Intelligent advertisement promotion system based on time-phased user behavior analysis

Abstract

The invention relates to an intelligent advertisement promotion system based on time-phased user behavior analysis. The system comprises a user behavior data acquisition module, a time sequence analysis module, an advertisement content matching module, a time-phased delivery module and a feedback optimization module. By collecting user data and through a time sequence analysis technology, behavior characteristics of a user in different time periods are mined, and the field of interest of the user is accurately recognized; dynamically screening and matching the optimal advertisement content according to the preference and behavior of the user in the current time period; and the advertisement putting efficiency and the conversion rate are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of advertising promotion, and particularly relates to an intelligent advertising promotion system based on time - segmented user behavior analysis. Background Art

[0002] With the rapid development of the advertising market, precise advertising placement has become an important means to improve conversion rates. However, there are many deficiencies in the traditional advertising promotion methods in terms of strategy design and technical implementation: First, the time factor is ignored. Existing advertising strategies are usually based on static user portraits, relying on the overall characteristics and long - term preferences of users, lacking the ability to dynamically capture the behavioral changes of users at different times, resulting in the disconnection between advertising content and users' actual needs. Second, the population segmentation is insufficient. Although some advertising strategies stratify users through dimensions such as age and region, they ignore the dynamic stratification in the time dimension and cannot meet the personalized needs of users at different times, resulting in the missed promotion opportunities. Third, the optimization of placement effects is insufficient. The real - time monitoring and feedback mechanism of advertising placement effects is imperfect, making it difficult to adjust placement strategies in a timely manner, resulting in a low advertising conversion rate.

[0003] A large amount of data analysis shows that there are significant changes in users' consumption behaviors in the time dimension, with obvious differences in consumption preferences at different times, and the consumption habits of different user groups also show diversity in the time dimension. Currently, this dynamic change based on the time dimension and user group characteristics indicates that single - and static - advertising strategies are difficult to effectively meet users' needs and also difficult to fully utilize advertising promotion opportunities during high - value periods. It is necessary to develop an intelligent advertising promotion system based on time - segmented user behavior analysis. By dynamically analyzing the behavioral changes of users in the time dimension and combining the characteristics of user groups, formulating time - segmented precise advertising placement strategies has become the key technical path to improve advertising placement efficiency and conversion rates. Summary of the Invention

[0004] In order to solve the current situation of the lack of dynamic analysis in the time dimension, low matching degree between advertising content and user needs, and untimely optimization of placement effects in the existing advertising promotion process, the present invention proposes an intelligent advertising promotion system based on time - segmented user behavior analysis; to achieve the above - mentioned purpose, the technical solutions adopted by the present invention are specifically as follows: An intelligent advertising promotion system based on time - segmented user behavior analysis, the system includes a user behavior data collection module, a time - series analysis module, an advertising content matching module, a time - segmented placement module, and a feedback optimization module; the five modules transmit data in sequence; Data collection module: The module collects behavior data such as users' browsing, clicking, and purchasing records through platform user behavior and records the specific time dimension when they occur; Time series analysis module: The time series analysis module receives the data collected by the data collection module, and uses a long short-term memory network (LSTM) model to analyze the user's behavior dynamics at different time periods, and analyzes the time period feature data of the user's behavior; Advertising content matching module: The advertising content matching module receives the time period feature data of the user's behavior, and based on the interest preferences of the user's behavior time period, screens the matching advertising content from the advertising library carried by the advertising content matching module itself, and generates the data of the placement plan; Time period placement module: The time period placement module receives the data of the placement plan and pushes the matching advertising content during the user's active time period; Feedback optimization module: According to the click and conversion data of the customers who push the matching advertising content collected by the data collection module, analyze the push effect, and dynamically optimize the effect data to form data.

[0005] Further, when the data collection module collects data, it is necessary to perform uniqueness detection based on the combination key of the user ID, timestamp, and behavior type, and delete redundant records.

[0006] Further, the data collected by the data collection module needs to be normalized, and the feature values of the collected data are scaled to the range of 0 to 1; the formula is as follows: : Original feature value; : Minimum value of the feature; : Maximum value of the feature; : Value after normalization.

[0007] Further, the time series analysis module organizes the data into a sequence in chronological order to form an input matrix, and the matrix dimension is "number of time steps * number of features"; a time series feature matrix is formed.

[0008] Further, input the time series feature matrix into the LSTM model, capture the long-term dependencies and dynamic changes of the user's behavior in the time dimension through the forget gate, input gate, and output gate, and the model predicts the domain preference distribution of each time window based on the data, and outputs the domain preference distribution of each time window.

[0009] Further, the advertising content in the advertising content matching module is added with domain labels according to categories to identify the target domain to which the advertisement belongs.

[0010] Further, vectorize the advertisements with domain labels. The vectorization process: a) Define the dimension of vectorization, b) Define the assignment of vector values, c) Output the vectorized result of the advertisement.

[0011] Further, based on the cosine similarity model, calculate the cosine similarity and sort the advertisements, and select the advertisement content. The formula is as follows: : User preference vector; : Advertisement tag vector; : Magnitude of the user preference vector; : Magnitude of the advertisement tag vector.

[0012] Advantages of the present invention: The system includes a user behavior data acquisition module, a time series analysis module, an advertisement content matching module, a time - segmented delivery module, and a feedback optimization module. By collecting user data and using time series analysis technology, it mines the behavioral characteristics of users at different time periods, accurately identifies their interest areas, dynamically filters and matches the optimal advertisement content according to the preferences and behaviors of users in the current time period, which is the key technical path to improve the advertisement delivery efficiency and conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is the architecture diagram of the intelligent advertisement promotion system based on time - segmented user behavior analysis of the present invention; Figure 2 is the time series analysis flow chart of the intelligent advertisement promotion system based on time - segmented user behavior analysis of the present invention; Figure 3 is the advertisement content matching flow chart of the intelligent advertisement promotion system based on time - segmented user behavior analysis of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0014] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: As shown in the architecture diagram of the embodiment of the intelligent advertisement promotion system based on time - segmented user behavior analysis of the present invention, it includes a user behavior data acquisition module, a time series analysis module, an advertisement content matching module, a time - segmented delivery module, and a feedback optimization module. The five modules transmit data in sequence.

[0015] Data acquisition module: a Collect user browsing, clicking, and purchasing record and other behavioral data through the platform (such as e - commerce, short - video platforms, etc.) user behavior, and record the specific time dimension when they occur; b Perform uniqueness detection based on the combined key of user ID, time dimension, and behavior type, and delete redundant records. That is, deduplicate the collected behavioral data, eliminate redundant data caused by possible network latency and user repeated operations, remove false data, and improve the data processing efficiency.

[0016] Normalize the characteristic values to scale the eigenvalues of the collected data to the range of 0 to 1. Eliminate the large differences that may exist in the units and ranges of the numerical characteristics of user behavior data (such as the number of views, click weights, purchase amounts). For example, the number of clicks is 2 times, but the purchase amount is 2000 yuan, which may cause the analysis model to be unable to accurately measure the contribution of click records.

[0017] The formula used is: : The original eigenvalue; : The minimum value of the feature; : The maximum value of the feature; : The value after normalization.

[0018] Time series analysis module: Its flowchart is as Figure 2 shown.

[0019] a Divide the user behavior data into time windows, and divide a day into several windows according to hours, time periods (such as morning, afternoon, evening) or the time periods when users are active.

[0020] b In each time window, count the eigenvalue features such as the number of views, clicks, and purchase amounts of users.

[0021] c Normalize the statistical results to eliminate the influence of eigenvalue range differences on model analysis.

[0022] d The time series analysis module organizes the data into a sequence in chronological order to form an input matrix, and the matrix dimension is "number of time steps * number of features"; form a time series feature matrix.

[0023] e The time series analysis module receives the data collected by the data collection module, and uses the long short-term memory network (LSTM) model to analyze the behavior dynamics of users in different time periods, and analyzes the time period feature data of user behavior; the specific process is to capture the long-term dependence and dynamic changes of user behavior in the time dimension through the forget gate, input gate and output gate. The model predicts the domain preference distribution of each time window according to the data, and outputs the domain preference distribution of each time window. Among them, the domain preference distribution is a probability vector, indicating the user's preference for different advertising domains in each time window. The domain preference distribution of each time window can be regarded as the interest vector of the user in the current time period, that is, the user preference vector.

[0024] Advertising content matching module: The content matching flowchart is as Figure 3 shown; the advertising content matching module receives the time period feature data of user behavior, and according to the interest preferences of the time period of user behavior, screens the matching advertising content from the advertising library carried by the advertising content matching module itself, and generates the data of the placement plan.

[0025] Add domain tags to the advertising content according to categories to identify the target domain to which the advertisement belongs. For example, advertisement 1 is for fashion and clothing, and advertisement 2 is for electronic products. Each advertisement supports multiple domain tag annotations to adapt to the diverse interest preferences of users. In other instances, other methods (such as identifying based on content or generation method) can be adopted.

[0026] Vectorize the tagged advertisements, and the process is as follows: a Define the dimension of vectorization: The dimension of the advertisement tag vector is the same as the total number of all domain tags in the advertisement library. For example, if the domain tag set of the advertisement library is 5 tags: food, healthy drinks, electronic products, fashion and clothing, entertainment, then the vector dimension is 5, and each dimension corresponds to a tag; b Define the assignment rule of vector values: When the advertisement belongs to a certain domain, the value of the corresponding dimension is 1; When the advertisement belongs to multiple domains, the value of the corresponding dimension is evenly divided according to the number of tags; The dimension value that does not belong to a certain domain is 0.

[0027] c Output the vectorization result of the advertisement. For example, the tags of advertisement 1 are: [healthy drinks, food], and the vectorization result is: .

[0028] Based on the cosine similarity model, calculate the matching degree between the user's interest domain preferences at different times and the advertisement tags respectively, where the cosine similarity is used to measure the similarity between the user preference vector and the advertisement tag vector. Sort the advertisements according to the calculated cosine similarity and select the advertisement content; The formula is as follows: : User preference vector; : Advertisement tag vector; : The norm of the user preference vector; : The norm of the advertisement tag vector.

[0029] Sort the advertisements according to the calculated cosine similarity, and select the advertisement content with the highest similarity as the optimal recommendation for the current time period. Repeat this process for different time periods in turn to calculate the optimal recommendation for each time period.

[0030] Time-slot delivery module: The time-slot delivery module receives the data of the delivery plan and pushes the matching advertisement content during the time periods when the user is active.

[0031] Feedback optimization module: Collect key metrics such as click-through rate, conversion rate, and ad interaction time in real time from ad placements. Classify the ad performance data according to the user behavior active time periods (such as morning, afternoon, evening), analyze the ad placement performance in different time periods, and according to the effect evaluation results, classify the ads into three types: good performance, average performance, and poor performance. Analyze the ads with poor performance in combination with the time-segmented user preference data to draw relevant conclusions, such as: the ad content does not match the time segment preference, the ad placement time period is unreasonable, etc. Based on the time-segmented effect evaluation results of ad placements, conduct strategy optimization, such as: a Adjust the recommended fields of the ads; b Reduce the number of ad displays in inactive time periods to save budget; Layer optimize the target audience of the ads according to the real-time changes in user behavior data.

[0032] The system includes a user behavior data collection module, a time series analysis module, an ad content matching module, a time-segmented placement module, and a feedback optimization module; by collecting user data, using time series analysis technology, mining the behavior characteristics of users in different time periods, accurately identifying their interest fields; dynamically screening and matching the optimal ad content according to the preferences and behaviors of users in the current time period; the key technical path to improve ad placement efficiency and conversion rate.

[0033] In the present invention, unless otherwise clearly specified and defined, terms such as "installation", "connection", "connection", "fixation", etc. shall be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0034] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and substitutions can be made, and these improvements and substitutions should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent advertising promotion system based on time-divided user behavior analysis, characterized in that: The system includes a user behavior data collection module, a time series analysis module, an advertising content matching module, a time-slot delivery module, and a feedback optimization module; the five modules transmit data in sequence; Data collection module: collects user browsing, clicking and purchasing records and other behavior data through platform user behavior, and records the specific time dimension of its occurrence; Time series analysis module: The time series analysis module receives the data collected by the data collection module, and uses the long short-term memory network (LSTM) model to analyze the user's behavior dynamics in different time periods, and analyzes the time period characteristic data of the user's behavior; Advertisement content matching module: The advertisement content matching module receives the time period characteristic data of the user's behavior, selects matching advertisement content from the advertisement library carried by the advertisement content matching module itself according to the interest preference of the user's behavior time period, and generates data for the delivery plan; Time-divided delivery module: The time-divided delivery module receives data of the delivery plan and pushes matching advertising content during the user's active time period; Feedback optimization module: collects click and conversion data of customers of push matching advertising content according to the data collection module, analyzes the push effect, and dynamically optimizes the effect data to form data.

2. The intelligent advertising promotion system based on time-divided user behavior analysis according to claim 1 is characterized in that: The data collection module collects data and performs uniqueness detection based on the combination key of user ID, timestamp and behavior type, and deletes redundant records.

3. The intelligent advertising promotion system based on time-divided user behavior analysis according to claim 2 is characterized in that: The data collected by the data acquisition module needs to be normalized to scale the characteristic values ​​of the collected data to the range of 0 to 1; the formula is as follows: : original eigenvalue; : minimum value of the feature; : The maximum value of the feature; : Normalized value.

4. The intelligent advertising promotion system based on time-divided user behavior analysis according to claims 1 to 3 is characterized in that: The time series analysis module organizes the data into a sequence in chronological order to form an input matrix whose dimension is "number of time steps*number of features" to form a time series feature matrix.

5. The intelligent advertising promotion system based on time-divided user behavior analysis according to claim 4 is characterized in that: The time series feature matrix is ​​input into the LSTM model, and the long-term dependence and dynamic changes of user behavior in the time dimension are captured through the forget gate, input gate and output gate. The model predicts the domain preference distribution of each time window based on the data and outputs the domain preference distribution of each time window.

6. The intelligent advertising promotion system based on time-divided user behavior analysis according to claim 1 is characterized in that: The advertisement content in the advertisement content matching module is added with domain tags according to categories, so as to identify the target domain to which the advertisement belongs.

7. The intelligent advertising promotion system based on time-divided user behavior analysis according to claim 6 is characterized in that: The advertisements with domain labels are vectorized. The vectorization process is as follows: a) define the vectorization dimension, b) define the assignment of vector values, and c) output the advertisement vectorization results.

8. The intelligent advertising promotion system based on time-divided user behavior analysis according to claim 7 is characterized in that: Based on the cosine similarity model, the cosine similarity is calculated to sort the advertisements and select the advertisement content; the formula used is as follows: : user preference vector; : advertisement label vector; : The modulus of the user preference vector; : The modulus of the ad label vector.

Images