimension (Dhour), daily fractal dimension (Dday), and geometric mean porosity (LG), forming a fractal dataset that provides multi-dimensional input for subsequent robustness analysis. Next, the fractal robustness assessment module utilizes the fractal dataset to calculate and output the multi-temporal fractal robustness index FCSms through a structural combination of slow-burn factors, stability factors, and complexity and spatial normalization factors. Based on the historical distribution of videos of the same category, it determines the threshold Fth for the multi-temporal fractal robustness index and compares the real-time calculated FCSms with the threshold value to determine whether the video's traffic structure across multiple time scales possesses sustained robustness. When the determination result triggers a transaction priority strategy, the priority ranking module combines the effective fractal dimension Def, the multi-temporal fractal robustness index FCSms, the scale consistency slope variance Xslope, the future exposure availability EA, and the burst skewness SB to calculate and output the fractal survival delivery window score FSWms. It then sorts all videos that trigger the transaction priority strategy by score, generating a priority queue and categorizing them into priority loading, regular loading, and basic layer loading levels to optimize the allocation of advertising resource loading order and intensity. Finally, the delivery window optimization module determines a fixed delivery window length (twindow) based on the multi-time dimension fractal robustness index (FCSms), forming a fixed delivery time interval [ts, te]. Within the window, advertising resources are allocated according to the ranking results of the fractal survival delivery window score (FSWms), ensuring that high-potential videos receive priority access to high-value exposure resources when different video windows overlap and resources are limited. Through the above implementation methods, this system achieves multi-scale feature quantification of video traffic, accurate robustness assessment, and hierarchical priority allocation of advertising resources. This not only improves the timeliness and accuracy of advertising delivery but also reduces resource waste, maximizes the commercial conversion capabilities of high-potential videos, and achieves the goal of optimal traffic value allocation in a dynamic competitive environment.

[0063] Example 2: Please refer to Figure 1 and Figure 2 Specifically: the multi-dimensional traffic acquisition module includes an acquisition unit and a transmission unit;

[0064] The data collection unit collects video playback data from short video platforms at minute, hour, and day time scales by setting up data collection tools on the short video platforms.

[0065] Data acquisition tools include data acquisition terminals and acquisition control programs;

[0066] The data acquisition terminal obtains raw playback data that has not been aggregated or compressed by deploying the backend data interface node of the short video platform between the platform's data exchange layer and business processing layer. The data acquisition terminal has data buffering, real-time data packaging and network communication functions.

[0067] The data acquisition control program runs within the data acquisition terminal, calls the platform's open API, collects log streams from the short video platform, and extracts video playback data and additional parameters from the log streams.

[0068] Video play count data includes a unique video identifier, corresponding timestamp, scale consistency slope variance Xslope, popularity burst delay time TB, reference delay time TR, future exposure availability EA, burst skewness SB, minute-level play count sequence Pmin, hour-level play count sequence Phour, and day-level play count sequence Pday.

[0069] The minute-level playback sequence Pmin has a sampling step of 1 minute and a box counting scale r∈{1,2,4,8,16} minutes;

[0070] The hourly playback sequence Phour has a sampling step of 1 hour and a box counting scale r∈{1,2,4,8} hours;

[0071] The sampling step size for the daily play count sequence Pday is 1 day, and the box counting scale is r∈{1,2,4} days.

[0072] in:

[0073] The scale consistency slope variance Xslope represents the stability of the rate of change of video views at different time scales, such as minutes, hours, and days. First, the slope sequence of the change in video views is calculated at each time scale.

[0074] Then, compare the consistency of slope changes at different scales and take the variance value as the scale consistency slope variance Xslope. The smaller the variance, the more consistent the playback trend at each scale. The larger the variance, the greater the difference in changes at different scales. It is used to measure the multi-scale stability of video traffic and is one of the important inputs for fractal persistence judgment.

[0075] The "Trendy Explosion Delay Time (TB)" refers to the time interval from the video's release to the first significant increase in views, representing the "Trendy Explosion." It reflects the lag in the video's popularity and is useful for determining whether a video is suitable for short-term or long-term advertising campaigns.

[0076] The reference delay time (TR) is the relative difference in delay time between the target video and the average delay time of the popularity surge of similar videos. In fractal analysis, it is used as a reference quantity to correct the prediction model of the delivery window.

[0077] Future Exposure Availability (EA) forecasts the total amount of exposure a video may receive in the platform's recommendation algorithm over a period of time. This forecast is based on the ARIMA model combined with historical playback trends, current platform recommendation weights, and user activity models, and directly impacts bidding strategies and ad delivery order in advertising transactions.

[0078] The burst skewness (SB) reflects the symmetry of the video's popularity burst curve, describes the difference in play volume distribution before and after the burst, judges the morphological characteristics of the video's popularity life cycle, and assists in selecting the duration of the campaign.

[0079] The transmission unit packages video playback data into data packets with unique video identifiers and corresponding timestamps according to time scales, and sends them to the cross-scale fractal analysis module in an encrypted manner via the high-speed message queue transmission protocol MQTT. After receiving the data packets, the cross-scale fractal analysis module reconstructs the order of the data packets and generates a receipt confirmation acknowledgment.

[0080] In this embodiment, the system's multi-dimensional traffic acquisition module works collaboratively with the acquisition unit and the transmission unit. The acquisition unit collects video playback data and additional parameters in real time at minute, hour, and day timescales on the short video platform. The minute-level data captures short-term fluctuations and immediate hotspots, the hour-level data reflects mid-term recommendations and activity rhythms, and the day-level data depicts the long-tail effect and sustained appeal, ensuring the comprehensiveness and stability of the traffic structure across multiple timescales. The data acquisition terminal is deployed between the platform's data exchange layer and business processing layer, directly acquiring raw data without aggregation or compression. This avoids data accuracy loss caused by internal platform aggregation, sampling, or caching, thereby obtaining a high-fidelity traffic curve in subsequent fractal analysis and reducing the risk of underestimating fractal dimensions. The scale consistency slope variance (Xslope) is used to measure the consistency of the trend of video playback changes across multiple timescales. A smaller variance indicates a stable trend, which can effectively identify videos suitable for continuous distribution and eliminate short-lived traffic. The dual-parameter structure of the heat burst delay time (TB) and the reference delay time (TR) is used to distinguish between instant burst types. For slow-burning videos, to avoid window matching errors, for example, slow-burning videos can be allocated a longer delivery period to release value; Future Exposure Availability (EA) is based on the ARIMA model to predict the future realizable traffic scale, ensuring that advertising resources match potential exposure capabilities and preventing over-investment or supply shortages; Burst Skewness (SB) reflects the symmetry of the popularity burst curve, distinguishing between concentrated bursts and steady growth patterns, and guiding the timing of placement to avoid a decline in conversion rates; The transmission unit transmits data encrypted via MQTT and reconstructs the time sequence to ensure the integrity and consistency of the time series, preventing fractal calculation deviations caused by data disorder or tampering; The above implementation process, through precise collection, robust feature extraction, and secure transmission across the entire chain, not only improves the authenticity and stability of fractal analysis input data, but also ensures, in a physical sense, that the acquired multi-timescale traffic features can accurately reflect the dissemination rhythm and life cycle characteristics of short videos, thus providing a high-quality data foundation for subsequent robustness assessment and prioritization, achieving precision in advertising timing and resource allocation and maximizing returns.

[0081] Example 3: Please refer to Figure 1 Specifically: the cross-scale fractal analysis module includes a data preprocessing unit, which is used to preprocess video playback data to generate a standardized dataset;

[0082] Preprocessing includes time synchronization, unit unification, and data structuring and storage;

[0083] Time synchronization processing aligns the minute-level playback sequence Pmin, hour-level playback sequence Phour, and day-level playback sequence Pday in the video playback data with timestamps based on a global time reference. It fills in missing time data points using an interpolation algorithm, where the interpolation method is linear interpolation, to ensure that data at different time scales are comparable at the same reference time.

[0084] The dimensionless processing standardizes the values ​​of all parameters in the video playback data by applying the Z-Score standardization method, so that parameters with different physical dimensions are converted into dimensionless values ​​with a mean of 0 and a standard deviation of 1.

[0085] Data structuring and storage involves grouping and storing the video playback data, after unifying the dimensions, into a standardized dataset according to time scale and parameter type. Each record is then assigned a unique video identifier and a corresponding timestamp to ensure that it can be directly accessed in subsequent fractal calculations.

[0086] The cross-scale fractal analysis module also includes a fractal feature extraction unit, which is used to extract features from the standardized dataset to obtain a fractal dataset;

[0087] The fractal dataset includes minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, day-level fractal dimension Dday, geometric mean porosity LG, minute-level porosity Amin, hour-level porosity Ahour, day-level porosity Aday, scale consistency slope variance Xslope, heat burst delay time TB, reference delay time TR, future exposure availability EA, and burst skewness SB.

[0088] Specifically, the minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, and day-level fractal dimension Dday are calculated by extracting minute-level playback volume sequences Pmin, hour-level playback volume sequences Phour, and day-level playback volume sequences Pday from a standardized dataset. The playback volume sequences at each time scale are arranged in ascending order by timestamp, and a two-dimensional curve of time versus playback volume is constructed. Based on box counting, the fractal dimensions of the minute-level, hour-level, and day-level curves are calculated respectively. The minute-level calculation result is denoted as the minute-level fractal dimension Dmin, the hour-level calculation result as the hour-level fractal dimension Dhour, and the day-level calculation result as the day-level fractal dimension Dday. The steps are as follows:

[0089] Step 1: Set multiple box counting scales ri, with the unit being the minimum time interval of the corresponding time scale;

[0090] Step 2: Divide the time axis into intervals using multiple box counting scales ri, and calculate the minimum number of boxes N(ri) required to cover the curve;

[0091] Step 3: Calculate lnN(ri) and ln(1 / ri) and perform linear regression fitting to obtain the fractal dimension. Where d represents the integral function and ln represents the natural logarithm;

[0092] Minute-level porosity Amin, hour-level porosity Ahour, and day-level porosity Aday are obtained using a sliding box porosity calculation method. The steps are as follows: Select a preset set of box counting scales R for the minute-level playback volume sequence Pmin, hour-level playback volume sequence Phour, and day-level playback volume sequence Pday at the corresponding time scales. Slide the playback volume time series according to the box counting scale ri to obtain the quality Mr, where the quality Mr is defined as the binary occupied quality. Calculate the porosity curve and take the exponent of the arithmetic mean of each box counting scale ri to obtain the scale porosity.

[0093] Minute-level porosity Amin: Represents the "sparseness" of the video playback curve at the minute-level time scale, that is, the proportion of time with playback below zero in the entire minute-level time series, and the distribution density of these low-value intervals on the time axis;

[0094] Hourly porosity Ahour: This represents the distribution and proportion of low play count intervals within an hourly timescale, also based on the play count curve.

[0095] Aday porosity: Represents the proportion of "empty windows" in the number of views over a long period on a day-scale time scale, such as the proportion of low-popularity periods of a video during the day.

[0096] Geometric mean porosity (LG) is obtained by averaging minute-level porosity (Amin), hour-level porosity (Ahour), and day-level porosity (Aday), and is used to penalize the complexity of hollow structures.

[0097] In this embodiment, the cross-scale fractal analysis module of the system first completes time synchronization processing, dimension unification processing, and data structuring and storage through the data preprocessing unit: Time synchronization processing aligns the minute-level playback volume sequence Pmin, hour-level playback volume sequence Phour, and day-level playback volume sequence Pday according to a unified benchmark and fills the gaps with linear interpolation, aiming to eliminate spurious fluctuations caused by cross-scale time misalignment; Dimension unification processing performs Z-score standardization on all parameters, converting different physical dimensions into comparable dimensionless values, avoiding the "drowning" of fine-grained signals by a large-scale quantity; Data structuring and storage groups the standardized multi-scale data according to "time scale × parameter type" and assigns a unique video identifier. The system incorporates timestamps to ensure that fractal calculations can be directly invoked without loss and with traceability. Subsequently, the fractal feature extraction unit extracts minute-level fractal dimensions Dmin, hour-level fractal dimensions Dhour, and day-level fractal dimensions Dday from the time-play count curve using box counting to characterize self-similar complexity. The sliding box porosity method is used to obtain minute-level porosity Amin, hour-level porosity Ahour, day-level porosity Aday, and geometric mean porosity LG to quantify the "window" sparsity of traffic. At the same time, the scale consistency slope variance Xslope, heat burst delay time TB, reference delay time TR, future exposure availability EA, and burst skewness SB are retained as robustness and rhythm discrimination factors. This approach is implemented because real-world platform environments often experience issues such as "holiday breakpoints," "time drift caused by batch new product launches," and "scale imbalance due to peak sales events," which directly skew fractal dimension and porosity assessments. Alignment, standardization, and structuring effectively mitigate data misalignment and dimensional bias. Furthermore, using a combination of fractal dimension and porosity to simultaneously characterize "complexity and sparsity" avoids misjudgments caused by relying solely on peak values ​​or averages, such as mistaking short spikes for long tails or prematurely eliminating slow-moving videos. The resulting benefits include: more robust feature inputs, significantly improved cross-scale comparability, and more reliable differentiation between long tails and short peaks. This directly enhances the accuracy of subsequent multi-time-dimensional fractal robustness index (FCSms) and fractal survival deployment window score (FSWms), thereby reducing mismatches and resource waste on the deployment side and improving the profitability of priority ranking and window configuration.

[0098] Example 4: Please refer to Figure 1 Specifically: the fractal robustness assessment module includes a fractal robustness analysis unit;

[0099] The fractal robustness analysis unit extracts the heat burst delay time TB and reference delay time TR from the fractal dataset to form a slow-heat factor, constructs the scale consistency slope variance Xslope to form a stability factor, and constructs the complexity and spatial normalization factor by combining the minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, and day-level fractal dimension Dday with the geometric mean porosity LG. By multiplying and summing the slow-heat factor, stability factor, and complexity and spatial normalization factor, the multi-time-dimensional fractal robustness index FCSms is calculated and output. This index is used to measure the complexity, stability, void structure characteristics, and heat growth pattern of a short video at different time scales, and reflects its long-term sustainability potential through a single value.

[0100] The multi-time-dimensional fractal enduring robustness index FCSms is calculated and output using the following algorithm formula;

[0101] ;

[0102] In the formula, ln represents the natural logarithm;

[0103] Basic source:

[0104] Box-counting Dimension: A fundamental quantity in fractal geometry, representing the minimum number of boxes N(ri) for a given box size r on a time-play count curve. Linear regression approximates the slope;

[0105] Porosity / Interstitiality: A commonly used "sparse / void" index in texture statistics and fractal analysis; slider quality Mr:

[0106] The concept of scale consistency: The variance of the regression slope over multiple box counting scales ri measures cross-scale stability; the smaller the variance, the more stable the system.

[0107] Log-stabilization of time ratios: Stabilizing the time ratios Taking ln(1+⋅), we obtain a dimensionless, robust "slow-burning" enhancement term;

[0108] Structural changes, deductions based on the above.

[0109] Fractal complexity fusion: using As a geometrically averaged fractal dimension spanning three scales (minute / hour / day), subjective weighting is avoided;

[0110] Gap penalty normalization: Apply a logarithmic penalty of 1+ln(LG) to the "hollow complexity"; as the overall sparsity increases, the exponent is naturally suppressed;

[0111] Stability gain / suppression: with Suppress samples that are unstable across scales;

[0112] Slow-heat amplification: with Reward the "slow-burning long-tail" characteristic;

[0113] in, This represents the slow-burn factor. When the heat burst delay time TB is greater than the reference delay time TR, the index increases, indicating a preference for slow-burning videos with "strong staying power". When the heat burst delay time TB=0, this item is 0, suppressing samples that quickly decay after an immediate peak.

[0114] The stability factor is indicated by the slope across scales being more stable (Xslope→0), with the factor increasing to 1. If the multi-scale morphology is unstable (Xslope is large), the overall index is suppressed.

[0115] Representing complexity and spatial normalization factor, the higher the fractal dimension, the stronger the wave complexity / self-similarity.

[0116] Denominator: When the geometric mean porosity LG→1 (uniform and continuous), ln(LG)→0, no penalty; when the geometric mean porosity LG increases and becomes more sparse and void, ln(⋅) increases, forming a logarithmic penalty to avoid "hollow complexity".

[0117] Verification of dimensional consistency and reasonableness:

[0118] Fractal dimension D*: a dimension quantity, naturally dimensionless;

[0119] Geometric mean porosity LG: Composed of the expected / variance ratio, dimensionless; ln(LG) is also dimensionless, 1+ln(⋅) is dimensionless;

[0120] Scale consistency slope variance Xslope: The variance of the regression slope, where the slope is dimensionless and the variance is also dimensionless;

[0121] Dimensionless;

[0122] Time ratio Time / time is dimensionless; ln(1+⋅) is dimensionless;

[0123] Multiplication and division structure: dimensionless × dimensionless × dimensionless. The multi-time dimension fractal continuous robustness index FCSms is dimensionless, which is consistent with common sense.

[0124] The fractal robustness assessment module also includes assessment units;

[0125] The evaluation unit calculates the multi-time-dimensional fractal robustness index FCSms in batches on the historical sample set of the same product category, obtains the empirical distribution, takes the 80th percentile as the initial multi-time-dimensional fractal robustness index threshold Fth, and re-evaluates it on a monthly rolling basis.

[0126] Next, a preliminary comparative evaluation is conducted between the real-time acquired multi-time-dimensional fractal robustness index FCSms and the multi-time-dimensional fractal robustness index threshold Fth to determine the robustness of the short video's traffic structure across multiple time scales, and triggering based on the preliminary comparative evaluation results; the specific evaluation content is as follows:

[0127] When the multi-time dimension fractal continuous robustness index FCSms is greater than or equal to the multi-time dimension fractal continuous robustness index threshold Fth, the current short video traffic structure is determined to be robust, and the transaction priority strategy is triggered at this time.

[0128] When the multi-time dimension fractal robustness index FCSms is less than the multi-time dimension fractal robustness index threshold Fth, the current short video traffic structure is determined to be unstable, and the conventional strategy should be applied.

[0129] In this embodiment, the fractal robustness analysis unit of the fractal robustness assessment module of the system first constructs three factors from the fractal dataset and multiplies and summarizes them to output a multi-time-dimensional fractal sustained robustness index FCSms: First, the slow-heating factor is composed of the heat release delay time TB and the reference delay time TR, which aims to distinguish between "immediate burst" and "slow-heating long tail". Physically, a larger TB / TR indicates that the heat release is more persistent and can maintain effective reach within a longer window. Second, the stability factor is composed of the scale consistency slope variance Xslope, which aims to constrain the trend consistency across minutes / hours / days. Physically, a smaller Xslope indicates greater stability. The more similar the scale curves are, the less likely they are to experience a "short surge followed by a sharp drop," thus reducing the risk of deployment. Thirdly, the complexity and spatial normalization factor is composed of the geometric mean and geometric mean porosity LG of the minute-level fractal dimension Dmin, the hour-level fractal dimension Dhour, and the day-level fractal dimension Dday. The purpose is to simultaneously characterize self-similar complexity and "window" sparsity. Physically, a higher fractal dimension indicates a richer structure, while a larger 1+ln(LG) indicates a more hollow flow that needs to be suppressed. The reason for using this structured combination is that in real-world scenarios, situations such as "short spikes," "holiday breakpoints," and "recommendation strategy switching" often lead to... Misjudgments based on a single indicator, such as judging short-lived peak videos as high-potential ones by only looking at peak values, can be effectively avoided by linking TB / TR, Xslope, and fractal dimension—LG. The evaluation unit then batch-calculates the multi-time-dimensional fractal robustness index FCSms on historical samples of the same category and constructs a multi-time-dimensional fractal robustness index threshold Fth using the 80th percentile. The purpose is to provide an objective, time-rolling admission line, physically defining a lower bound for "robustness" based on population statistical steady state. During online judgment, if the multi-time-dimensional fractal robustness index FCSms ≥ the multi-time-dimensional fractal robustness index threshold Fth, a trading priority strategy is directly triggered; otherwise, proceed... For example, in the conventional path, for a video that is "slow to heat up, has low porosity, and is stable across scales," the multi-temporal fractal robustness index FCSms is usually higher than the multi-temporal fractal robustness index threshold Fth, allowing for stable conversion within a longer window. Conversely, videos with "single-peak steep descent, high porosity, and large Xslope" will be compressed to prevent high-value resources from being occupied by instantaneous noise. The resulting benefits are: more noise-resistant admission criteria, reduced short-lived content mistakenly entering high windows, better matching of windows and rhythm, preventing slow-heating videos from being eliminated prematurely, and significantly reducing placement risks and resource waste, thereby improving the hit rate and overall ROI of subsequent priority ranking and placement window optimization stages.

[0130] Example 5: Please refer to Figure 1 Specifically: the priority ranking module includes a fractal survival deployment window analysis unit;

[0131] The fractal survival deployment window analysis unit obtains the effective fractal dimension Deff by averaging the minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, and day-level fractal dimension Dday after triggering the transaction priority strategy.

[0132] Simultaneously, the multi-time-dimensional fractal robustness index FCSms, output from the fractal robustness assessment module, is combined with the effective fractal dimension Deff, as well as the scale consistency slope variance Xslope, future exposure availability EA, and burst skewness SB to calculate and output the fractal survival deployment window score FSWms.

[0133] Fractal Survival Deployment Window Score (FSWms) is calculated and output using the following algorithm formula;

[0134] ;

[0135] In the formula, ln represents the natural logarithm;

[0136] Molecules: Reflect "realizability" With “cross-scale sustainable gains” FCSms, ;

[0137] Denominator: Suppressing "overall sparsity" "Cross-scale instability" "Peak outbreak risk" ;

[0138] The origin, modification and derivation of the formula: fractal dimension (box-counting), porosity / latency (Lacunarity) and long-range correlation / power-law tail idea (time series, survival analysis);

[0139] The three-scale fractal dimensions are synthesized into an effective fractal dimension Deff, and then... To approximate the "long-tail retention factor", when Deff→2, the exponent→0 and the retention factor→1, indicating a more stable tail; when Deff decreases, the retention factor<1; the logarithmic return of the marketable inventory is expressed as ln(1+EA) to avoid numerical imbalance caused by the scale difference.

[0140] By introducing "global sparsity", "cross-scale instability" and "peak morphology" into the denominator as three independent stability suppression terms, interpretability and monotonicity are maintained.

[0141] Structurally, artificial weights are avoided, and a product-based normalized "structural combination" is adopted, which works in conjunction with the multi-time-dimensional fractal robustness index FCSms: the multi-time-dimensional fractal robustness index FCSms determines the time-side entry and window, and the fractal survival deployment window score FSWms determines the resource-side ranking and loading intensity.

[0142] Analysis of dimensional consistency and rationality: FCSms, Deff, LG, Xslope, and SB are all dimensionless;

[0143] ln(1+EA) converts the scale quantity into a dimensionless quantity;

[0144] The numerator, denominator, and power term are all dimensionless, and the fractal survival deployment window score FSWms is dimensionless, while the dimensions are consistent.

[0145] The priority sorting module also includes sorting units;

[0146] The sorting unit generates a priority queue for all videos that trigger the transaction priority strategy, sorting them in descending order of their Fractal Survival Window (FSWms) scores. The specific sorting steps are as follows:

[0147] The 85th percentile P85 and 95th percentile P95 of the fractal survival delivery window score (FSWms) distribution based on historical short videos are used as step thresholds. The real-time fractal survival delivery window score (FSWms) is compared and evaluated with the step thresholds, and a priority queue is generated based on the secondary comparison and evaluation.

[0148] When the fractal survival delivery window score FSWms ≥ 95 quantile P95, it is classified as a secondary priced item and is loaded preferentially.

[0149] When 85 percentile P85 ≤ Fractal Survival Delivery Window Score FSWms < 95 percentile P95, it is classified as Level 1 premium loading, and regular loading is performed.

[0150] When the fractal survival delivery window score FSWms < 85 quantile P85, it is classified as a three-level valence and classified as a basic layer load;

[0151] Among them: Priority loading means that the video receives the first batch of ad resources at the start of the delivery window, and the loading ratio is relatively high. For example, the allocated exposure quota is ≥ 30% of the total available exposure. It enjoys the highest resource guarantee priority during resource competition. Even if resources are scarce, the exposure quota of the video will be guaranteed not to be reduced. In the bidding system, it can be given a higher starting bid coefficient or priority allocation of high-value traffic entrances, such as homepage recommendations and hot spot flow positions, to ensure that the highest-rated video receives a lot of exposure in the early stage of the delivery window, accelerate the formation of a stable viewing curve and improve conversion efficiency.

[0152] Regular loading means that after the delivery window starts, the ad resources are loaded at a regular cycle with a moderate loading ratio, such as an exposure quota of 10%-20% of the total available exposure. It is in the middle priority when resources are competitive. The exposure may be reduced appropriately when resources are tight, but it will not be completely eliminated. It is mainly placed in ordinary recommendation positions or regular traffic entrances, without occupying the core golden exposure positions, ensuring that this type of video gets a continuous but relatively gentle delivery rhythm. It is suitable for potential videos or content in the observation period.

[0153] Basic load means that exposure is only received within the delivery window when resources are plentiful. The load ratio is low, for example, the exposure quota accounts for less than 5% of the total available exposure. It is the first to be cut or delayed during resource competition, and there may be a long gap period. It is mainly placed in low traffic or marginal positions. The exposure value and potential revenue from click conversion are relatively low. This reduces the waste of resources on low-rated videos and concentrates resources on high-return videos. However, basic exposure is still retained to prevent missing out on individual long-tail videos that can "turn things around" later.

[0154] In this embodiment, after triggering the transaction priority strategy, the fractal survival deployment window analysis unit of the system first calculates the effective fractal dimension Deff using the minute-level fractal dimension Dmin, the hour-level fractal dimension Dhour, and the day-level fractal dimension Dday. This is used to characterize the "roughness / long-tail potential" of the cross-scale curve. Subsequently, the multi-time-dimensional fractal robustness index FCSms and the future exposure availability EA are incorporated into the numerator of the fractal survival deployment window score FSWms to reflect "sustainability × realizability". At the same time, the geometric mean porosity LG, the scale consistency slope variance Xslope, and the burst skewness SB are placed in the denominator to suppress "hollow complexity, cross-scale instability, and sharpness". The purpose of "peak risk" is to proactively lower the fractal survival window score (FSWms) when a scenario with a "short, sharp peak but seemingly ample future exposure availability (EA)" occurs. This is because the burst skewness (SB) and geometric mean porosity (LG) are relatively large, thus preventing the misallocation of high-value exposure to content that is about to decline. Conversely, if the multi-temporal fractal robustness indices (FCSms) of two videos are similar, and video A has a slightly lower future exposure availability (EA) but a smaller scale consistency slope variance (Xslope) and a geometric mean porosity (LG) close to 1, then video A has a higher fractal survival window score (FSWms), thus gaining priority in resource acquisition. Physically, this means that video A has a higher "stable and continuous time occupancy" than video A. "Short-term peaks" can be converted into effective reach within the window period; the ranking unit sets a tiered threshold based on the historical 85th percentile (P85) and 95th percentile (P95) of the Fractal Survival Scale (FSWms) delivery window score, classifying candidate videos into priority loading categories. This ensures that the delivery window receives ≥30% exposure from the start, seizing high-value entry points, followed by regular loading with 10%–20% exposure, and loading at a neutral and basic pace. Videos with less than or equal to 5% exposure are loaded only when resources are abundant. The significance lies in mapping continuous scores to stable and executable operational levels, resisting frequent repetitions caused by intraday fluctuations. The reason for adopting the "numerator gain × denominator suppression" structure and the "quantile ladder" strategy is because the platform... Common issues in operations, such as "holiday traffic disturbances, temporary adjustments to algorithm weights, and synchronous peaks caused by batch new product launches," can amplify misjudgments of single indicators. By using multi-factor normalization and tiered stratification of the Fractal Survival Window Score (FSWms), mis-targeting and over-targeting can be effectively avoided. The resulting benefits are: without changing the fixed targeting window determined by the multi-time-dimensional fractal robustness index (FCSms), the Fractal Survival Window Score (FSWms) achieves precise sorting of resource loading intensity and targeting order within the window, significantly reducing exposure waste and mismatch risks, improving the reach and conversion rate of high-potential content during key periods, thereby increasing overall ROI and inventory turnover efficiency.

[0155] Example 6: Please refer to Figure 1 Specifically: the delivery window optimization module includes a delivery optimization unit;

[0156] The delivery optimization unit determines a fixed delivery window length twindow based solely on the multi-time dimension fractal robustness index FCSms when the target short video passes the initial comparative evaluation. The specific determination form is: twindow = t0 × (1 + FCSms), where t0 represents the basic delivery duration set by the platform, which is set according to the short video platform strategy, such as 24 hours or 48 hours.

[0157] And register to form a fixed delivery window [ts, te], where ts represents the start time of the delivery window and te represents the end time of the delivery window;

[0158] The ad delivery optimization unit does not accept any additions or subtractions to the fixed ad delivery window length twindow by the Fractal Survival Ad Delivery Window Score (FSWms). When the fixed ad delivery windows [ts,te] of different videos overlap and platform resources are limited, the loading intensity and starting order of ad resources are allocated only within their respective fixed ad delivery windows [ts,te] based on the ranking results of the Fractal Survival Ad Delivery Window Score (FSWms), keeping the start and end boundaries of each fixed ad delivery window [ts,te] unchanged.

[0159] In this embodiment, after the target short video passes the initial comparative evaluation, the system's delivery optimization unit determines the delivery window length (twindow) solely based on the multi-time-dimensional fractal robustness index (FCSms) and registers it to form a fixed delivery window [ts, te]. The reason for "only considering the multi-time-dimensional fractal robustness index (FCSms) for timing and not considering the fractal survival delivery window score (FSWms) for timing changes" is to decouple "time decisions" from "resource decisions": On the time side, the multi-time-dimensional fractal robustness index (FCSms) reflects the stability and continuity across minutes / hours / days, ensuring the physical meaning of lifecycle matching (the higher the multi-time-dimensional fractal robustness index (FCSms), the more stable the curve and the more obvious the long tail, and a longer observable period should be given); on the resource side, within the predetermined window, the loading intensity and delivery sequence are allocated according to the fractal survival delivery window score (FSWms), avoiding scheduling jitter and resource mismatch caused by frequent window changes due to short-term noise, such as temporary peaks during holidays causing the fractal survival delivery window score (FSWms) to fluctuate. While the window can be extended, forcibly extending it would lock in inefficient periods as well. When different video windows overlap and resources are limited, loading is only performed within their respective [ts, te] based on the fractal survival delivery window score FSWms, keeping the boundaries unchanged. This avoids "crowding out" and chain rescheduling. For example, if t0 = 24 hours and a video's multi-time dimension fractal continuous robustness index FCSms = 1.2, then twindow = 52.8 hours. After the system locks [ts, te], it will not change the window due to subsequent slight fluctuations in the fractal survival delivery window score FSWms. Instead, it will only increase the starting order and exposure ratio within the window. The beneficial effects of this are: stable windows, predictable scheduling, reduced rescheduling costs and the risk of incorrect delivery; time matching is more in line with the real life cycle, reducing the structural waste of "short peak long window" or "long tail short window"; when resources conflict, the fractal survival delivery window score FSWms is used to finely allocate resources within a fixed window, improving the accessibility and conversion rate of high-potential content during key periods, thereby improving overall ROI and inventory turnover efficiency.

[0160] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart system for dynamic trading and multi-dimensional value assessment of short video traffic, characterized in that: include: Multi-dimensional traffic acquisition module: Collects video playback data from short video platforms in real time using acquisition tools, and transmits the video playback data to the cross-scale fractal analysis module; The multi-dimensional traffic acquisition module includes an acquisition unit; The acquisition unit collects video playback data from the short video platform at minute, hour, and day time scales by setting up acquisition tools on the short video platform. Cross-scale fractal analysis module: By preprocessing video playback data, a standardized dataset is obtained, and based on the standardized dataset, feature extraction is performed to obtain a fractal dataset; Fractal robustness assessment module: Calculates and outputs the multi-time-dimensional fractal robustness index FCSms based on the fractal dataset, and presets the multi-time-dimensional fractal robustness index threshold Fth for preliminary comparative assessment; The fractal robustness assessment module includes a fractal robustness analysis unit; The fractal robustness analysis unit extracts the heat burst delay time TB and reference delay time TR from the fractal dataset to form a slow-heating factor, constructs the scale consistency slope variance Xslope to form a stability factor, and constructs the complexity and spatial normalization factor by combining the minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, and day-level fractal dimension Dday with the geometric mean porosity LG. Then, the slow-heating factor, stability factor, and complexity and spatial normalization factor are multiplied and summarized to calculate and output the multi-time-dimensional fractal sustained robustness index FCSms. It is used to measure the complexity, stability, gap structure characteristics, and popularity growth pattern of a short video at different time scales, and reflects its long-term sustainability potential through a single value. The scale consistency slope variance Xslope represents the rate of change of video views at different time scales, minutes, hours and days. First, the slope sequence of the view change is calculated at each time scale. Then compare the consistency of slope changes at different scales, and take the variance value as the scale consistency slope variance Xslope. The "Hotness Explosion Delay Time (TB)" refers to the time interval from the video's release to the first significant increase in views, which is considered a "hotness explosion." Reference delay time (TR) represents the relative difference in delay time of the target video compared to the average delay time of the surge in popularity of similar videos. It is used as a reference quantity in fractal analysis. Geometric mean porosity LG is obtained by averaging minute-level porosity Amin, hour-level porosity Ahour, and day-level porosity Aday. Minute-level porosity Amin: represents the proportion of time in the entire minute-level time series where the number of plays is below zero, and the distribution density of these low-value intervals on the time axis; Hourly porosity Ahour: This represents the distribution and proportion of low play count intervals within an hourly timescale, also based on the play count curve. Aday porosity: Represents the percentage of empty windows in the number of plays over a long period of time on a day-scale time scale. Priority ranking module: Triggers trading priority strategies through preliminary comparison and evaluation, calculates and outputs fractal survival deployment window score FSWms based on the multi-time dimension fractal robustness index FCSms, and prioritizes video trading based on the fractal survival deployment window score FSWms. The ad delivery window optimization module determines the ad delivery window length (twindow) based on the multi-time dimension fractal robustness index (FCSms) and allocates ad resources and delivery time based on the ranking results of the fractal survival ad delivery window score (FSWms). The delivery window optimization module includes a delivery optimization unit; The delivery optimization unit determines a fixed delivery window length twindow based solely on the multi-time dimension fractal robustness index FCSms when the target short video undergoes preliminary comparative evaluation. Specifically, twindow = t0 × (1 + FCSms), where t0 represents the basic delivery duration set by the platform. And register to form a fixed delivery window [ts, te], where ts represents the start time of the delivery window and te represents the end time of the delivery window; The ad delivery optimization unit does not accept any additions or subtractions to the fixed ad delivery window length twindow by the Fractal Survival Ad Delivery Window Score (FSWms). When the fixed ad delivery windows [ts,te] of different videos overlap and platform resources are limited, the loading intensity and starting order of ad resources are allocated only within their respective fixed ad delivery windows [ts,te] based on the ranking results of the Fractal Survival Ad Delivery Window Score (FSWms), keeping the start and end boundaries of each fixed ad delivery window [ts,te] unchanged.

2. The intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic according to claim 1, characterized in that: The data acquisition tool includes a data acquisition terminal and an acquisition control program; The data acquisition terminal is deployed on the backend data interface node of the short video platform, located between the platform's data exchange layer and business processing layer, to acquire raw playback data that has not been aggregated or compressed. The data acquisition terminal has data buffering, real-time data packaging, and network communication functions. The acquisition control program runs within the data acquisition terminal, calls the platform's open API, collects log streams from the short video platform, and extracts video playback data and additional parameters from the log streams. The video play count data includes a unique video identifier, a corresponding timestamp, scale consistency slope variance Xslope, popularity burst delay time TB, reference delay time TR, future exposure availability EA, burst skew SB, minute-level play count sequence Pmin, hour-level play count sequence Phour, and day-level play count sequence Pday. The minute-level playback sequence Pmin has a sampling step of 1 minute and a box counting scale r∈{1,2,4,8,16} minutes; The hourly playback sequence Phour has a sampling step of 1 hour and a box counting scale r∈{1,2,4,8} hours; The sampling step size for the daily play count sequence Pday is 1 day, and the box counting scale is r∈{1,2,4} days. Future Exposure Availability (EA) represents the predicted total exposure a video may receive in the platform's recommendation algorithm over a future period. It is based on the ARIMA model combined with historical playback trends, current platform recommendation weights, and user activity models. The burst skewness (SB) indicates the symmetry of the video popularity burst curve, describing the difference in play count distribution before and after the burst.

3. The intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic according to claim 2, characterized in that: The multi-dimensional traffic acquisition module also includes a transmission unit; The transmission unit packages video playback data into data packets with unique video identifiers and corresponding timestamps according to time scales, and sends them to the cross-scale fractal analysis module in an encrypted manner via the high-speed message queue transmission protocol MQTT. After receiving the data packets, the cross-scale fractal analysis module reconstructs the order of the data packets and generates a receipt confirmation acknowledgment.

4. The intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic according to claim 3, characterized in that: The cross-scale fractal analysis module includes a data preprocessing unit, which is used to preprocess video playback data to generate a standardized dataset; The preprocessing includes time synchronization processing, unit unification processing, and data structuring and storage. The time synchronization process aligns the minute-level playback sequence Pmin, hour-level playback sequence Phour, and day-level playback sequence Pday in the video playback data with timestamps according to a global time base, and fills in missing time data points using an interpolation algorithm, wherein the interpolation method is linear interpolation. The dimensionless processing involves applying the Z-Score standardization method to all parameters in the video playback data to convert parameters with different physical dimensions into dimensionless values ​​with a mean of 0 and a standard deviation of 1. The data structuring and storage involves grouping and storing the video playback data, after unifying the units of measurement, into a standardized dataset according to time scale and parameter type, and adding a unique video identifier and corresponding timestamp to each record.

5. The intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic according to claim 4, characterized in that: The cross-scale fractal analysis module also includes a fractal feature extraction unit, which is used to extract features from the standardized dataset to obtain a fractal dataset; The fractal dataset includes minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, day-level fractal dimension Dday, geometric mean porosity LG, minute-level porosity Amin, hour-level porosity Ahour, day-level porosity Aday, scale consistency slope variance Xslope, heat burst delay time TB, reference delay time TR, future exposure availability EA, and burst skewness SB.

6. The intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic according to claim 5, characterized in that: The fractal robustness assessment module includes an assessment unit; The evaluation unit calculates the multi-time-dimensional fractal robustness index FCSms in batches on the historical sample set of the same product category, obtains the empirical distribution, takes the 80th percentile as the initial multi-time-dimensional fractal robustness index threshold Fth, and re-evaluates it on a monthly rolling basis. Next, a preliminary comparative evaluation is conducted between the real-time acquired multi-time-dimensional fractal robustness index FCSms and the multi-time-dimensional fractal robustness index threshold Fth to determine the robustness of the short video's traffic structure across multiple time scales, and triggering based on the preliminary comparative evaluation results; the specific evaluation content is as follows: When the multi-time dimension fractal continuous robustness index FCSms is greater than or equal to the multi-time dimension fractal continuous robustness index threshold Fth, the current short video traffic structure is determined to be robust, and the transaction priority strategy is triggered at this time. When the multi-time dimension fractal robustness index FCSms is less than the multi-time dimension fractal robustness index threshold Fth, the current short video traffic structure is determined to be unstable, and the conventional strategy should be applied.

7. The intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic according to claim 6, characterized in that: The priority ranking module includes a fractal survival deployment window analysis unit; The fractal survival deployment window analysis unit obtains the effective fractal dimension Deff by averaging the minute-level fractal dimension Dmin, hour-level fractal dimension Dhour, and day-level fractal dimension Dday after triggering the transaction priority strategy. Simultaneously, the multi-time-dimensional fractal robustness index FCSms, output from the fractal robustness assessment module, is combined with the effective fractal dimension Deff, as well as the scale consistency slope variance Xslope, future exposure availability EA, and burst skewness SB to calculate and output the fractal survival deployment window score FSWms.

8. The intelligent system for dynamic trading and multi-dimensional value assessment of short video traffic according to claim 7, characterized in that: The priority sorting module also includes a sorting unit; The sorting unit sorts all videos that trigger the transaction priority strategy by generating a priority queue in descending order of the Fractal Survival Window Score (FSWms). The specific sorting steps are as follows: The 85th percentile P85 and 95th percentile P95 of the fractal survival delivery window score (FSWms) distribution based on historical short videos are used as step thresholds. The real-time fractal survival delivery window score (FSWms) is compared and evaluated with the step thresholds, and a priority queue is generated based on the secondary comparison and evaluation. When the fractal survival delivery window score FSWms ≥ 95 quantile P95, it is classified as a secondary priced item and is loaded preferentially. When 85 percentile P85 ≤ Fractal Survival Delivery Window Score FSWms < 95 percentile P95, it is classified as Level 1 premium loading, and regular loading is performed. When the fractal survival delivery window score FSWms < 85 quantile P85, it is classified as a three-level priced system and classified as a basic layer load.

Images