Accurate analysis method and system for advertisement putting audiences

Abstract

The invention relates to the technical field of e-commerce marketing, in particular to an accurate analysis method and system for advertisement putting audiences, and the method comprises the following steps: obtaining the interaction data of the advertisement putting audiences, calculating the variation amplitude relative to a reference behavior, calling a scene triggering factor, and carrying out the comparison and screening of the scene triggering factor and a threshold value, and obtaining a short-term behavior influence coefficient. According to the method, after user attribution probability calculation is combined with short-term interaction data weight correction, noise interference reduction, classification precision improvement and abnormal value elimination, differential weights are given through interaction data type parameters, so that interest feature extraction better conforms to a user behavior mode; the accuracy of interest vectors is enhanced through calculation of the change rate of short-time interaction behaviors, interest feature weights are corrected in combination with feedback data, real-time optimization of advertisement push content is achieved, the interest vectors are complemented through high-activity user similarity matching, interest speculation of users with insufficient data is more accurate, and the advertisement reach rate and the delivery effectiveness are improved.

Description

Technical Field

[0001] The present invention relates to the field of e-commerce marketing technology, and in particular to a method and system for accurately analyzing an advertisement delivery audience. Background Art

[0002] The field of e-commerce marketing technology includes core contents such as precision marketing based on user behavior analysis, personalized recommendation systems, dynamic pricing strategies, and advertising optimization. The core of this technical field is to use user data for market analysis, analyze users' browsing, search, and purchase history through computer systems, and explore their interests and preferences to improve the matching degree of goods or services. E-commerce marketing not only involves advertising, but also covers recommendation algorithms, user portrait construction, click-through rate prediction, etc. It mainly relies on data mining, statistical analysis, and machine learning methods to enhance market competitiveness and user experience. The development of this field has transformed the marketing method from the traditional wide-ranging model to a more refined and personalized direction, improving the accuracy and conversion rate of advertising.

[0003] Among them, the method of accurate analysis of advertising audiences refers to a technical method that uses modeling analysis and rule matching based on user behavior data of e-commerce platforms to screen out target user groups with specific characteristics and optimize advertising delivery strategies. This method involves feature extraction based on dimensions such as user historical browsing records, search keywords, purchase behaviors, and social interaction data, and uses classification algorithms to divide users into interest tags. Subsequently, by matching advertising content with user interest tags, the advertising display frequency and content optimization strategy are adjusted in combination with real-time user behavior data. In addition, this method is based on the conversion rate prediction model and combines the bidding mechanism to calculate the advertising delivery cost to ensure the rationality of advertising budget allocation. Finally, the advertising delivery effect is continuously optimized through the feedback mechanism, and the user group screening rules are adjusted to improve the accuracy and reach of the advertisement.

[0004] Existing technologies rely on static data and are difficult to adapt to the dynamic changes in user interests, resulting in delayed advertising push. User classification labels are dominated by historical behavior and fail to effectively identify short-term interest changes, resulting in obvious interest solidification problems. The outlier processing method is single, and noise data affects the accuracy of analysis. Interest feature extraction lacks hierarchical distinction, and the definition of long-term and short-term interests is vague, reducing advertising matching. Similar highly active audience information is not fully utilized, resulting in increased interest prediction deviations for users with insufficient data, affecting advertising recommendation effects and conversion rates. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method and system for accurately analyzing the audience for advertising.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a method for accurately analyzing the audience of advertisement delivery, comprising the following steps: S1: Obtain the interactive data of the advertising audience, calculate the change compared with the baseline behavior, call the scenario trigger factor and compare and filter with the threshold, and obtain the short-term behavior impact coefficient; S2: Based on the short-term behavior influence coefficient, the attribution ratio of the advertising audience is counted, the attribution probability is calculated, and the short-term advertising interaction data is called to filter and correct the weight to obtain the audience attribution drift probability; S3: Based on the audience attribution drift probability, the initial interactive behavior data of the advertising audience is screened, outliers are removed, advertising interactive data type parameters are called, differentiated weights are assigned, cluster analysis is performed, interest feature primitives are screened, and the short-term interactive behavior change rate is calculated to obtain an interest vector primitive set; S4: Based on the interest vector primitive set, calling the advertisement feedback data to determine the feature impact, calculating the interest feature weight correction value, updating the interest vector distribution, screening the long-term advertisement interest feature, and calculating the interest vector weight adjustment value; S5: Based on the interest vector weight adjustment value, call the advertising audience interaction behavior similarity calculation factor, match the most similar highly active advertising audience, select its highest weighted advertising interest feature, infer and complete the current advertising audience interest vector, and obtain the interest vector reconstruction value.

[0007] The short-term behavior impact coefficient includes the change amplitude, scene trigger factor, and threshold comparison screening result; the audience attribution drift probability includes attribution proportion, attribution probability, and short-term advertising interaction data screening correction weight; the interest vector primitive set includes interest feature primitives, short-term interactive behavior change rate, and differentiation weight; the long-term advertising interest feature includes interest vector distribution, interest feature weight correction value, and advertising feedback data judgment feature influence; the interest vector reconstruction value includes interest vector weight adjustment value, advertising audience interaction behavior similarity calculation factor, and highest weight advertising interest feature.

[0008] As a further solution of the present invention, the step of obtaining the short-term behavior influence coefficient is specifically as follows: S101: Obtaining interaction data of the advertising audience, extracting multiple interaction parameters such as click-through rate, browsing time, and bounce rate, calculating the average value and standard deviation of the multiple interaction parameters in the delivery cycle, establishing a data distribution model, calling the benchmark behavior data, calculating the deviation of the interaction parameters compared to the benchmark behavior, and obtaining the interaction parameter deviation value; S102: Based on the interaction parameter offset value, the fluctuation trend of multiple parameters is calculated, the short-term variation range of multiple parameters is calculated by using the offset mean and variance, and compared with the variation range of the benchmark behavior, the fluctuation level of multiple parameters compared with the benchmark is determined, and the fluctuation intensity of the interaction parameters is calculated; S103: calling the interaction parameter fluctuation intensity, matching it with the scene trigger factor and the threshold, screening the behavior change items that meet the conditions, and calculating the proportion of their impact on short-term decision-making using the formula: ; Get the short-term behavior impact coefficient; in, represents the short-term behavior impact coefficient, Representative The current value of the interaction parameter, Representative The baseline values ​​of the interaction parameters, Representative The standard deviation of the interaction parameter, Representative The weights of the interaction parameters, Represents the total number of interaction parameters.

[0009] As a further solution of the present invention, the step of obtaining the audience attribution drift probability is specifically as follows: S201: Based on the short-term behavior impact coefficient, the attribution ratio of the audience for advertising delivery is counted, the distribution of multiple audience groups under differentiated advertising categories is calculated, the data source is analyzed, abnormal samples are removed, the audience ratio of multiple categories is normalized, data that meets the statistical standards is screened, the audience category distribution is calculated according to the normalized ratio, the standardized ratio value is obtained, and the audience attribution ratio matrix is ​​constructed; S202: calling the audience attribution ratio matrix, and setting the audience behavior correction weight according to the normalized ratio and the short-term advertising interaction data, using the formula: ; Calculate the audience's attribution probability under multiple advertising categories and generate an audience attribution probability matrix; in, Representative audience In the advertising category The probability of belonging to For the audience In the advertising category The proportion of attributable Representative audience The short-term behavior impact coefficient, For the audience In the Ad Interaction Data category The interaction strength under Represents interaction category For advertising categories The influence weight of is the total number of interaction categories, is the total number of advertising categories; S203: calling the audience attribution probability matrix, screening short-term advertising interaction data to correct weights, analyzing audience short-term interaction behaviors, setting attribution change thresholds, calculating audience weight adjustment values ​​under differentiated advertising categories, and obtaining audience attribution drift probability in combination with short-term interaction characteristics.

[0010] As a further solution of the present invention, the step of acquiring the interest vector primitive set is specifically: S301: Based on the initial interactive behavior data of the advertising audience, screen the data samples that meet the audience attribution drift probability standard, remove the abnormal values ​​that do not meet the standard, assign differentiated weights to multiple data types according to the advertising interaction data type parameters, calculate the weighted interaction frequencies of the multiple data types, and obtain the weighted interaction frequency value; S302: calling the weighted interaction frequency value, using a cluster analysis method, grouping data according to the interaction frequency distribution pattern, screening representative behavior feature data points in each category, and obtaining interest feature primitives; S303: Based on the interest feature primitives, the short-term interaction behavior change rate is calculated, and the interaction change trend under the differentiated time window is analyzed, using the formula: ; Get the interest vector primitive set; in, represents the interest vector primitive set, Representative The interaction frequency value at each time point, Representative A point in time, represents the time mean, represents the total number of data points, Calculates the total amount of interactive change over a short time window.

[0011] As a further solution of the present invention, the step of obtaining the interest vector weight adjustment value is specifically: S401: Based on the interest vector primitive set, calling the advertisement feedback data, determining the influence of multiple interest features, calculating the influence factors of multiple interest features according to the advertisement exposure times, click-through rate, and user stay time, and calculating the weight correction value according to the feature influence factor, adjusting the correction parameter in combination with the performance of the differentiated features in the advertisement delivery, and obtaining the interest feature weight correction matrix; S402: calling the interest feature weight correction matrix to update the interest vector distribution, using the formula: ; Obtain updated interest vector distribution through calculation, filter out long-term stable advertising interest features, and integrate them into a long-term advertising interest feature set; in, represents the updated interest vector distribution, represents the original interest vector, Represents the weight correction value in the interest feature correction matrix, Represents the influencing factors of multiple interest features in advertising feedback data, represents the total number of features of interest, The modulus length of the original interest vector is increased by 1 to balance the adjustment amplitude of the interest feature; S403: Based on the long-term advertising interest feature set, the overall criticality of multiple interest features is calculated, and the updated interest vector distribution is called. According to the mean, variance and influencing factor of the interest features in the differentiated advertising delivery cycle, the overall weight of the interest features is calculated, and the calculated distribution of the interest vector is adjusted in combination with the weight information to obtain the interest vector weight adjustment value.

[0012] As a further solution of the present invention, the step of obtaining the interest vector reconstruction value is specifically as follows: S501: Based on the interest vector weight adjustment value, call the advertisement audience interaction behavior fit calculation factor, calculate the fit values ​​between all advertisement audiences and the current advertisement audience, compare the fit data, select the advertisement audience with the highest fit, extract its identification information, call the matching algorithm for verification, and determine the matching value of the same type of advertisement audience; S502: Call the matching value of the same type of advertising audience, extract all advertising interest features of the corresponding advertising audience, calculate the matching weight of the advertising interest features, and select the advertising interest features with the highest weight value, using the formula: ; Adjust the current advertising audience interest vector, calculate and obtain the highest weight advertising interest feature influence; in, Represents the highest weight advertising interest feature influence, Represents the calculation factor of the fit degree of the interactive behavior of the advertising audience, Represents the weight of advertising interest features, Represents the interest characteristic value of the most similar advertising audience, Represents the interest feature value of the current advertising audience. The total number of characteristics representing advertising interests. Represents the advertising behavior characteristic value of the same type of advertising audience, Represents the advertising behavior characteristic value of the current advertising audience, The total number of advertising behavior characteristics; S503: Call the highest weighted advertising interest feature influence, adjust the current advertising audience interest vector, recalculate weight values ​​for all interest features, analyze the interest vector structure after weight adjustment, complete missing interest feature information, and calculate and generate an interest vector reconstruction value.

[0013] A system for accurately analyzing the audience for advertising delivery, the system for accurately analyzing the audience for advertising delivery is used to execute the above-mentioned method for accurately analyzing the audience for advertising delivery, and the system comprises: The interactive data processing module obtains the interactive data of the advertising audience, calculates the change range compared with the benchmark interactive behavior, calls the scene trigger factor and the threshold for comparison and screening, selects the qualified data as the short-term behavior influence factor, and calculates and generates the short-term behavior influence coefficient based on the change trend of the short-term behavior influence factor; The audience attribution analysis module counts the attribution proportion of the advertising audience based on the short-term behavior influence coefficient, calculates the attribution probability, calls the short-term advertising interaction data to filter and correct the weight, and obtains the audience attribution drift probability; The interest feature clustering module screens the initial interaction data of the advertising audience based on the audience attribution drift probability, removes outliers, calls the advertisement interaction data type parameters to assign differentiated weights, performs cluster analysis, calculates the short-term interaction behavior change rate, and generates an interest vector primitive set; The interest vector optimization module calls the advertisement feedback data based on the interest vector primitive set, determines the feature impact, calculates the interest feature weight correction value, updates the interest vector distribution, screens the long-term advertisement interest feature, and calculates and generates the interest vector weight adjustment value; The audience interest matching module calls the advertising audience interaction behavior fitting calculation factor based on the interest vector weight adjustment value, matches the most similar highly active advertising audience, selects the highest weighted advertising interest feature, infers and completes the current advertising audience interest vector, and generates an interest vector reconstruction value.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by calculating the magnitude of behavioral changes and combining scene trigger factors to screen short-term behavioral impacts, it is ensured that advertising can adapt to the dynamic changes in user interests. The user attribution probability calculation is combined with short-term interaction data to correct weights, reduce noise interference, and improve classification accuracy. After the outliers are removed, differentiated weights are assigned through the interaction data type parameters to make the interest feature extraction more in line with user behavior patterns. The calculation of the change rate of short-term interactive behavior enhances the accuracy of the interest vector, and the interest feature weights are corrected in combination with feedback data to achieve real-time optimization of advertising push content. Highly active users are matched with similarity to complete the interest vector, making the interest speculation of users with insufficient data more accurate, thereby improving advertising reach and delivery effectiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 A flow chart of the steps for obtaining the short-term behavior influence coefficient of the present invention; Figure 3 A flow chart of the steps for obtaining the audience attribution drift probability of the present invention; Figure 4 is a flow chart of the steps for obtaining the interest vector primitive set of the present invention; Figure 5 A flowchart of the steps for obtaining the interest vector weight adjustment value of the present invention; Figure 6 This is a flow chart of the steps for obtaining the interest vector reconstruction value of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with 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 intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0018] Embodiment 1 See also Figure 1 The present invention provides a technical solution: a method for accurately analyzing the audience of an advertisement, comprising the following steps: S1: Obtain the interactive data of the advertising audience, calculate the change compared with the baseline behavior, call the scenario trigger factor and compare and filter with the threshold, and obtain the short-term behavior impact coefficient; S2: Based on the short-term behavior influence coefficient, the attribution ratio of the advertising audience is counted, the attribution probability is calculated, and the short-term advertising interaction data is called to filter and correct the weight to obtain the audience attribution drift probability; S3: Based on the audience attribution drift probability, filter the initial interactive behavior data of the advertising audience, remove outliers, call the advertising interaction data type parameters, assign differentiated weights, perform cluster analysis, filter interest feature primitives, calculate the short-term interactive behavior change rate, and obtain the interest vector primitive set; S4: Based on the interest vector primitive set, call the advertising feedback data to determine the feature impact, calculate the interest feature weight correction value, update the interest vector distribution, filter the long-term advertising interest features, and calculate the interest vector weight adjustment value; S5: Based on the interest vector weight adjustment value, call the advertising audience interaction behavior similarity calculation factor, match the most similar highly active advertising audience, select its highest weighted advertising interest feature, infer and complete the current advertising audience interest vector, and obtain the interest vector reconstruction value.

[0019] The short-term behavior impact coefficient includes the change amplitude, scene trigger factor, and threshold comparison screening results. The audience attribution drift probability includes the attribution proportion, attribution probability, and short-term advertising interaction data screening correction weight. The interest vector primitive set includes interest feature primitives, short-term interactive behavior change rate, and differentiation weight. The long-term advertising interest characteristics include interest vector distribution, interest feature weight correction value, and advertising feedback data judgment feature influence. The interest vector reconstruction value includes interest vector weight adjustment value, advertising audience interaction behavior similarity calculation factor, and the highest weight advertising interest feature.

[0020] See also Figure 2 , the steps to obtain the short-term behavior influence coefficient are as follows: S101: Obtaining interaction data of the advertising audience, extracting multiple interaction parameters such as click-through rate, browsing time, and bounce rate, calculating the average value and standard deviation of the multiple interaction parameters in the delivery cycle, establishing a data distribution model, calling the benchmark behavior data, calculating the deviation of the interaction parameters compared to the benchmark behavior, and obtaining the interaction parameter deviation value; First, you need to extract the audience's click-through rate (CTR), browsing time (T), bounce rate (BR) and other key interaction parameters from the log data of the advertising platform or the embedding system. Set CTR to represent the ratio of clicks to impressions, T to represent the duration from the user entering the page to leaving the page, and BR to represent the proportion of users who leave immediately after visiting only a single page. Record the single interaction behavior data of each user, merge the data of the same user in different time periods to ensure data integrity. After obtaining the complete data set, use data cleaning technology to eliminate outliers, such as robot click behavior with CTR far higher than the average level, or invalid visit behavior with T less than 1 second. After cleaning, the data will enter the statistical processing stage. For each interaction parameter, calculate its average value and standard deviation during the entire advertising delivery cycle. The average value is calculated as: the sum of the parameter values ​​of all users / total number of users, and the standard deviation is calculated as: the square root of the mean of the sum of the squares of the deviations of the parameter values ​​of all users from the average value. Taking the click-through rate as an example, if the click data of an advertisement within 10 days is as follows: Table 1: Advertisement click-through rate data As shown in Table 1, the mean click rate is calculated as: ; Calculate the standard deviation: ; After obtaining the mean and standard deviation of all interaction parameters, a normal distribution data model is established, with the mean as the central value and the standard deviation measuring the data fluctuation range. The distribution of each parameter is visualized through the normal distribution curve, and benchmark behavior data is further introduced. The benchmark behavior data can be the historical data of the advertisement, industry standard data, or specific competitor data. The mean and standard deviation of the benchmark behavior are calculated in the same way, and the difference between the current period and the benchmark behavior data is calculated as follows: ; in, is the value of an interactive parameter in the current period, is the mean value of this parameter for the baseline behavior, is the standard deviation of the parameter. If calculated based on click-through rate data, assuming the mean of the benchmark click-through rate is , standard deviation ,but ; When the offset value When the absolute value is large, it means that the current offset of the parameter is more significant, which may mean that the change in the delivery strategy has a greater impact on user interaction. After the offset values ​​of all interaction parameters are calculated, the offset of the interaction parameters at this stage is recorded.

[0021] S102: Based on the interaction parameter offset value, the fluctuation trend of multiple parameters is calculated, and the short-term variation range of multiple parameters is calculated by using the offset mean and variance, and compared with the variation range of the benchmark behavior, so as to determine the fluctuation level of multiple parameters compared with the benchmark, and calculate the fluctuation intensity of the interaction parameters; The specific operation includes calculating the daily average volatility of each parameter, using the sliding window method to calculate the short-term change range, and converting continuous The interactive parameter data of the day is used as a window to calculate the mean and variance of each parameter in the window. For example, if the sliding window is set to 3 days, the mean in the window is calculated: ; Variance calculation method: ; Taking the bounce rate data as an example, suppose the bounce rate data for 3 days is ,but: ; ; Then, compare the change range of the benchmark behavior data to determine the current data fluctuation level. Assume that the change range of the benchmark behavior data is (i.e. the upper limit of the standard bounce rate change), the fluctuation level of this cycle can be divided into: : Low volatility; : Medium volatility; : High volatility; because , which belongs to the medium volatility level, and finally the interaction parameter fluctuation intensity is obtained, and the data fluctuation of this period is marked.

[0022] S103: Call the interaction parameter fluctuation intensity, match it with the scene trigger factor and threshold, screen the behavior change items that meet the conditions, and calculate the proportion of its impact on short-term decision-making using the formula: ; Get the short-term behavior impact coefficient; in, represents the short-term behavior impact coefficient, Representative The current value of the interaction parameter, Representative The baseline values ​​of the interaction parameters, Representative The standard deviation of the interaction parameter, Representative The weights of the interaction parameters, Represents the total number of interaction parameters.

[0023] formula: ; Set bounce rate weight , click-through rate weight , browsing time weight , respectively calculate: ; The calculation results show that the short-term behavior impact coefficient in this stage is 0.613, which is used for subsequent advertising optimization analysis.

[0024] See also Figure 3 , the specific steps for obtaining the audience attribution drift probability are: S201: Based on the short-term behavior impact coefficient, count the attribution ratio of the advertising audience, calculate the distribution of multiple audience groups under differentiated advertising categories, analyze the data source, remove abnormal samples, normalize the audience ratio of multiple categories, screen the data that meets the statistical standards, calculate the audience category distribution according to the normalized ratio, obtain the standardized ratio value, and construct the audience attribution ratio matrix; First, extract the behavioral data of the advertising audience, preliminarily classify the audience according to the advertising category, calculate the proportion of each type of audience in different advertising categories, and count the distribution of the audience in each category. Among them, the audience behavior data includes browsing, clicking, purchasing and other behaviors. Assuming that an audience has 50 browsing times and 10 clicks on Category A ads, and 30 browsing times and 5 clicks on Category B ads, then its behavioral weight under Category A ads can be calculated as (50×0.6+10×0.4) / (50+10+30+5)≈0.55, and its behavioral weight under Category B ads is (30×0.6+5×0.4) / (50+10+30+5)≈0.33. Based on the statistical results, construct the advertising category distribution of multiple audience groups. Then, analyze the data source and classify the collected audience behavior data according to different advertising delivery platforms, device types, and regional characteristics. In the process of data cleaning, abnormal samples are removed. For example, if an audience clicks on an advertisement more than 1,000 times in a day, which obviously exceeds the normal user behavior pattern, the data should be removed as an outlier. Furthermore, the audience proportion of each category is normalized so that the sum of the audience proportion of all advertising categories is 1. For example, the initial proportions of an audience in five advertising categories are 0.3, 0.25, 0.2, 0.15 and 0.1, respectively. After normalization, they are adjusted to 0.3 / (0.3+0.25+0.2+0.15+0.1)≈0.26, and so on. Finally, data that meets the statistical standards are selected, and the final distribution results of the audience categories are calculated according to the normalized ratio to obtain the standardized audience attribution proportion matrix. Each element in the matrix represents the standardized proportion of an audience in a certain advertising category, as shown in Table 2.

[0025] Table 2 Audience attribution ratio matrix (example) As shown in Table 2, the U1 audience has the largest standardized share in Category A ads, while Category E has the smallest share. This matrix will serve as the basis for subsequent audience attribution probability calculations.

[0026] S202: Call the audience attribution ratio matrix, and set the audience behavior correction weight based on the normalized ratio and combined with the short-term advertising interaction data, using the formula: ; Calculate the audience's attribution probability under multiple advertising categories and generate an audience attribution probability matrix; in, Representative audience In the advertising category The probability of belonging to For the audience In the advertising category The proportion of attributable Representative audience The short-term behavior impact coefficient, For the audience In the Ad Interaction Data category The interaction strength under Represents interaction category For advertising categories The influence weight of is the total number of interaction categories, is the total number of advertising categories; First, the initial probability of the audience belonging to the advertising category is calculated as follows: ,in Representative audience In the advertising category The proportion of attributable Represents the audience's short-term behavior influence coefficient, assuming Determined by the number of ad clicks by the audience in the last 7 days. If U1 has clicked on ads 50 times in total in the past 7 days, then , and then calculate the audience's interaction intensity under different interaction categories For example, if the number of interactions of U1 in categories A, B, and C are 20, 15, and 10 respectively, then the normalized calculation can be , and so on, combined with the influence weight of the interaction category on the advertising category , assuming that the influence weights of advertising categories A, B, and C on interaction categories X, Y, and Z are 0.4, 0.35, and 0.25, respectively, then calculate: ; Through the above calculation, we can get the probability of U1 belonging to category A advertisement. Similarly, we can calculate the probability of all audiences belonging to each advertisement category, and finally generate the audience belonging probability matrix, as shown in Table 3.

[0027] Table 3 Audience attribution probability matrix (example) As shown in Table 3, the probability of audience U1 belonging to category A ads is the highest, reflecting the revised result of the audience’s recent advertising interaction behavior and the short-term behavior influence coefficient on advertising attribution.

[0028] S203: Call the audience attribution probability matrix, filter the short-term advertising interaction data to correct the weight, analyze the audience's short-term interaction behavior, set the attribution change threshold, calculate the audience's weight adjustment value under the differentiated advertising category, and obtain the audience attribution drift probability in combination with the short-term interaction characteristics.

[0029] If the probability of an audience belonging to a certain advertising category changes by more than ±0.05 in the past three days, it is considered that its behavior pattern has drifted, and the weight adjustment value of the audience under different advertising categories is calculated. For example, if the probability of U1 belonging to category A advertising drops from 0.30 to 0.24, the weight is adjusted. , combined with short-term interaction characteristics, the audience attribution drift probability is calculated. If the audience's click-through rate in a certain advertising category drops by more than 50% within three days, and the attribution probability drops by more than 0.05, the attribution probability of the advertising category is further adjusted. For example, if U1's click-through rate in category A ads drops from 100 times to 40 times, its attribution drift probability increases, and finally an audience attribution drift matrix is ​​generated, indicating the attribution changes of different audiences in each advertising category, and then the advertising target group is dynamically adjusted in the advertising delivery strategy.

[0030] See also Figure 4 , the steps for obtaining the interest vector primitive set are as follows: S301: Based on the initial interactive behavior data of the advertising audience, screen the data samples that meet the audience attribution drift probability standard, remove the outliers that do not meet the standard, assign differentiated weights to multiple data types according to the advertising interactive data type parameters, calculate the weighted interactive frequencies of the multiple data types, and obtain the weighted interactive frequency value; First, the collected user interaction data is screened to ensure that the data samples meet the set audience attribution drift probability standards. This process involves calculating the initial interaction frequency of each user and comparing it with the set probability threshold to screen out abnormal data points. For example, if a user's interaction frequency is far below 20% of the sample mean or far above 200% of the mean, the user's data points will be eliminated. The specific standards can be set based on the audience stability index in the historical advertising data. Then, based on the different types of advertising interaction data, differentiated weights are assigned to them. When setting the weight standard, the contribution of each data type to user interest can be referred to. For example, the weight of click behavior is set to 0.6, the weight of stay time is set to 0.3, and the weight of sharing behavior is set to 0.8. According to the weight calculation formula: ; in, is the comprehensive weighted interaction frequency of a user. represents the interaction value of data type j, Represents the weight value of data type j. For example, if a user clicks 10 times, stays for 300 seconds, and shares 2 times, the calculation is as follows: Finally, the weighted interaction frequency values ​​of all users' data are calculated in the above manner and a data set is formed to obtain all weighted interaction frequency data that meet the standards.

[0031] S302: calling the weighted interaction frequency value, using a cluster analysis method, grouping data according to the interaction frequency distribution pattern, screening representative behavior feature data points in each category, and obtaining interest feature primitives; First, according to the distribution of interaction frequency, the data is normalized to ensure that different user data are comparable at the same scale. For example, for user data with a weighted interaction frequency range of [0,500], the following normalization method can be used: ; Assume that the maximum value in the data set , minimum , if a user's interaction frequency , then calculate the normalized value: ; After normalization, the K-means clustering method is used to group the data. Assuming that the number of groups is set to 3, the mean of each cluster center is calculated respectively, and the user data points are classified according to the minimum Euclidean distance principle. After the data grouping is completed, the representative behavioral feature data points in each category are further screened. The screening criteria can be based on the stability of the user's interaction frequency, that is, the mean interaction frequency of users in each category is calculated, and several users closest to the mean are selected as representative data points. For example, if the mean interaction frequency of a group of users is 120, the interaction data of the five users closest to 120 are screened in this group to obtain the interest feature primitives.

[0032] S303: Based on the interest feature primitives, calculate the change rate of short-term interaction behavior and analyze the interaction change trend under the differentiated time window, using the formula: ; Get interest vector primitive set; in, represents the interest vector primitive set, Representative The interaction frequency value at each time point, Representative A point in time, represents the time mean, represents the total number of data points, Calculates the total amount of interactive change over a short time window.

[0033] formula: ; First, extract the interaction frequency data within the short-term window and calculate the frequency difference between adjacent time points. For example, the interaction frequency data of a user at 5 time points are: ; Calculate the frequency change at adjacent time points: ; The total change after summation: ; Calculate the mean time , assuming the user's time point is : ; Calculate the denominator of the time standard deviation: ; Compute the interest vector primitive set: ; Finally, we get the interest vector primitive set ,This result shows that the change rate of the user's short-term interactive behavior is within the normal range.

[0034] See also Figure 5 , the specific steps for obtaining the interest vector weight adjustment value are: S401: Based on the interest vector primitive set, calling the advertisement feedback data, determining the influence of multiple interest features, calculating the influence factors of multiple interest features according to the advertisement exposure times, click-through rate, and user stay time, and calculating the weight correction value according to the feature influence factor, adjusting the correction parameter in combination with the performance of the differentiated features in advertisement delivery, and obtaining the interest feature weight correction matrix; First, we call the advertising feedback data. This process involves collecting historical data on user-ad interaction from different advertising delivery platforms, including ad exposure times, click-through rates, and user dwell time. Assuming that an ad is exposed 10,000 times, with a click-through rate of 3.5% and an average user dwell time of 8 seconds, these data need to be standardized so that they are within the same numerical range for further calculations. By setting the standardized interval of exposure times to , the minimum-maximum normalization method can be used for conversion, and the normalization formula is: ; Assuming that the minimum value of the number of exposures is 5000 and the maximum value is 50000, the normalized value of the number of exposures is calculated as follows: ; Similarly, the click rate and user dwell time are normalized to enter the same numerical space. The normalized click rate and dwell time are 0.35 and 0.25 respectively. Next, the influence factors of multiple interest features are calculated based on these normalized values. The calculation of the influence factors can be done by weighted summation. Assume that the weight coefficients are , then the impact factor The calculation is as follows: ; ; Subsequently, the weight correction value is calculated based on the influence factor of the interest feature. The weight correction value is dynamically adjusted by comparing the influence factor with the preset benchmark value. Assuming the benchmark value is 0.2, when the influence factor is greater than the benchmark value, the weight correction value is set to , otherwise it is set to zero, so for this example: ; Combined with the performance of differentiated features in advertising, the weighted average method is used to adjust the correction parameters. Assuming that the conversion rate of an advertisement in a specific population is 5% higher than the average conversion rate, the correction value is readjusted to make it more in line with the interest preferences of the target users. Finally, the interest feature weight correction matrix is ​​obtained, as shown in Table 4.

[0035] Table 4 Interest feature weight correction matrix As shown in Table 4, the weight correction value of the interest feature is calculated based on the advertising feedback data, and is adjusted on this basis to optimize the matching degree of the interest vector, thereby improving the accuracy of advertising delivery.

[0036] S402: Call the interest feature weight correction matrix to update the interest vector distribution using the formula: ; Obtain updated interest vector distribution through calculation, filter out long-term stable advertising interest features, and integrate them into a long-term advertising interest feature set; in, represents the updated interest vector distribution, represents the original interest vector, Represents the weight correction value in the interest feature correction matrix, Represents the influencing factors of multiple interest features in advertising feedback data, represents the total number of features of interest, The modulus length of the original interest vector is increased by 1 to balance the adjustment amplitude of the interest feature; First, get the interest vector The original value of , assuming that the user's original interest vector is , the weight correction value in the correction matrix As shown in Table 4, the impact factor Calculated from the previous steps, using the given formula: ; Perform update calculations, assuming that the total number of features of interest is , calculate the denominator: ; Calculate the numerator part: ; ; Finally, the updated interest vector distribution is calculated: ; After the interest vector distribution is updated, a threshold is set to filter the long-term stable advertising interest features. Assuming the stability threshold is 0.4, when When it is greater than 0.4, the interest feature is included in the long-term advertising interest feature set, otherwise it is removed. , so the interest features are retained and integrated into the long-term advertising interest feature set.

[0037] S403: Based on the long-term advertising interest feature set, calculate the overall criticality of multiple interest features, call the updated interest vector distribution, calculate the overall weight of the interest feature according to the mean, variance and influencing factor of the interest feature in the differentiated advertising delivery cycle, and adjust the calculated distribution of the interest vector in combination with the weight information to obtain the interest vector weight adjustment value.

[0038] First, call the updated interest vector distribution, as calculated above , the overall weight of the interest feature is calculated based on the mean, variance and influencing factor of the interest feature in different advertising delivery cycles. Assuming that in the historical advertising delivery cycle, the mean of the interest feature is 0.38 and the variance is 0.02, the overall criticality can be calculated by weighting: ; Among them, the weight coefficient is set , calculated: ; ; Finally, the interest vector weight adjustment value is calculated according to the overall criticality, and the calculation distribution of the interest vector is adjusted based on the weight information, resulting in a final interest vector weight adjustment value of 0.320675, which is used to further optimize the advertising delivery strategy and make the interest vector more accurate in practical applications.

[0039] See also Figure 6 , the steps for obtaining the interest vector reconstruction value are as follows: S501: Based on the interest vector weight adjustment value, call the advertisement audience interaction behavior fit calculation factor, calculate the fit values ​​between all advertisement audiences and the current advertisement audience, compare the fit data, select the advertisement audience with the highest fit, extract its identification information, call the matching algorithm for verification, and determine the matching value of the same type of advertisement audience; It is necessary to call the advertising audience interaction behavior fit calculation factor and obtain the interaction behavior data of all advertising audiences from the database. The behavior data of each advertising audience includes multiple dimensions such as the number of ad clicks, dwell time, conversion rate, etc. For these dimensional parameters, calculate their means and standard deviations respectively, and normalize the data through standardization methods. Subsequently, compare the current advertising audience's behavior data with all advertising audience data one by one, calculate the Euclidean distance between the two, and define the one with the smaller distance as the object with higher fit. On this basis, set the fit threshold to screen out the advertising audience with the highest fit and extract its identification information, such as user ID, device ID or encrypted identification. Then, call the matching algorithm to perform secondary matching on the interest characteristics of the advertising audience and the current advertising audience, extract all its advertising interest characteristics, and calculate the cosine similarity based on these interest characteristics and the interest vector of the current advertising audience to obtain the similarity score, and finally determine the matching value of the same type of advertising audience.

[0040] To further illustrate the calculation process, assume that the database stores the interaction data of the following three advertising audiences: Table 5 Advertising audience interaction behavior data table First, calculate the Euclidean distance between each ad audience and the current ad audience: ; ; ; It can be seen that the advertising audience B has the highest fit with the current advertising audience, so B is selected as the matching object, and its interest feature data is extracted, and the matching weight of the advertising interest feature is subsequently calculated.

[0041] S502: Call the matching value of the same type of advertising audience, extract all advertising interest features of the corresponding advertising audience, calculate the matching weight of the advertising interest features, and select the advertising interest features with the highest weight value, using the formula: ; Adjust the current advertising audience interest vector, calculate and obtain the highest weight advertising interest feature influence; in, Represents the highest weight advertising interest feature influence, Represents the calculation factor of the fit degree of the interactive behavior of the advertising audience, Represents the weight of advertising interest features, Represents the interest characteristic value of the most similar advertising audience, Represents the interest feature value of the current advertising audience. The total number of characteristics representing advertising interests. Represents the advertising behavior characteristic value of the same type of advertising audience, Represents the advertising behavior characteristic value of the current advertising audience, The total number of advertising behavior characteristics; The specific steps include: 1. Read the advertising interest feature data of the matching advertising audience, which includes multiple dimensions such as keywords, browsing categories, and click frequency, cross-compare these interest features with the interest features of the current advertising audience, and count the common interest feature items; 2. Calculate the matching weight of each interest feature using the formula: in: Represents the calculation factors of the fit of the advertising audience's interactive behavior, such as click-through rate, browsing time, etc. Represents the weight of advertising interest features, and its value is calculated by weighting the advertising click volume and purchase behavior data; and Represent the interest feature values ​​of the matching advertising audience and the current advertising audience, such as the number of clicks on a certain category of advertising; and Respectively represent the advertising behavior characteristic values ​​of the matching advertising audience and the current advertising audience, such as advertising conversion rate; and Represents the total number of interest features and behavioral features respectively.

[0042] Assume that the interest characteristics of matching advertising audience B are as follows: Table 6 Advertising interest feature matching data table Calculate the matching weight: ; ; ; Through calculation, the advertising interest feature with the highest weight is "electronic products", and its influence is applied to the current advertising audience interest vector.

[0043] S503: Call the highest weighted advertising interest feature influence, adjust the current advertising audience interest vector, recalculate the weight values ​​for all interest features, analyze the interest vector structure after the weight adjustment, complete the missing interest feature information, and calculate and generate the interest vector reconstruction value.

[0044] The specific execution process includes: 1. Recalculate the weight values ​​of all interest features: Use the aforementioned matching weight calculation method to reassign all interest features of the current advertising audience and correct the original interest vector; 2. Analyze the interest vector structure after weight adjustment: Compare the interest distribution before and after the weight adjustment, calculate whether the proportion of each feature has changed significantly, and ensure that the key interest features remain stable; 3. Complete the missing interest feature information: Based on the interest data of the matching advertising audience, screen the missing but highly relevant interest features and fill them in; 4. Calculate the interest vector reconstruction value: Use the normalization method to regularize the completed interest vector and finally generate a new interest vector structure.

[0045] Assume that the original interest vector of the current advertising audience is as follows: ; After the adjustment, since electronic products have the highest impact, the weights are changed to: ; The final interest vector reconstruction value is: ; The results show that the influence weight of electronic product-related interest features on the current advertising audience increases. At the same time, after the original interest vector is adjusted, the overall interest distribution is more in line with the behavioral preferences of the matching advertising audience, and ensures that the sum of all interest features remains normalized, that is, the sum of all interest weights is 1.

[0046] A system for accurately analyzing the audience for advertising delivery, which is used to execute the above-mentioned method for accurately analyzing the audience for advertising delivery, and includes: The interactive data processing module obtains the interactive data of the advertising audience, calculates the change range compared with the benchmark interactive behavior, calls the scene trigger factor and the threshold for comparison and screening, selects the qualified data as the short-term behavior influence factor, and calculates and generates the short-term behavior influence coefficient based on the change trend of the short-term behavior influence factor; The audience attribution analysis module counts the attribution ratio of the advertising audience based on the short-term behavior influence coefficient, calculates the attribution probability, calls short-term advertising interaction data to filter and modify the weight, and obtains the audience attribution drift probability; The interest feature clustering module screens the initial interaction data of the advertising audience based on the audience attribution drift probability, removes outliers, calls the advertising interaction data type parameters to assign differentiated weights, performs clustering analysis, calculates the change rate of short-term interaction behavior, and generates an interest vector primitive set; The interest vector optimization module calls the advertising feedback data based on the interest vector primitive set, determines the feature impact, calculates the interest feature weight correction value, updates the interest vector distribution, screens the long-term advertising interest features, and calculates and generates the interest vector weight adjustment value; The audience interest matching module adjusts the interest vector weight value, calls the advertising audience interaction behavior fitting calculation factor, matches the most similar highly active advertising audience, selects the highest weighted advertising interest feature, infers and completes the current advertising audience interest vector, and generates an interest vector reconstruction value.

[0047] 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 method for accurately analyzing the audience of advertisement delivery, characterized in that: The following steps are involved: S1: Obtain the interactive data of the advertising audience, calculate the change compared with the baseline behavior, call the scenario trigger factor and compare and filter with the threshold, and obtain the short-term behavior impact coefficient; S2: Based on the short-term behavior influence coefficient, the attribution ratio of the advertising audience is counted, the attribution probability is calculated, and the short-term advertising interaction data is called to filter and correct the weight to obtain the audience attribution drift probability; S3: Based on the audience attribution drift probability, the initial interactive behavior data of the advertising audience is screened, outliers are removed, advertising interactive data type parameters are called, differentiated weights are assigned, cluster analysis is performed, interest feature primitives are screened, and the short-term interactive behavior change rate is calculated to obtain an interest vector primitive set; S4: Based on the interest vector primitive set, calling the advertisement feedback data to determine the feature impact, calculating the interest feature weight correction value, updating the interest vector distribution, screening the long-term advertisement interest feature, and calculating the interest vector weight adjustment value; S5: Based on the interest vector weight adjustment value, call the advertising audience interaction behavior similarity calculation factor, match the most similar highly active advertising audience, select its highest weighted advertising interest feature, infer and complete the current advertising audience interest vector, and obtain the interest vector reconstruction value.

2. The method for accurate analysis of advertising audience according to claim 1, characterized in that: The short-term behavior impact coefficient includes the change amplitude, scene trigger factor, and threshold comparison screening result; the audience attribution drift probability includes attribution proportion, attribution probability, and short-term advertising interaction data screening correction weight; the interest vector primitive set includes interest feature primitives, short-term interactive behavior change rate, and differentiation weight; the long-term advertising interest feature includes interest vector distribution, interest feature weight correction value, and advertising feedback data judgment feature influence; the interest vector reconstruction value includes interest vector weight adjustment value, advertising audience interaction behavior similarity calculation factor, and highest weight advertising interest feature.

3. The method for accurate analysis of advertising audience according to claim 2, characterized in that: The steps for obtaining the short-term behavior influence coefficient are specifically as follows: S101: Obtaining interaction data of the advertising audience, extracting multiple interaction parameters such as click-through rate, browsing time, and bounce rate, calculating the average value and standard deviation of the multiple interaction parameters in the delivery cycle, establishing a data distribution model, calling the benchmark behavior data, calculating the deviation of the interaction parameters compared to the benchmark behavior, and obtaining the interaction parameter deviation value; S102: Based on the interaction parameter offset value, the fluctuation trend of multiple parameters is calculated, the short-term variation range of multiple parameters is calculated by using the offset mean and variance, and compared with the variation range of the benchmark behavior, the fluctuation level of multiple parameters compared with the benchmark is determined, and the fluctuation intensity of the interaction parameters is calculated; S103: calling the interaction parameter fluctuation intensity, matching it with the scene trigger factor and the threshold, screening the behavior change items that meet the conditions, and calculating the proportion of their impact on short-term decision-making using the formula: ； Get the short-term behavior impact coefficient; in, represents the short-term behavior impact coefficient, Representative The current value of the interaction parameter, Representative The baseline values ​​of the interaction parameters, Representative The standard deviation of the interaction parameter, Representative The weights of the interaction parameters, Represents the total number of interaction parameters.

4. The method for accurate analysis of advertising audience according to claim 3, characterized in that: The steps for obtaining the audience attribution drift probability are specifically as follows: S201: Based on the short-term behavior impact coefficient, the attribution ratio of the audience for advertising delivery is counted, the distribution of multiple audience groups under differentiated advertising categories is calculated, the data source is analyzed, abnormal samples are removed, the audience ratio of multiple categories is normalized, data that meets the statistical standards is screened, the audience category distribution is calculated according to the normalized ratio, the standardized ratio value is obtained, and the audience attribution ratio matrix is ​​constructed; S202: calling the audience attribution ratio matrix, and setting the audience behavior correction weight according to the normalized ratio and the short-term advertising interaction data, using the formula: ； Calculate the audience's attribution probability under multiple advertising categories and generate an audience attribution probability matrix; in, Representative audience In the advertising category The probability of belonging to For the audience In the advertising category The proportion of attributable Representative audience The short-term behavior impact coefficient, For the audience In the Ad Interaction Data category The interaction strength under Represents interaction category For advertising categories The influence weight of is the total number of interaction categories, is the total number of advertising categories; S203: calling the audience attribution probability matrix, screening short-term advertising interaction data to correct weights, analyzing audience short-term interaction behaviors, setting attribution change thresholds, calculating audience weight adjustment values ​​under differentiated advertising categories, and obtaining audience attribution drift probability in combination with short-term interaction characteristics.

5. The method for accurate analysis of advertising audience according to claim 4, characterized in that: The steps of obtaining the interest vector primitive set are specifically as follows: S301: Based on the initial interactive behavior data of the advertising audience, screen the data samples that meet the audience attribution drift probability standard, remove the abnormal values ​​that do not meet the standard, assign differentiated weights to multiple data types according to the advertising interaction data type parameters, calculate the weighted interaction frequencies of the multiple data types, and obtain the weighted interaction frequency value; S302: calling the weighted interaction frequency value, using a cluster analysis method, grouping data according to the interaction frequency distribution pattern, screening representative behavior feature data points in each category, and obtaining interest feature primitives; S303: Based on the interest feature primitives, the short-term interaction behavior change rate is calculated, and the interaction change trend under the differentiated time window is analyzed, using the formula: ； Get interest vector primitive set; in, represents the interest vector primitive set, Representative The interaction frequency value at each time point, Representative A point in time, represents the time mean, represents the total number of data points, Calculates the total amount of interactive change over a short time window.

6. The method for accurate analysis of advertising audience according to claim 5, characterized in that: The step of obtaining the interest vector weight adjustment value is specifically as follows: S401: Based on the interest vector primitive set, calling the advertisement feedback data, determining the influence of multiple interest features, calculating the influence factors of multiple interest features according to the advertisement exposure times, click-through rate, and user stay time, and calculating the weight correction value according to the feature influence factor, adjusting the correction parameter in combination with the performance of the differentiated features in the advertisement delivery, and obtaining the interest feature weight correction matrix; S402: calling the interest feature weight correction matrix to update the interest vector distribution, using the formula: ； Obtain updated interest vector distribution through calculation, filter out long-term stable advertising interest features, and integrate them into a long-term advertising interest feature set; in, represents the updated interest vector distribution, represents the original interest vector, Represents the weight correction value in the interest feature correction matrix, Represents the influencing factors of multiple interest features in advertising feedback data, represents the total number of features of interest, The modulus length of the original interest vector is increased by 1 to balance the adjustment amplitude of the interest feature; S403: Based on the long-term advertising interest feature set, the overall criticality of multiple interest features is calculated, and the updated interest vector distribution is called. According to the mean, variance and influencing factor of the interest features in the differentiated advertising delivery cycle, the overall weight of the interest features is calculated, and the calculated distribution of the interest vector is adjusted in combination with the weight information to obtain the interest vector weight adjustment value.

7. The method for accurate analysis of advertising audience according to claim 6, characterized in that: The steps for obtaining the interest vector reconstruction value are specifically as follows: S501: Based on the interest vector weight adjustment value, call the advertisement audience interaction behavior fit calculation factor, calculate the fit values ​​between all advertisement audiences and the current advertisement audience, compare the fit data, select the advertisement audience with the highest fit, extract its identification information, call the matching algorithm for verification, and determine the matching value of the same type of advertisement audience; S502: Call the matching value of the same type of advertising audience, extract all advertising interest features of the corresponding advertising audience, calculate the matching weight of the advertising interest features, and select the advertising interest features with the highest weight value, using the formula: ； Adjust the current advertising audience interest vector, calculate and obtain the highest weight advertising interest feature influence; in, Represents the highest weight advertising interest feature influence, Represents the calculation factor of the fit degree of the interactive behavior of the advertising audience, Represents the weight of advertising interest features, Represents the interest characteristic value of the most similar advertising audience, Represents the interest feature value of the current advertising audience. The total number of characteristics representing advertising interests, Represents the advertising behavior characteristic value of the same type of advertising audience, Represents the advertising behavior characteristic value of the current advertising audience, The total number of characteristics representing advertising behavior; S503: Call the highest weighted advertising interest feature influence, adjust the current advertising audience interest vector, recalculate weight values ​​for all interest features, analyze the interest vector structure after weight adjustment, complete missing interest feature information, and calculate and generate an interest vector reconstruction value.

8. An accurate analysis system for advertising audiences, characterized in that: According to the method for accurately analyzing the advertising audience according to any one of claims 1 to 7, the system comprises: The interactive data processing module obtains the interactive data of the advertising audience, calculates the change range compared with the benchmark interactive behavior, calls the scene trigger factor and the threshold for comparison and screening, selects the qualified data as the short-term behavior influence factor, and calculates and generates the short-term behavior influence coefficient based on the change trend of the short-term behavior influence factor; The audience attribution analysis module counts the attribution proportion of the advertising audience based on the short-term behavior influence coefficient, calculates the attribution probability, calls the short-term advertising interaction data to filter and correct the weight, and obtains the audience attribution drift probability; The interest feature clustering module screens the initial interaction data of the advertising audience based on the audience attribution drift probability, removes outliers, calls the advertisement interaction data type parameters to assign differentiated weights, performs cluster analysis, calculates the short-term interaction behavior change rate, and generates an interest vector primitive set; The interest vector optimization module calls the advertisement feedback data based on the interest vector primitive set, determines the feature impact, calculates the interest feature weight correction value, updates the interest vector distribution, screens the long-term advertisement interest feature, and calculates and generates the interest vector weight adjustment value; The audience interest matching module calls the advertising audience interaction behavior fitting calculation factor based on the interest vector weight adjustment value, matches the most similar highly active advertising audience, selects the highest weighted advertising interest feature, infers and completes the current advertising audience interest vector, and generates an interest vector reconstruction value.

Images