Research On User Profiling Recommendation Algorithm Based On E-commerce

With the rapid development of mobile e-commerce,commodity information and user behavior data have shown an explosive growth trend,leading to the emergence of "information overload".The proposal of a personalized recommendation system has alleviated this problem to a large extent.The personalized recommendation system aims to solve the"information overload" and analyzes user behavior data to mine users'interests and preferences and make active recommendations.Although the research on recommendation algorithms has made great progress,there are certain challenges in data sparsity,static nature of user characteristics,and interpretability of recommendation results.Therefore,in the context of e-commerce,this paper combines real e-commerce data to study the data sparsity of e-commerce platform,the interpretability of recommendation results,and the static nature of user characteristics.A hybrid recommendation algorithm based on forgetting curve and automatic feature construction is carried out.The main research work is as follows:1)Summarize the theory and technology of current popular recommendation systems,introduce in detail the data sets and data preprocessing methods used in this article,deeply analyze the challenges and existing problems of current recommendation systems,and give targeted optimization solutions.2)Aiming at the interpretability of recommendation results,this paper proposes a cascade model based on automatic feature construction algorithm,which has strong interpretability for user features and optimizes the calculation speed of the algorithm model to a certain extent.3)Aiming at the problem of data sparseness and staticity of user characteristics,this paper proposes a hybrid recommendation algorithm based on user dynamic interest factors.The algorithm first constructs a user behavior profile based on the user's implicit feedback,and then according to Ebbinghaus The periodic law of the forgetting curve constructs a dynamic interest factor matrix.When predicting the user's purchase results,the weight of the prediction results is increased according to the distance of the user's behavior time,and a hybrid recommendation model that integrates the user's dynamic interest factors is realized.This hybrid model is compared with other models in this article.Recommendation model has better interpretability and accuracy of recommendation results.