Search

CN-121981757-A - Accurate marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning

CN121981757ACN 121981757 ACN121981757 ACN 121981757ACN-121981757-A

Abstract

The invention discloses a precise marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning, and relates to the technical field of intelligent marketing. The invention integrates behavior pattern clustering and motivation depth deducing technology, ensures data quality through multi-source data acquisition and intelligent cleaning, utilizes a clustering algorithm system to identify a user behavior pattern, combines an entropy method and a stacked self-encoder to excavate deep motivations, finally realizes accurate motivation identification by means of LightGBM classification, effectively solves the problem that the traditional method can only analyze surface behavior data but cannot understand the actual intention of a user, promotes the cognition of the user to realize deepening from behavior appearance to motivation essence, enables enterprises to accurately grasp the actual demand and price psychological expectation of the user, thereby remarkably improving the accuracy and user conversion rate of marketing activities and realizing the essence improvement of marketing effects.

Inventors

  • SONG GUANGNIAN
  • SONG JINGRU

Assignees

  • 成都艾迪梅斯科技有限责任公司

Dates

Publication Date
20260505
Application Date
20260225

Claims (10)

  1. 1. The accurate marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning is characterized by comprising the following components: The behavior pattern clustering module acquires user interaction records and browsing tracks in real time by acquiring a multi-source data channel, and performs grouping processing on the records and tracks by adopting a clustering algorithm to acquire user behavior pattern classification; The motivation depth deducing module extracts relevant dimension indexes from a preset feature library according to the classification of the user behavior modes, calculates a behavior complexity quantization value through an entropy method, activates the depth analyzing module if the complexity exceeds a threshold value, and obtains a deep motivation deducing result by stacking the self-encoders; The strategy dynamic mapping module is used for acquiring current market change data according to the deep motivation inference result, carrying out association mapping by fusing the hierarchical motivation inference result through a Hungary algorithm, and determining the instant strategy adjustment requirement; The pricing optimization deduction module is used for optimizing the reinforcement learning model through the near-end strategy if the instant strategy adjustment requirement exists, and simulating a multi-scene interaction path by combining Monte Carlo tree search to obtain an optimized pricing scheme; the quota efficiency configuration module is used for extracting key parameters from the optimized pricing scheme, judging the resource allocation efficiency by adopting an efficiency evaluation method, and adjusting the allocation proportion by a genetic algorithm to obtain quota configuration avoiding waste; The personalized touch enhancement module analyzes opportunity capturing probability through XGBoost probability evaluation models according to quota configuration, and adjusts touch logic by adopting thompson sampling if the probability is lower than a threshold value to obtain an enhanced personalized scheme; The effect monitoring and evaluating module is deployed into the marketing system through an enhanced personalized scheme, tracks the execution effect by adopting a real-time monitoring method, and judges the gain change by combining a trend analysis method to obtain a final gain optimization index.
  2. 2. The precision marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning of claim 1, wherein the behavior pattern clustering module comprises: Collecting user interaction records and browsing tracks in real time through a multi-source data channel, cleaning by adopting a pre-established filtering mechanism, detecting missing values or abnormal values in data through an isolated forest algorithm, and obtaining a processed behavior data set; Based on the processed behavior data set, grouping analysis is carried out on the user interaction records and the browsing tracks by using a clustering algorithm to obtain a preliminary classification result of the user behavior patterns, and then core interaction features and track features in each behavior pattern group are extracted to obtain behavior feature sets of each group; The pattern comparison is carried out on the behavior feature set, and the matching degree between the behavior feature set and the preset typical pattern features is measured by adopting a similarity calculation method, so that refined user behavior pattern classification is obtained; The method comprises the steps of classifying refined user behavior modes, analyzing the relevance among modes by using a graph neural network, judging the interaction frequency and track overlapping condition among the modes, constructing a dynamic relevance graph of the user behavior modes, identifying a main transition path in the user behavior modes by using a Markov chain model, calculating a state transition probability matrix, and determining the behavior migration rule of the user among different modes.
  3. 3. The precision marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning of claim 1, wherein the motivational depth inference module comprises: S11, classifying according to a user behavior mode, extracting a data source field corresponding to a key dimension index from a preset feature library to obtain a preliminary complexity evaluation result, calculating a quantized value of the behavior complexity by adopting an entropy method, and classifying the quantized value as a high-complexity behavior if the quantized value exceeds a preset threshold; S12, activating a processing flow of a depth analysis module according to the record of the high complex behavior, performing multi-layer feature decomposition through stacking the self-encoders to obtain potential motivation clues, and performing classification inference by adopting LightGBM classification algorithm in combination with logic mapping of deep motivation and motivation inference to judge the final motivation category; S13, generating a correlation record of behaviors and motivations according to the final motivation category, constructing a complete user behavior analysis file containing user ID, behavior characteristics and motivation category, automatically updating key dimension data weight in a preset feature library by adopting online gradient descent, and performing continuous iterative optimization.
  4. 4. The precision marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning of claim 3, wherein the multi-layer feature decomposition by stacking the self-encoders comprises constructing a depth network structure stacked by a plurality of self-encoders, compressing and abstracting the input high-dimensional features layer by the encoder part, extracting potential low-dimensional representations contained in the data, reconstructing the original input from the low-dimensional representations by the decoder part, learning the network to the most essential feature mode in the data by minimizing reconstruction errors, and taking the output of the final encoding layer as a potential motivation clue obtained by multi-layer nonlinear decomposition after the training is completed.
  5. 5. The precision marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning of claim 1, wherein the policy dynamic mapping module comprises: s21, acquiring latest market change data from a market monitoring system, processing the data by adopting a structured analysis technology to obtain the processed market dynamic information, carrying out data matching and association mapping by a Hungary algorithm in combination with a pre-established deep motivation inference result, constructing a market change and user motivation association matrix, and determining a preliminary fusion analysis result; S22, based on a preliminary fusion analysis result, adopting a C4.5 decision tree algorithm to carry out logic comparison, and if the variation trend in the fusion analysis result exceeds a preset threshold, extracting corresponding market variation key points to obtain trigger conditions of strategy adjustment; s23, acquiring historical decision data related to an instant strategy according to a trigger condition of strategy adjustment, and comparing the similarity of the current market change key point and the historical decision data by adopting a Jaccard similarity algorithm to judge the specific direction of strategy adjustment; S24, based on the specific direction of policy adjustment, combining with the priority rule in the service requirement, if the adjustment direction is consistent with the priority rule, generating a corresponding instant policy update scheme, determining the final adjustment content, S25, analyzing and adjusting the suitability of the content and the current market change trend by adopting a Bayesian network model, calculating the adaptation probability, automatically generating an execution instruction sequence by means of priority ordering of policy execution, and carrying out system distribution on the sequence by adopting a message queue mechanism to finish dynamic update of the instant policy.
  6. 6. The precise marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning of claim 5, wherein the data matching and association mapping are performed through a Hungary algorithm, and the market change and user motivation association matrix is constructed, wherein the market change and user motivation association matrix is constructed by taking market dynamic characteristics as row vectors and user motivation characteristics as column vectors, constructing benefit matrices by calculating synergy scores of various market-motivation combinations, and solving an optimal matching scheme within polynomial time by using the Hungary algorithm, so as to form a sparse association matrix showing a synergistic effect, wherein the matrix reflects relevant user motivation groups under specific market change.
  7. 7. The precision marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning of claim 1, wherein the pricing optimization deduction module comprises: acquiring dynamic change data in the current market environment by analyzing the association between the instant strategy and the adjustment requirement, determining the trigger condition of strategy adjustment based on logistic regression, screening core characteristics by a characteristic selection algorithm, and judging to start a reinforcement learning model for simulation; Adopting a near-end strategy optimization reinforcement learning model, carrying out simulation calculation on various scenes by combining Monte Carlo tree search, generating a plurality of groups of interaction path combinations, deducing potential benefit values of each path through a summer ratio, screening out a plurality of groups of interaction path combinations with optimal benefit performance according to the deduced potential benefit values, and adopting a weighted voting method to fuse results to determine an optimized pricing scheme suitable for the current market environment; If the difference between the optimized pricing scheme and the current pricing exceeds a preset threshold, applying the new scheme to a strategy updating module through a system automatic updating mechanism to obtain an adjusted execution basis, and carrying out data synchronization on a pricing adjustment link by adopting a distributed data synchronization technology to obtain a final pricing execution result; and recording market feedback data after each adjustment according to a final pricing execution result, updating reinforcement learning model parameters by adopting a time sequence differential learning algorithm, and judging whether the interaction path is required to be further optimized or not.
  8. 8. The precision marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning of claim 1, wherein the quota efficiency configuration module comprises: extracting key parameter data from the optimized pricing scheme, classifying each parameter, eliminating dimension influence through Z-Score standardization operation, analyzing the current state of resource allocation, and determining resource allocation items to be adjusted; Based on marked resource allocation items, adopting data envelope analysis to calculate the technical efficiency value of each allocation item, then adopting a hierarchical analysis method to analyze key points of waste control by combining historical data of quota configuration, and determining a redundant part in the quota configuration by judging matrix calculation weight; adjusting the allocation proportion of the quota allocation and redundancy part by adopting a genetic algorithm, maximizing the resource utilization efficiency as an objective function, analyzing the overall effect of resource utilization by adopting random forest regression, calculating an efficiency index, and determining a final quota allocation result; and generating an execution instruction of resource allocation through a final quota configuration result, and updating data of each resource utilization unit by adopting a caching technology to acquire an optimized resource allocation state.
  9. 9. The precision marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning of claim 1, wherein the personalized touch enhancement module comprises: Obtaining quota configuration data, carrying out structural processing on each configuration content, extracting key field information to obtain detailed distribution conditions of quota configuration, associating distribution characteristics of quota configuration with core indexes through XGBoost probability evaluation models, calculating capture probability values corresponding to each configuration, and determining probability evaluation results; Judging whether a preset threshold value is reached or not according to a probability evaluation result, adopting a Thompson sampling generation logic optimized primary scheme, combining a generation rule of a personalized scheme, and adopting a collaborative filtering algorithm to construct a customized strategy of an enhanced effect so as to obtain output content of the personalized scheme; And through the output content of the personalized scheme, adopting the pearson correlation coefficient to analyze the matching degree between configuration analysis and scheme generation, using a data comparison tool to verify the suitability of the scheme and quota configuration, judging the adaptation result and obtaining the final enhanced scheme.
  10. 10. The precision marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning of claim 1, wherein the effect monitoring assessment module comprises: Deploying the enhanced personalized scheme to a marketing system, completing configuration synchronization of scheme parameters through a system interface, collecting user interaction data in the scheme execution process from the marketing system in real time, and constructing a dynamic monitoring data set; performing time sequence segmentation on the dynamic monitoring data set by adopting a sliding window algorithm, calculating core execution indexes in each window, performing trend analysis on the indexes of each window by combining an index smoothing method, and predicting profit trend; comparing the predicted trend with a preset profit target, triggering an abnormal investigation mechanism if the actual index deviates from the target, positioning key influencing factors which do not reach the standard, integrating a window analysis result, trend prediction data and an abnormal investigation conclusion, and calculating a final profit optimization index to form a complete scheme effect evaluation report.

Description

Accurate marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning Technical Field The invention relates to the technical field of intelligent marketing, in particular to a precise marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning. Background In the current commercial environment, accurate marketing and dynamic pricing are key fields for improving the competitiveness of enterprises, and bear important missions for optimizing resource allocation and improving user experience. Along with the advancement of the digital wave, how to get insight into the demands of users and formulate the optimal strategies through technical means has become a core subject for enterprises to pursue growth, and the importance of the field is self-evident, and the importance of the field directly relates to whether enterprises can realize win-win of income maximization and user satisfaction in vigorous market competition. However, most solutions on the current market often look at the mind when dealing with complex user behaviors and market changes, many methods lack comprehensive understanding of deep motivations behind the user behaviors, and it is also difficult to flexibly adjust strategies under different scenes, especially when faced with instantaneous changes of user demands and multi-scene interactions, the existing methods often cannot achieve real-time response and strategy coordination, resulting in resource waste and opportunity missing. This limitation not only affects marketing effectiveness, but also limits the adaptability of the enterprise in dynamic environments. The challenge in a deeper level is how to achieve refinement of user value perception and immediacy of policy execution at the technical level, firstly, complexity of user behavior makes it difficult to capture real intention and potential requirement thereof simply by means of surface data, the complexity further evolves into urgent requirement for deep pattern mining of user behavior, and after deep requirement of user is mined, how to seamlessly link these insights with specific marketing and pricing policies, and the two problems are interwoven, the former determines accuracy of the insights, the latter affects efficiency of converting the insights into actual commercial value, for example, when a certain user browses goods, the system may recognize that the system is interested in a certain product, but due to unable to accurately judge price acceptance range and hesitation reasons, finally pushed preferential or information may completely miss targets, resulting in user loss. Disclosure of Invention The invention aims to provide a precise marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning, and through behavior pattern clustering and motivation depth inference technology, the cognition deepening from user behavior appearance to motivation essence is realized, and the marketing precision and conversion rate are remarkably improved. The aim of the invention can be achieved by the following technical scheme: The application provides a precise marketing and dynamic pricing integrated platform based on user behavior analysis and machine learning, which comprises the following components: The behavior pattern clustering module acquires user interaction records and browsing tracks in real time by acquiring a multi-source data channel, and performs grouping processing on the records and tracks by adopting a clustering algorithm to acquire user behavior pattern classification; The motivation depth deducing module extracts relevant dimension indexes from a preset feature library according to the classification of the user behavior modes, calculates a behavior complexity quantization value through an entropy method, activates the depth analyzing module if the complexity exceeds a threshold value, and obtains a deep motivation deducing result by stacking the self-encoders; The strategy dynamic mapping module is used for acquiring current market change data according to the deep motivation inference result, carrying out association mapping by fusing the hierarchical motivation inference result through a Hungary algorithm, and determining the instant strategy adjustment requirement; The pricing optimization deduction module is used for optimizing the reinforcement learning model through the near-end strategy if the instant strategy adjustment requirement exists, and simulating a multi-scene interaction path by combining Monte Carlo tree search to obtain an optimized pricing scheme; the quota efficiency configuration module is used for extracting key parameters from the optimized pricing scheme, judging the resource allocation efficiency by adopting an efficiency evaluation method, and adjusting the allocation proportion by a genetic algorithm to obtain quota configuration avoiding waste; The personalized touch enhan