Research Of Collaborative Recommendation Based On Total Surplus Maximization And Item Context Regularization

More and more product and transaction information is being generated in e-commerce,which makes it difficult for users to quickly find their favorite products.At the same time,e-commerce companies are faced with the problem that how to provide users with satisfactory products in time to improve sales.Recommender system has emerged to solve such problems.However,most recommender systems are based on the direct analysis of the data generated by users instead of analysis from economic perspective.A new recommendation algorithm is proposed by combining total surplus maximization in economics and item similarity context regularization.First,the item similarity matrix is constructed from user consumption record considering consumption association and time locality.Second,a personalized user utilization model is obtained by combining latent factor model and item similarity context regularization.We obtain objective function leveraging the Law of Zero Surplus for the Last Unit and train the model with user consumption records.Finally,user consumption prediction model is obtained by maximizing total surplus,thus maximizing consumer and producer surplus.To address the problem of sparse consumption matrix,item similarity context is taken into account of and bring potentially more accurate user utilization prediction,and better performance in user behavior prediction.The experiments are based on the data from Ta-Feng which is well-known grocery store website.Experimental results show that our model have a considerable improvement over the previous collaborative filtering model and Total Surplus Maximization model.