Research On Shopping Behavior Recognition Method Of Physical Stores Based On RFID And IMU Data Fusion

The shopping behavior data of physical store is of great significance for grasping the trend of the market and analyzing the preferences of users.Through effective shopping behavior data analysis,the marketing service strategy can be adjusted in a timely manner according to user preferences and market trends,and more effective personalized recommendation services can be provided to users.However,most of the existing methods analyze the overall trend and wind changes of the market in isolation from the perspective of commodities,and it is difficult to effectively associate users with interactive products,resulting in the inability to provide recommendation services that are more in line with personal preferences for user preferences.In this paper,it conducts research on the identification of shopping behavior of physical store,and proposes a shopping behavior recognition method for physical stores based on the fusion of RFID and IMU data.This method not only identifies the trend trend and user’s shopping intention,but also provides a practical solution to realize the correlation and matching of users and interactive goods,so as to achieve effective analysis of user preferences and shopping tendencies.Firstly,for popular commodities and trends,this paper implements the Seq-UT network architecture by introducing the recurrent time-step mechanism and the Seq Pooling module into the Transformer,and identifies and analyzes the RFID data change patterns generated by the passive interaction process of commodities to determine more popular commodities.Secondly,regarding the user’s purchase intention and the degree of rigid demand,this paper designs and implements a Seq-TNT network that simultaneously characterizes the sequential relationship of time series data and the correlation of different sequences by nesting the Transformer module.In this way,the purpose of identifying just-needed users with strong purchasing needs can be achieved.Finally,for the shopping tendency of different users,this paper proposes a skip-grambased method for calculating the correlation between user and commodity behavior,and combines the time-domain difference features to calculate the matching score.In this way,shopping preference information of different users can be obtained.In simulated shopping environment,the accuracy is over 87%.