Short video traffic futures transaction and value evaluation system

Abstract

The invention relates to the technical field of value evaluation, in particular to a short video traffic futures transaction and value evaluation system, which realizes accurate comparison and matching of cross-platform traffic through deep mining of periodic feature offset, accurate recognition of click intensity difference between platforms and standardized adjustment, and improves the accuracy of value evaluation of the short video traffic futures transaction and value evaluation. The value evaluation has higher applicability and operability, the periodic structure of the data is analyzed through the causal forest algorithm, potential influence factors in flow fluctuation can be deeply identified, the predictability of short video flow and the stability of future value evaluation are improved, and through accurate judgment of the long and short-term memory network in conversion deviation, the value evaluation accuracy is improved. According to the method, the operation response characteristics of the content can be in butt joint with market demands, quantitative prediction of future traffic value is realized, a feasible decision basis is provided for a platform and a content creator, the decision quality of market participants is improved, and the efficiency and reliability of traffic futures trading are optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of value evaluation, and particularly to a short video traffic futures trading and value evaluation system. Background Art

[0002] The technical field of value evaluation aims to objectively and accurately measure and predict the economic value of assets, resources, products or services through systematic and quantitative means, so as to support various commercial and financial activities such as pricing, investment decision-making, risk control, and resource allocation, help the entity reasonably judge the current and potential value of assets, and then optimize trading strategies and decision-making behaviors.

[0003] A short video traffic futures trading and value evaluation system aims to provide a predictable basis for the future monetization of traffic by accurately evaluating the market value of short video traffic. Through the analysis of real-time data, historical performance and market trends of short video traffic, the market value of traffic is evaluated, and through the way of futures trading, stakeholders such as content creators and platform operators can expect and trade the value of future traffic.

[0004] Although the prior art has insufficient analysis of traffic fluctuations in the short video platform, resulting in the evaluation results being difficult to reflect the periodic changes and potential market structural adjustments in traffic fluctuations, and failing to effectively capture the jump segments and interaction peaks of traffic growth. Therefore, in actual operation, it is impossible to accurately locate the key traffic growth periods, resulting in large errors in traffic value prediction. Moreover, the existing methods cannot effectively compare the traffic differences between platforms across platforms, lack accurate modeling of periodic fluctuation characteristics, making the existing evaluation system unable to comprehensively reflect the changes in platform structure and the dynamic fluctuations of market demand, lacking sufficient flexibility and accuracy, and affecting the reliability of traffic futures trading and the rationality of market decision-making. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies in the prior art and propose a short video traffic futures trading and value evaluation system.

[0006] To achieve the above purpose, the present invention adopts the following technical solution: A short video traffic futures trading and value evaluation system includes: A trading period construction module: According to the content release records of a specified account in the short video platform, extract the release time difference and the growth jump value, screen the jump segments and extract the interaction peaks and popularity tags, combine the active time blocks, and establish a set of platform trading intervals; Period offset recognition module: Based on the block data of the platform trading interval set, analyze the click perturbation amplitude and conversion jump distance, extract the jump-up distance and center of gravity position, and use the causal forest algorithm to screen the periodic jump structure difference sections to obtain the periodic feature offset clusters; Platform structure mapping module: Based on the periodic feature offset clusters, compare the platform structure feature strength and center of gravity displacement, perform standardized mapping adjustment, measure the click intensity and structure peak position difference between platforms, unify the scale of the traffic feature vectors, and establish a cross-platform equivalent feature set; Value response derivation module: Based on the jump-up gradient and conversion center of gravity position synchronously inherited in the cross-platform equivalent feature set, use the long short-term memory network, and identify the high-response feature combinations by judging the distribution of conversion deviation values under the segmented intervention groups to establish a content operation response characteristic map; Contract value quantification module: Based on the content operation response characteristic map, extract the product of the jump-up frequency and conversion displacement to construct a pricing factor, correct the valuation interval, and establish the futures contract valuation output result.

[0007] As a further solution of the present invention, the trading time period construction module includes: Release time extraction sub-module: Based on the content release records of a specified account in the short video platform, calculate the difference in the upload time points of adjacent contents, sort the difference sequence in ascending order and divide the high-frequency difference intervals, construct non-uniform time period boundaries according to the interval mutation nodes, and perform time period label numbering processing to generate a time interval distribution sequence; Behavior jump recognition sub-module: Based on the time interval distribution sequence, extract the amplitude changes of the cumulative difference of click values, the difference in the number of likes, and the total number of comments in each time period, synchronously compare each amplitude change with the range of the previous time period, and perform the operations of generating a difference matrix between results and screening marked values to obtain a content jump activity index set; Active block combination sub-module: Based on the content jump activity index set, extract the time points where the jump activity values are located, perform continuous segment construction judgment on the time points and splice the time periods, combine the change values of the appearance density of the label keywords in each time period to screen the label active areas, perform the intersection operation of the two and mark the interval boundaries to establish a platform trading interval set.

[0008] As a further solution of the present invention, the period offset recognition module includes: Structure difference measurement sub-module: Based on the platform trading interval set, extract the total number of clicks and the total number of interactions in each trading section, calculate the difference between adjacent click values with time increment and convert it into a trend slope, identify the change points of the appearance frequencies of likes and comments within a unit time, measure the time distance value from them to the click jump-up points and establish a sequence to obtain a periodic behavior structure difference group; Center of gravity deviation extraction sub-module: Based on the periodic behavior structure difference group, extract the center point position of the high-frequency value section of the behavior in the structure sequence, measure the distance between the centers of gravity of the front and rear high-frequency sections in chronological order, perform normalization and stratification processing on the distance values, generate position change labels according to the relative displacement, integrate all offset records and sort and number them to obtain the behavior response center of gravity offset sequence; Offset cluster screening sub-module: Based on the behavior response center of gravity offset sequence, use the causal forest algorithm to calculate the coincidence ratio of the center of gravity displacement value between adjacent cycles and the trend direction of the structure difference, set the judgment threshold according to the proportion of the trend direction consistency, group and aggregate the cycles that meet the conditions, number them in sequence and construct the boundary ranges of each group to obtain the periodic feature offset clusters.

[0009] As a further solution of the present invention, the causal forest algorithm is in accordance with the formula:

[0010] Where: represents the proportion of the consistency of the trend direction between the i th cycle and the th cycle, represents the angle between the center of gravity displacement direction of the th cycle and the structure difference trend, represents the angle between the center of gravity displacement direction of the th cycle and the structure difference trend, represents the center of gravity displacement distance of the th cycle, represents the center of gravity displacement distance of the th cycle, represents the smoothness of the center of gravity displacement trajectory of the th cycle, represents the smoothness of the center of gravity displacement trajectory of the th cycle, represents a constant for angle normalization, represents the weight coefficient of the center of gravity displacement distance, represents the weight coefficient of the smoothness of the center of gravity displacement trajectory, represents the adjustment coefficient of the normalization constant.

[0011] As a further scheme of the present invention, the center point position of the high-frequency value segment of the behavior in the structural sequence is extracted. The high-frequency value segment of the behavior is the time period in the periodic structure where the number of clicks, likes and comments continue to appear in a concentrated manner. First, a sliding window is used to scan the periodic structure difference sequence, and the continuous time period in which the local statistical value exceeds the set behavior intensity threshold is recorded. The weighted central time point of the behavior frequency in the segment is calculated to obtain the true center of gravity of the actual behavior aggregation trend. Then, the center of gravity distance between the front and rear high-frequency bands is normalized, and the interval length is used as the basis for layered archiving. It is marked as positive offset, negative offset and no offset according to the displacement direction to generate a position change label.

[0012] As a further solution of the present invention, the platform structure mapping module includes: Structural parameter extraction submodule: Based on the periodic feature offset cluster, the extreme value difference of the total click value in each platform within a fixed time length is extracted, the center of gravity point position is calibrated by minute segment and the time position frequency is summarized, the line segment length between the start and end points of the span is calculated and the time vector is assembled to generate a platform feature parameter set; Difference comparison and calibration submodule: Based on the platform characteristic parameter set, the click difference between the target platform and the source platform is calculated in absolute value in the time vector sequence and the difference sign direction is determined, the displacement between the center of gravity points is counted and the position of the anisotropic change sequence is marked, the structural amplitude interval is divided and grouped by platform number, and the structural characteristic deviation sequence is obtained; Scale unified reconstruction submodule: Based on the structural feature deviation sequence, amplitude line segment compression and stretching are performed according to the click difference ratio of each group, the displacement direction of the time reference point is adjusted according to the center of gravity displacement sign, the offset channel vector is extracted after splicing the vectors within the group, platform intersection screening and reorganization alignment are performed, and a cross-platform equivalent feature set is established.

[0013] As a further solution of the present invention, the value response derivation module includes: Intervention parameter division submodule: Based on the cross-platform equivalent feature set, the three parameters of tag quantity, release time, and duration are aggregated and divided into segments according to the account dimension, and the total click difference and conversion rate change interval of each type of video are calculated and classified into the parameter intersection set, and a number mapping relationship is established and associated with sample groups to generate intervention behavior classification groups; Response matching extraction submodule: Based on the intervention behavior classification group, the overlapping areas of the click growth segments, conversion rate increase points and comment abnormality occurrence time nodes of the videos in the group are compared, the overlapping time periods of the three types of change positions are extracted, the response time window is determined according to the sequence distribution, and the content with synchronous jump characteristics is screened out to obtain the intervention response feature set; Structural combination judgment submodule: Based on the intervention response feature set, a long short-term memory network is used to pair the click difference of each group of jump feature points with the direction of the corresponding intervention parameter curve, and path reconstruction and superposition frequency statistics are performed on the group with consistent directions. High-frequency co-occurrence structures are extracted and numbering vectors are constructed. The combined structure sequence is integrated to generate an archive set, and a content operation response characteristic map is established.

[0014] As a further solution of the present invention, the long short-term memory network is according to the formula:

[0015] in: Indicates that at the current time step The obtained structural combination state vector, Indicates that the jump feature point is at time step The input vector is Represents the time step The hidden state of , as the information memory of the historical structure combination path, Represents the mapping weight matrix between the input features and the current state, Represents the weight matrix from the previous hidden state to the current state, represents the bias term in the state update, represents the click difference calibration coefficient, Represents the normalized fluctuation of the click difference between the jump feature points, represents the response density adjustment factor, Indicates the concentration of response points per unit time, represents the path direction continuity weight coefficient, represents the directional consistency score between three consecutive jump points, represents the structural rhythm modulation coefficient, represents the standard deviation of the time interval between adjacent structure combinations, Represents the activation function.

[0016] As a further solution of the present invention, the parameter intersection set is included, and the three parameters of the number of video tags, release time and video duration under the account dimension are used as division criteria. First, the video data is segmented within each parameter dimension, and the total click-through rate and conversion rate change values ​​of the corresponding videos in each segment are counted. In the three-dimensional parameter space, the overlapping areas with significant changes in click-through rate and conversion rate are found, that is, the cross-segments with active intervention responses in multiple dimensions at the same time are included in the parameter intersection set as key intervention areas.

[0017] As a further solution of the present invention, the contract value quantification module includes: Frequency calculation sub-module: Based on the content operation response characteristic map, extract the indexes of the points with sudden increase in the number of clicks in each video, extract consecutive jump points with an interval less than a fixed threshold in chronological order, calculate the number of each group of jump points and the average interval length, mark the sample numbers and synchronously record the frequency values, and generate the statistical indicators of the jump increase period; Offset construction sub-module: Based on the statistical indicators of the jump increase period, extract the time points of the positions of the number of likes and the high-frequency segments of the corresponding conversion numbers, calculate the absolute interval value between the two time points, perform a product operation item by item according to the interval value and the frequency value, generate a product mean vector grouped by content numbers, and obtain the pricing adjustment structure factor after dividing the standard interval range; Valuation output sub-module: Based on the pricing adjustment structure factor, extract the total click value and the total amplitude of the conversion rate change of each piece of content during the trading cycle, replace the upper and lower limits of the original valuation range after performing multiplication adjustment according to the adjustment factor respectively, construct a new click valuation bandwidth, re-encode and map according to the content numbers, and establish the valuation output result of the futures contract.

[0018] Compared with the prior art, the advantages and positive effects of the present invention are as follows: 1. In the present invention, through the in-depth excavation of the offset of periodic characteristics, the click intensity differences between platforms are accurately identified, and through standardized adjustment, the accurate comparison and matching of cross-platform traffic are realized, making the value evaluation more applicable and operable; 2. In the present invention, by analyzing the periodic structure of data through the causal forest algorithm, the potential influencing factors in traffic fluctuations can be deeply identified, improving the predictability of short-video traffic and the stability of future value evaluation; 3. In the present invention, through the accurate judgment of the conversion deviation by the long short-term memory network, the operation response characteristics of the content can be connected with the market demand, realizing the quantitative prediction of the future traffic value, providing a practical decision-making basis for platforms and content creators, and improving the decision-making quality of market participants and optimizing the efficiency and reliability of traffic futures trading. Brief Description of the Drawings

[0019] Figure 1 is the system flow chart of the present invention; Figure 2 is the flow chart of the present invention; Figure 3 is the schematic diagram of the system framework of the present invention. Detailed Embodiment

[0020] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0021] Please refer to Figure 1 and Figure 2 , the present invention provides a technical solution: a short video traffic futures trading and value evaluation system includes: Trading period construction module: According to the content release records of a specified account in the short video platform, extract the release time difference and the amplitude jump value, screen the jump segments and extract the interaction peaks and popularity tags, combine the active time blocks, and establish a platform trading interval set; Period offset identification module: Based on the block data of the platform trading interval set, analyze the click perturbation amplitude and the conversion jump distance, extract the jump-up distance and the center of gravity position, and use the causal forest algorithm to screen the period jump structure difference segments to obtain the periodic feature offset clusters; Platform structure mapping module: Based on the periodic feature offset clusters, compare the platform structure feature strength and the center of gravity displacement, perform standardized mapping adjustment, measure the click strength and the structure peak position difference between platforms, unify the scale of the traffic feature vectors, and establish a cross-platform equivalent feature set; Value response derivation module: Based on the jump-up gradient and the conversion center of gravity position synchronously inherited in the cross-platform equivalent feature set, use the long short-term memory network, and identify the high-response feature combinations by judging the distribution of the conversion deviation values under the segmented intervention groups to establish a content operation response characteristic map; Contract value quantification module: Based on the content operation response characteristic map, extract the product of the jump increase frequency and the conversion displacement to construct a pricing factor, correct the valuation interval, and establish the futures contract valuation output result.

[0022] Please refer to Figure 3 , the trading period construction module includes: Release time extraction sub-module: Based on the content release records of a specified account in the short video platform, calculate the time difference between the upload time points of adjacent contents, sort the difference sequence in ascending order and divide the high-frequency difference intervals, construct the non-uniform time period boundaries according to the interval mutation nodes, and perform the time period label numbering process to generate the time interval distribution sequence; Behavior jump identification sub-module: Based on the time interval distribution sequence, extract the amplitude changes of the cumulative difference of click values, the difference in the number of likes, and the total number of comments in each time period, synchronously compare the ranges of each amplitude change with the previous time period, and perform the difference matrix generation and marked value screening operations between the results to obtain the content jump active index set; Active block combination sub-module: Based on the content jump active index set, extract the time points where the jump active values are located, perform the continuous segment construction judgment on the time points and splice the time periods, combine the change values of the appearance density of the label keywords in each time period to screen the label active areas, perform the intersection operation of the two and mark the interval boundaries to establish a platform trading interval set; Publication time extraction sub-module: Based on the content publication records of a specified account on a short video platform, the datetime.strptime function and timedelta object in the Python standard library datetime module are used to calculate the difference between the upload time points of adjacent content. The time format is set to "%Y-%m-%d %H:%M:%S", and the difference result is converted to the second-level unit. The numpy.diff method in NumPy is used to perform a first-order difference operation on the time series to generate a difference sequence. The numpy.sort function in numpy is used to sort the difference sequence in ascending order. And through the KDE kernel density estimation method, specifically the scipy.stats.gaussian_kde function, the estimation of the high-density segment in the sorting result is fitted to divide the high-frequency difference interval. The kernel function type is set to the Gaussian kernel, and the bandwidth parameter is automatically generated using the Silverman estimation method. Subsequently, the scipy.signal.find_peaks function is used to identify the mutation extreme points in the kernel density curve. The height parameter is set to 1.2 times the average value of the density curve, and the distance parameter is set to the window length with a minimum interval of 3. The non-uniform time period boundary is constructed based on the mutation point position, and the label numbering process is carried out according to the start and end time positions of the constructed sections. Finally, a time interval distribution sequence is generated; Behavior jump recognition sub-module: Based on the time interval distribution sequence, the groupby operation and agg aggregation function in the Pandas library are used to aggregate the click value, like frequency, and total comment volume within each time period. The sum function is used to generate the cumulative difference of the click value, the difference of the like frequency, and the difference of the total comment volume fields respectively. Subsequently, the numpy.ptp function in NumPy is used to perform the range calculation operation on the variable amplitude values of each time period. The corresponding extreme difference values of each type of behavior amplitude and the previous time period are compared item by item. The difference calculation is performed by constructing a two-dimensional difference matrix. The numpy.subtract function in NumPy is used to calculate the variable amplitude difference value element by element and mark the mutation points in the form of a boolean matrix. The position where the boolean value is True represents the significant variable amplitude point. Further, the numpy.where function in NumPy is called to filter the index points that meet the threshold criteria. The threshold criteria are set to the historical mean plus 1.5 times the standard deviation. The threshold calculation is implemented using the pandas.mean and std functions. Finally, the content jump activity index set is output; Active Block Combination Sub-module: Based on the content jump activity index set, extract the time points where all jump activity values are located. Use the itertools.groupby method and a custom window judgment function in Python to determine whether the interval between adjacent time points is within five minutes. Those that meet the conditions are constructed into continuous segments, and multiple consecutive time points that meet the conditions are merged using the time series concatenation function pandas.concat to generate time period blocks. Further extract the content label fields within each time period, use the TF-IDF keyword density extraction method, call the sklearn.feature_extraction.text.TfidfVectorizer function, set max_features to 200, ngram_range to (1, 2), and call the fit_transform method to vectorize the text to obtain the density value change vector of the label keywords. Then use the numpy.diff function of NumPy to calculate the change values of the keyword density sequence, use the scipy.signal.argrelextrema function to find local maxima for the change value sequence, take the intersection interval using the numpy.intersect1d method of NumPy for the jump blocks and density fluctuation blocks, perform interval boundary marking according to the start and end time points of the intersection result, and finally establish the platform trading interval set.

[0023] Please refer to Figure 3 , the cycle offset recognition module includes: Structure Difference Measurement Sub-module: Based on the platform trading interval set, extract the total number of clicks and the total number of interactions within each trading section, calculate the difference between adjacent click values under time increment and convert it into a trend slope, identify the change points in the occurrence frequencies of likes and comments per unit time, measure the time interval value between them and the click jump points and establish a sequence, and obtain the cycle behavior structure difference group; Center of Gravity Deviation Extraction Sub-module: Based on the cycle behavior structure difference group, extract the center point positions of the high-frequency value sections of the behavior in the structure sequence, measure the distance between the centers of gravity of the front and rear high-frequency sections in chronological order, perform normalization and stratification processing on the distance values, generate position change labels according to the relative displacement, integrate all offset records and sort and number them to obtain the behavior response center of gravity deviation sequence; Offset Cluster Screening Sub-module: Based on the behavior response center of gravity deviation sequence, use the causal forest algorithm to calculate the coincidence ratio of the center of gravity displacement value between adjacent cycles and the trend direction of the structure difference, set a judgment threshold according to the proportion of trend direction consistency, group and aggregate the cycles that meet the conditions, number them in sequence and construct the boundary ranges of each group to obtain the periodic feature offset cluster; Structural difference sub-module: Based on the platform trading interval set, extract the click data and interaction data within each trading section. Use the groupby and agg methods in Pandas to perform the sum function aggregation on the click values and interaction frequencies in each section respectively to generate the total click volume and total interaction volume. Use the numpy.diff method in NumPy to perform a first-order difference operation on the total click volume sequence to obtain the difference between adjacent click values. Call the numpy.gradient method in NumPy to calculate the derivative of the difference sequence and obtain the trend slope. Set the window width to 3 and use the first-order derivative with an edge-order of 1 for boundary processing. Use the scipy.signal.find_peaks function to identify the extreme values of the like sequence and comment frequency sequence respectively. Set the height parameter to the mean of the corresponding behavior sequence plus 1.2 times the standard deviation, and set the distance parameter to 5. Match the recognition results with the time positions of the click jump points through the combination function of numpy.abs and numpy.argmin in NumPy. Calculate the time interval values between each group of like and comment high-frequency points and click jump points and construct them into a time difference sequence. Finally, output the periodic behavior structure difference group; Center of gravity deviation extraction sub-module: Based on the periodic behavior structure difference group, use the sliding window scanning method combined with the weighted average algorithm to perform a moving process with a window width of 5 on the structure difference sequence and sort and extract the high-frequency value sections according to the internal difference amplitude of each section. Call the numpy.average method in NumPy, use the difference amplitude as the weight and the corresponding time index as the input vector to calculate the weighted time mean of each high-frequency section as the center point position of the section. Arrange the center of gravity points of each high-frequency section in chronological order. Call the numpy.diff function in NumPy to perform a difference on the distance values between adjacent center of gravity points to obtain the time displacement. Use the fit_transform method of the preprocessing.MinMaxScaler class in the scikit-learn library to perform a normalization operation on the difference sequence, set the feature_range parameter to (0, 1). Divide the displacement into three categories according to the normalized values, and set the interval boundaries to less than 0.33, 0.33 to 0.66, and greater than 0.66 respectively. The corresponding labels are marked as 0, 1, 2. Use the numpy.argsort function in NumPy to sort and number all the offset records according to the time position. Finally, generate the behavior response center of gravity offset sequence; Offset cluster screening sub-module: Based on the behavioral response center-of-gravity offset sequence, using the CausalForestDML model in the DoWhy causal inference library, with the center-of-gravity displacement value of each period and the slope value in the structural difference sequence as input features, call the model.fit method, pass in the parameter model_y as RandomForestRegressor(n_estimators=100, max_depth=4), model_t as RandomForestRegressor(n_estimators=50, max_depth=3), set the inference using the empirical standard, calculate the structural trend direction label for each period sample, use the numpy.sign function in NumPy to extract the positive and negative values of the trend direction and convert them into a one-dimensional direction vector, perform direction consistency matching with the previous period for each period segment, use the combination operation of numpy.equal and numpy.sum in NumPy to obtain the number of consistent trends within the period window, set the direction consistency threshold to 70% of the sliding window period number, that is, when the window width is 5, the threshold is set to 4, group and merge the consistent direction periods and number them sequentially using the enumerate function in Python, use the numpy.min and numpy.max functions in NumPy to extract the start and end boundaries of the numbered periods, and finally construct the periodic feature offset cluster.

[0024] Please refer to Figure 3 , the causal forest algorithm, according to the formula:

[0025] Where: represents the i ratio of the trend direction consistency between the period and the period, represents the angle between the center-of-gravity displacement direction of the period and the structural difference trend, represents the angle between the center-of-gravity displacement direction of the period and the structural difference trend, represents the center-of-gravity displacement distance of the period, represents the center-of-gravity displacement distance of the period, represents the smoothness of the center-of-gravity displacement trajectory of the period, represents the smoothness of the center-of-gravity displacement trajectory of the represents a constant used for angle normalization, The weight coefficient representing the center-of-gravity displacement distance The weight coefficient representing the smoothness of the center-of-gravity displacement trajectory The adjustment coefficient representing the normalization constant; Execution process: First, and represent the angle between the center-of-gravity displacement direction of the th cycle and the th cycle and the trend of the difference in short-video content traffic. By analyzing the trend direction of video traffic changes within each cycle, the consistency of the traffic change trend of short-video content over time is measured. Then and respectively represent the center-of-gravity displacement distance of the th cycle and the th cycle, which reflects the degree of traffic change of the short video within the cycle and measures the fluctuation range of short-video traffic. Then and are the smoothness of the center-of-gravity displacement trajectory of the th cycle and the th cycle, aiming to reflect the stability of traffic changes. A smaller smoothness indicates a drastic change in short-video traffic, while a larger smoothness indicates a stable traffic change. Through weighted summation using a formula, , and are respectively the weight coefficients of the center-of-gravity displacement distance, the smoothness of the center-of-gravity displacement trajectory, and the angle normalization constant. After weighting and normalization, the proportion of the trend consistency of traffic changes between two cycles is calculated, which is used to evaluate the stability of the traffic value of short-video content at different time periods, provide data support for futures trading decisions, help predict future traffic trends, and guide traders in short-video traffic futures trading.

[0026] Please refer to Figure 3 , extract the central point position of the high-frequency value section of the behavior in the structural sequence. The high-frequency value section of the behavior is the time period in the periodic structure where the click-through rate, the number of likes, and the number of comments continuously appear concentrated. First, use a sliding window to scan the difference sequence of the periodic structure, record the continuous time periods where the local statistical value exceeds the set behavior intensity threshold, calculate the weighted central time point of the behavior frequency within the section, obtain the true center of gravity of the actual behavior aggregation trend, and then normalize the center-of-gravity distance between the front and back high-frequency sections, perform hierarchical filing based on the interval length as the benchmark, and label it as positive offset, negative offset, and no offset according to the displacement direction to generate a position change label.

[0027] Please refer to Figure 3 , the platform structure mapping module includes: Structure parameter extraction sub-module: Based on the periodic feature offset clusters, extract the extreme value difference of the total click value within each platform in a fixed time period, calibrate the position of the center of gravity point in minute segments and summarize the time position frequencies, calculate the line segment length between the start and end points of the span and assemble the time vector to generate a set of platform feature parameters; Difference comparison and calibration sub-module: Based on the set of platform feature parameters, calculate the absolute value of the click difference between the target platform and the source platform in the order of the time vector and judge the sign direction of the difference, count the displacement amount between the center of gravity points and mark the positions of the out-of-direction change sequences, divide the structural amplitude change intervals and group them by platform number to obtain the structural feature deviation sequence; Scale unified reconstruction sub-module: Based on the structural feature deviation sequence, perform amplitude line segment compression and stretching according to the ratio of each group of click differences, adjust the displacement direction of the time reference point according to the center of gravity displacement sign, extract the offset channel vector after splicing the vectors within the group, perform platform intersection screening and recombination alignment to establish a cross-platform equivalent feature set; Structure parameter extraction sub-module: Based on the periodic feature offset clusters, use the groupby and resample methods in the Pandas library combined with NumPy functions to process the click data within each platform in a fixed time window, set the resample frequency to '1min', use the agg method to calculate the maximum and minimum values of the total click value per minute, call the numpy.ptp function of NumPy to extract the extreme value difference of each minute segment, use the scipy.signal.find_peaks function to identify the peak interval in the extreme value difference sequence, set the parameters as height to the average value of the difference plus 0.5 times the standard deviation, distance set to 10, identify the time index corresponding to the position of the center of gravity point, count the time index frequencies through the collections.Counter class and construct a frequency table, use the numpy.diff method in NumPy to perform a difference operation on the start and end points in the time index and obtain the line segment length, the time vector is assembled through the numpy.linspace function in NumPy, set the start and end points to the boundary timestamps of each line segment, and the number of intermediate interpolations is set to the length divided by one minute and rounded, and finally output the set of platform feature parameters; Difference comparison and calibration sub-module: Based on the platform feature parameter set, use the numpy.abs function in NumPy to calculate the absolute value of each item of the click value difference between the target platform and the source platform in the time vector order. Use the numpy.sign function in NumPy to judge the sign direction of each pair of differences. In the result, the positive and negative signs are marked as 1 and -1 respectively. Call the numpy.where function in NumPy to filter the time point positions with opposite sign directions. Further use the numpy.subtract function in NumPy to perform element-wise subtraction calculation on the centroid position values to obtain the displacement sequence. Call the numpy.histogram function in NumPy to divide all displacements into intervals. Set the number of bins to 5, and the range is set between the minimum and maximum values of all differences. Call the groupby method in Pandas to group the difference intervals by the platform_id field, and finally output the structural feature deviation sequence; Scale unified reconstruction sub-module: Based on the structural feature deviation sequence, use the combined operations of the numpy.divide function and the numpy.multiply function in NumPy. Set the maximum difference in each group as the normalization unit for the click difference, and perform compression and stretching processing on the difference value range for the amplitude line segment. The compression ratio is the ratio of the difference to the maximum value in the group, and the stretching ratio is the ratio of the difference to the minimum non-zero value in the group. Then call the numpy.sign function in NumPy to extract the centroid displacement sign and judge the direction. If it is a negative sign, move the time reference point forward by delta seconds, and the delta value is set to 10% of the current group line segment length. The movement operation is completed through the pandas.to_timedelta function combined with the shift method. After time alignment, use the numpy.concatenate function in NumPy to splice each group of vectors to obtain a continuous offset sequence. Then use the numpy.intersect1d function in NumPy to take the intersection of the target platform and the source platform in the spliced vectors, and finally output the cross-platform equivalent feature set.

[0028] Please refer to Figure 3 , and the value response derivation module includes: Intervention parameter division sub-module: Based on the cross-platform equivalent feature set, aggregate and divide the three parameters of the number of tags, release time, and duration by account dimension, calculate the total click difference and conversion rate change interval of each type of video and classify them into the parameter intersection set, establish a number mapping relationship and associate the sample group, and generate the intervention behavior classification group; Response matching extraction submodule: Based on the intervention behavior classification group, the overlapping areas of the click growth segments, conversion rate increase points and comment anomaly occurrence time nodes of the videos in the group are compared, the overlapping time periods of the three types of change positions are extracted, the response time window is determined according to the sequence distribution, and the content with synchronous jump characteristics is screened out to obtain the intervention response feature set; Structural combination judgment submodule: Based on the intervention response feature set, the long short-term memory network is used to pair the click difference of each group of jump feature points with the direction of the corresponding intervention parameter curve, and the path reconstruction and superposition frequency statistics are performed on the direction-consistent group, and the high-frequency co-occurrence structure is extracted and the numbering vector is constructed. The combined structure sequence is integrated to generate an archive set, and a content operation response characteristic map is established; Intervention parameter division submodule: Based on the cross-platform equivalent feature set, the groupby method of Pandas is used to aggregate the original data with the account identification field as the key value. The agg function is used to perform count operations on the label quantity field, the pandas.to_datetime standard format conversion is performed on the release time field, and then the hour field is used to extract the hour distribution. The mean operation is performed on the video duration field to generate three parameter columns. The numpy.histogram function of NumPy is used to perform interval segmentation on each parameter column. The bins parameter is set to the maximum value of the label quantity minus the minimum value divided by 4 and rounded up, the release time is divided into 24 bins according to the hourly segment, and the video duration is divided into 60 seconds as a bin segment. Then, the numpy.histogram function of NumPy is used to perform interval segmentation on each parameter column. The bins parameter is set to the maximum value of the label quantity minus the minimum value divided by 4 and rounded up, the release time is divided into 24 bins according to the hourly segment, and the video duration is divided into 60 seconds as a bin segment. The .digitize function maps the parameter value corresponding to each data item to the corresponding segment index, combines the three indexes to form a three-dimensional parameter identifier as a unique parameter segment identifier, uses NumPy's numpy.where and numpy.logical_and combination operations to extract data records belonging to the same parameter combination in each segment, and uses Pandas' groupby method to calculate the total difference in clicks and the range of conversion rate changes for each type of video according to the parameter combination. The conversion rate is the number of conversions divided by the total exposure, and the difference is the difference value obtained by subtracting the current record from the previous record. The summary results are classified into the parameter intersection set, and the number index is established using the enumerate function with the parameter label dictionary mapping, and the merge method of Pandas is called to map the parameter number to the original sample record, and finally output the intervention behavior classification group; Response matching extraction sub-module: Based on the intervention behavior classification groups, for each group of video sequences, the numpy.gradient method in NumPy is called to perform slope analysis on the click sequences to extract the growth segments. The window width is set to 5 for smooth slope extraction, and the position where the slope is greater than 0.8 times the maximum slope is selected as the starting point of the click growth segment. The conversion rate increase points are extracted by taking the difference of the conversion rate sequence using the numpy.diff function in NumPy, and the numpy.where function in NumPy is used to filter out the positive value positions as the conversion increase points. For the identification of abnormal comment nodes, the scipy.stats.zscore method is used to standardize the comment frequency sequence, and the numpy.where function in NumPy is used to filter out the positions where the absolute value of the z-score is greater than 2.5 as the abnormal points. Binary time identification sequences are generated for the time indices of the three types of nodes respectively, and the overlapping period is judged by combining with the numpy.logical_and function in NumPy. Segment cutting is performed according to the start and end time points of the overlapping area, and the numpy.bincount method in NumPy is used to count the frequencies of the time segment sequences. The positions where the coincidence frequency exceeds 2 times are identified as the response time windows, and finally the intervention response feature set is output; Structure combination determination sub-module: Based on the intervention response feature set, the LSTM model in the Keras module of TensorFlow is used to perform sequence structure analysis. The model layer is set to the Sequential structure. The first layer is the LSTM layer with 64 units, the activation function is set to tanh, and the input_shape is set to an input dimension of (10, 2). A Dense layer is connected later for regression output. The mean_squared_error is used as the loss function for the model, the optimizer is set to Adam, and the learning rate is 0.001. In the construction of the training data, the click difference of each group of jump feature points and the intervention parameter change curve are aligned according to time, and a two-dimensional input vector is constructed through the numpy.stack function in NumPy. After training, the direction consistency vector is output. The numpy.sign function in NumPy is used to judge whether the slope direction is consistent with the intervention parameter curve direction for direction comparison. The numpy.where function in NumPy is used to filter out the data pairs with consistent directions. For the groups with consistent directions, the DTW dynamic time warping algorithm is used, specifically the dtw.distance function in the dtaidistance library to reconstruct the path, and the numpy.bincount function in NumPy is used to count the occurrence frequencies of each path point. The points with a frequency greater than 2 times are extracted as high-frequency points. The numpy.argsort and numpy.unique functions in NumPy are called to combine and encode the high-frequency structures and generate vectors. Finally, the archived set is output to establish the content operation response characteristic map.

[0029] See also Figure 3 , long short-term memory network, according to the formula:

[0030] in: Indicates that at the current time step The obtained structural combination state vector, Indicates that the jump feature point is at time step The input vector is Represents the time step The hidden state of , as the information memory of the historical structure combination path, Represents the mapping weight matrix between the input features and the current state, Represents the weight matrix from the previous hidden state to the current state, represents the bias term in the state update, represents the click difference calibration coefficient, Represents the normalized fluctuation of the click difference between the jump feature points, represents the response density adjustment factor, Indicates the concentration of response points per unit time, represents the path direction continuity weight coefficient, represents the directional consistency score between three consecutive jump points, represents the structural rhythm modulation coefficient, represents the standard deviation of the time interval between adjacent structure combinations, represents the activation function; Execution process: First, the joint feature vector of the current time step is input into the network, combined with the structural state expression of the previous time step, and linearly transformed through the input weight matrix and the recursive weight matrix. The bias term in the superimposed state update constitutes the basic state input signal. Then, four adjustment factors are introduced to enhance the multi-dimensional recognition ability of structural features. The click difference volatility measures the amplitude of the jump point response change, the response aggregation reflects the structural density in a short period of time, the direction consistency score measures the continuity of the local trajectory, and the structural rhythm standard deviation measures the operation rhythm offset between adjacent structures. After weighted fusion of the adjustment coefficient, it participates in the state update expression. Finally, the structural combination state vector is output through the nonlinear activation function, and the dynamic modeling process of the structural path is continuously and progressively completed. The high-frequency co-occurrence pattern is extracted and the numbering vector is generated for subsequent structural archiving, combination classification and traffic value estimation.

[0031] See also Figure 3, classified into the parameter intersection set. Taking the three parameters of the number of video tags, release time, and video duration under the account dimension as the division criteria, first, within each parameter dimension, the video data is segmented, and the total click difference and conversion rate change value of the corresponding videos within each segment are counted. The overlapping area where both the click difference and the conversion rate change significantly is found in the three-dimensional parameter space, that is, the cross-segment with active intervention responses in multiple dimensions simultaneously, which is included in the parameter intersection set as the key intervention area.

[0032] Please refer to Figure 3 , the contract value quantification module includes: Frequency calculation sub-module: Based on the content operation response characteristic map, extract the indexes of the sudden increase points of the click count in each video, extract the continuous jump points with an interval less than a fixed threshold in chronological order, calculate the number of each group of jump points and the average interval length, mark the sample number and synchronously record the frequency value, and generate the jump increase cycle statistical index; Offset construction sub-module: Based on the jump increase cycle statistical index, according to the time points of the position of the like count extraction and the high-frequency segment of the corresponding conversion number, calculate the absolute interval value between the two time points, perform the product operation item by item according to the interval value and the frequency value, generate the product mean vector grouped by content number, and obtain the pricing adjustment structure factor after dividing the standard interval range; Valuation output sub-module: Based on the pricing adjustment structure factor, extract the total click value and the total amplitude of the conversion rate change of each piece of content within the trading cycle, replace the upper and lower limits of the original valuation range after multiplying and adjusting by the adjustment factor respectively, construct a new click valuation bandwidth, re-encode and map according to the content number, and establish the futures contract valuation output result; Frequency calculation sub-module: Based on the content operation response characteristic map, perform a first-order difference operation on the click count sequence of each video using the numpy.diff function of NumPy to extract the sudden increase points, call the numpy.where function of NumPy to filter the positions where the difference is greater than the set threshold as the jump points, the threshold is set to the sequence mean plus 1.5 times the standard deviation, obtain the jump point index list, call the numpy.diff function of NumPy to perform a second difference on the jump point index to obtain the time interval between adjacent jump points, use the numpy.where function of NumPy to filter the index positions where the interval is less than the set upper threshold, the time threshold is set to 300 seconds, divide the continuous jump points into subgroups, the subgroup formation is completed by iterative slicing using the itertools.groupby function in Python, call the len function to count the number of jump points in each group and use the numpy.mean function of NumPy to calculate the average length of the time interval of each group, record the corresponding sample number and jump frequency value of each group through the DataFrame structure of Pandas, and finally output the jump increase cycle statistical index; Offset construction submodule: Based on the jump-increase cycle statistical indicators, NumPy's numpy.argmax function is used to extract the maximum position index in the likes sequence as the high-frequency point position of likes. The high-frequency peak position of the conversion number is identified by the scipy.signal.find_peaks function. The height parameter is set to the mean of the conversion sequence plus 2 times the standard deviation, and the distance parameter is set to 5. The absolute time difference between the two types of time points is calculated using NumPy's numpy.abs function and the element index difference. The jump-increase frequency value in each group of samples and the above absolute interval value are multiplied item by item using NumPy's numpy.multiply function. Pandas' groupby function is called in combination with the agg method to perform grouping processing on all content numbers. The mean function is used in the aggregation method to generate the mean of each group of products. NumPy's numpy.histogram function is called to divide all mean vectors into standard intervals. The intervals are divided into five levels, and the boundary values ​​are set to linear divisions between the minimum and maximum values ​​of the overall mean. Finally, the pricing adjustment structure factor is output; Valuation output submodule: Based on the pricing adjustment structure factor, the groupby and agg functions of Pandas are combined to aggregate the click value and conversion rate of each content in the transaction cycle. The total click value is aggregated by the sum function, and the total change in conversion rate is calculated as the maximum value minus the minimum value of the conversion rate sequence. The numpy.multiply function of NumPy is called to perform multiplication adjustment on the above two original values ​​and the corresponding adjustment factor value, and the adjusted click valuation upper and lower limits are generated respectively. The numpy.vstack and numpy.transpose functions of NumPy are called to combine the upper and lower limits to form a two-dimensional valuation bandwidth vector, which is re-encoded and mapped to the main table using the reset_index method and merge function of Pandas according to the original content number, and finally the valuation output result of the futures contract is output.

[0033] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A short video traffic futures trading and value evaluation system, characterized in that The system includes: Transaction period construction module: According to the content release records of a specified account in the short video platform, extract the release time difference and the increase jump value, screen the jump segments, extract the interaction peaks and popularity tags, combine the active time blocks, and establish a set of platform trading intervals; Period offset identification module: Based on the block data of the platform trading interval set, analyze the click perturbation amplitude and the conversion jump distance, extract the jump distance and the center of gravity position, use the causal forest algorithm to screen the period jump structure difference sections, and obtain the periodic feature offset clusters; Platform structure mapping module: Based on the periodic feature offset clusters, compare the platform structure feature strength and the center of gravity displacement, perform standardized mapping adjustment, measure the click strength difference and the structure peak position difference between platforms, unify the scale of the traffic feature vectors, and establish a cross-platform equivalent feature set; Value response derivation module: Based on the jump gradient and the conversion center of gravity position synchronously inherited in the cross-platform equivalent feature set, use the long short-term memory network, and by segmentally judging the distribution of the conversion deviation values under the intervention grouping, identify the high-response feature combinations, and establish a content operation response characteristic map; Contract value quantification module: Based on the content operation response characteristic map, extract the product of the jump increase frequency and the conversion displacement to construct a pricing factor, correct the valuation interval, and establish the futures contract valuation output result.

2. The short video traffic futures trading and value evaluation system according to claim 1, wherein The transaction period construction module includes: Release time extraction sub-module: Based on the content release records of a specified account in the short video platform, calculate the time difference between the upload time points of adjacent contents, sort the difference sequence in ascending order and divide it into high-frequency difference intervals, construct non-uniform time period boundaries according to the interval mutation nodes, and perform time period label numbering processing to generate a time interval distribution sequence; Behavior jump identification sub-module: Based on the time interval distribution sequence, extract the amplitude changes of the cumulative difference of click values, the difference in the number of likes, and the total number of comments in each time period, synchronously compare each amplitude change with the range of the previous time period, and perform the generation of the difference matrix between the results and the screening operation of the marked values to obtain the content jump active index set; Active block combination sub-module: Based on the content jump active index set, extract the time points where the jump active values are located, perform continuous segment construction judgment on the time points and splice the time periods, combine the change values of the appearance density of the label keywords in each time period to screen the label active areas, perform the intersection operation of the two and mark the interval boundaries to establish a set of platform trading intervals.

3. The short video traffic futures trading and value evaluation system according to claim 1, characterized in that, The period offset identification module includes: Structure difference measurement sub-module: Based on the platform trading interval set, extract the total number of clicks and the total number of interactions in each trading section, calculate the difference between adjacent click values under the increasing time and convert it into a trend slope, identify the change points of the appearance frequencies of likes and comments within a unit time, measure the time distance value between it and the click jump point and establish a sequence to obtain the periodic behavior structure difference group; Center of gravity deviation extraction sub-module: Based on the periodic behavior structure difference group, extract the central point position of the high-frequency value section of the behavior in the structure sequence, measure the distance between the centers of gravity of the front and rear high-frequency sections in chronological order, perform normalization and stratification processing on the distance values, generate position change labels according to the relative displacement, integrate all offset records and sort and number them to obtain the behavior response center of gravity offset sequence; Offset cluster screening sub-module: Based on the behavior response center of gravity offset sequence, use the causal forest algorithm to calculate the coincidence ratio between the center of gravity displacement value between adjacent periods and the trend direction of the structure difference, set a judgment threshold according to the proportion of trend direction consistency, group and aggregate the periods that meet the conditions, number them in sequence and construct the boundary ranges of each group to obtain the periodic feature offset cluster.

4. The short video traffic futures trading and value evaluation system according to claim 3, characterized in that The causal forest algorithm, according to the formula: Among them: represents the proportion of the consistency of the trend direction between the i th cycle and the th cycle, represents the angle between the direction of the center-of-gravity displacement of the th cycle and the trend of the structural difference, represents the angle between the direction of the center-of-gravity displacement of the th cycle and the trend of the structural difference, represents the center-of-gravity displacement distance of the th cycle, represents the center-of-gravity displacement distance of the th cycle, represents the smoothness of the center-of-gravity displacement trajectory of the th cycle, represents the smoothness of the center-of-gravity displacement trajectory of the th cycle, represents a constant used for the normalization of the angle, represents the weight coefficient of the center-of-gravity displacement distance, represents the weight coefficient of the smoothness of the center-of-gravity displacement trajectory, represents the adjustment coefficient of the normalization constant.

5. The short video traffic futures trading and value evaluation system according to claim 3, wherein To extract the central point position of the high-frequency value section of the behavior in the structure sequence, the high-frequency value section of the behavior is the period when the click volume, like count, and comment count continuously appear concentratedly in the periodic structure. First, use a sliding window to scan the periodic structure difference sequence, record the continuous time periods when the local statistical value exceeds the set behavior intensity threshold, calculate the weighted central time point of the behavior frequency within the section to obtain the true center of gravity of the actual behavior aggregation trend, then normalize the center of gravity distance between the front and rear high-frequency sections, perform hierarchical filing based on the interval length, and label it as positive offset, negative offset, and no offset according to the displacement direction to generate position change labels.

6. The short video traffic futures trading and value evaluation system according to claim 1, characterized in that The platform structure mapping module includes: Structure parameter extraction sub-module: Based on the periodic feature offset cluster, extract the extreme value difference of the total click value within a fixed time length in each platform, calibrate the center of gravity point position in minute segments and summarize the time position frequencies, calculate the line segment length between the start and end points of the span and assemble the time vector to generate the platform feature parameter set; Difference comparison and calibration sub-module: Based on the platform feature parameter set, calculate the absolute value of the click difference between the target platform and the source platform in the order of the time vector and judge the difference sign direction, count the displacement amount between the center of gravity points and mark the position of the different-direction change sequence, divide the structure amplitude range and group by platform number to obtain the structure feature deviation sequence; Scale unified reconstruction sub-module: Based on the structure feature deviation sequence, perform amplitude line segment compression and stretching according to the click difference ratio of each group, adjust the displacement direction of the time reference point according to the center of gravity displacement sign, extract the offset channel vector after splicing the vectors within the group, perform platform intersection screening and recombination alignment to establish a cross-platform equivalent feature set.

7. The short video traffic futures trading and value evaluation system according to claim 1, characterized in that, The value response derivation module includes: Intervention parameter division sub-module: Based on the cross-platform equivalent feature set, aggregate and divide the three parameters of the label quantity, release time, and duration according to the account dimension, calculate the total click difference and conversion rate change range of each type of video respectively and include them in the parameter intersection set, establish a number mapping relationship and associate the sample group to generate the intervention behavior classification group; Response matching extraction submodule: Based on the intervention behavior classification group, the overlapping areas of the click growth segments, conversion rate increase points and comment abnormality occurrence time nodes of the videos in the group are compared, the overlapping time periods of the three types of change positions are extracted, the response time window is determined according to the sequence distribution, and the content with synchronous jump characteristics is screened out to obtain the intervention response feature set; Structural combination judgment submodule: Based on the intervention response feature set, a long short-term memory network is used to pair the click difference of each group of jump feature points with the direction of the corresponding intervention parameter curve, and path reconstruction and superposition frequency statistics are performed on the group with consistent directions. High-frequency co-occurrence structures are extracted and numbering vectors are constructed. The combined structure sequence is integrated to generate an archive set, and a content operation response characteristic map is established.

8. The short video traffic futures trading and value evaluation system according to claim 7, wherein The long short-term memory network is based on the formula: Wherein: represents the structural combination state vector obtained at the current time step represents the input vector of the jump feature point at the time step represents the hidden state at the time step which serves as the information memory of the historical structural combination path represents the mapping weight matrix between the input feature and the current state represents the weight matrix from the previous hidden state to the current state represents the bias term in the state update represents the click difference calibration coefficient represents the normalized fluctuation degree of the click difference between jump feature points represents the response density adjustment factor represents the response point aggregation degree per unit time represents the path direction continuity weight coefficient represents the direction consistency score between three consecutive jump points represents the structural rhythm modulation coefficient represents the standard deviation of the time interval between adjacent structural combinations represents the activation function​​ 9. The short video traffic futures trading and value evaluation system according to claim 7, characterized in that, The parameters are included in the intersection set of parameters, and the three parameters of the number of video tags, release time and video duration under the account dimension are used as the division criteria. First, the video data is segmented in each parameter dimension, and the total click-through rate and conversion rate change values ​​of the corresponding videos in each segment are counted. In the three-dimensional parameter space, the overlapping areas with significant changes in click-through rate and conversion rate are found, that is, the cross-segments with active intervention responses in multiple dimensions at the same time are included in the parameter intersection set as the key intervention areas.

10. The short video traffic futures trading and value evaluation system according to claim 1, characterized in that The contract value quantification module includes: Frequency calculation submodule: Based on the content operation response characteristic map, extract the index of the click count sudden increase point in each video, extract the continuous jump points with an interval less than a fixed threshold in chronological order, calculate the number and average interval length of each group of jump points, mark the sample number and synchronously record the frequency value, and generate the jump increase cycle statistical index; Offset construction submodule: Based on the jump increase cycle statistical indicators, extract the position of the likes number and the time point of the high frequency band of the corresponding conversion number, calculate the absolute interval value of the two time points, perform product operation on the interval value and the frequency value item by item, generate the product mean vector by grouping the content number, divide the standard interval range, and obtain the pricing adjustment structure factor; Valuation output submodule: Based on the pricing adjustment structure factor, the total click value and the total change in conversion rate of each content in the transaction cycle are extracted, and the upper and lower limits of the original valuation range are replaced after multiplication adjustment according to the adjustment factor, and a new click valuation bandwidth is constructed. The mapping is recoded according to the content number to establish the futures contract valuation output result.

Images