Study On Recommendation Algorithm Based On User Consumption Behaviour

With the rapid development of Web2.0,the information in Internet is growing in a super high speed.Especially in the area of electronic commerce,mass merchandise appeared in various e-commerce sites.What ordinary consumers care about is how to get useful things for themselves from the mass of information,making life easier;Businesses need is to recommend their own products effectively to the appropriate groups,making business to maximize profits.In this context,the study of recommendation system,especially the study on recommendation algorithm based on user consumption behavior becomes very important.The main work of this paper include the following two aspects:1)The research of expenditure aware rating prediction for recommendation.The rating score prediction is widely studied in recommender system,many additional information have been employed to promote recommendations.Expenditure information on each transaction between users and items is widely available at E-commerce websites,this work try to use the expenditure given by the user to promote the performance of rating prediction.Through the analysis of the data,this work find three correlations between expenditures and rating scores.Based on these three correlations and matrix factorization method,this paper proposed the expenditure aware rating prediction for recommendation method.In addition,this work solves the problem of data sparsity and noise problems of expenditure by using the method of augmented matrix and expenditure discrete.Based on Dianping’s real dataset,this work conduct lots of experiments on the expenditure aware rating prediction for recommendation method.Compared with state-out-the-art baselines,the method that this paper proposed can outperform other methods.In addition,the method that proposed obtains meaningful personalized users’ responses and businesses’ associations on expenditure grades,which are useful to exploit the characteristics of users and businesses.2)The research of recommend potential repeat buyers for merchants based on user consumption behavior.Many merchants use the discount method to attract new buyers on e-commerce sites.Merchants hope the new buyers can be converted into repeat buyers,but most new buyers are one-time hunters This paper proposed a multi-classification model recommendation algorithm.In this algorithm,this work design a two-level cascade integration framework for multi-classification algorithm model.In the first level,this work produces a large number of different classification results by using a variety of different types of classifiers and combinations of features.In order to get better recommendation performance,in the second level,this work uses logistic regression to merge different classification results produced by the first level.Based on the experiment on the real dataset from TMall.com,this work verify the effectiveness of this algorithm.In this paper,in order to deploy the algorithm to the cloud platform,the algorithm also made the corresponding design in the feature extraction stage.