Hybrid Recommender System Based on Personal Behavior Mining

Recommender systems are mostly well known for their applications in e-commerce sites and are mostly static models. Classical personalized recommender algorithm includes item-based collaborative filtering method applied in Amazon, matrix factorization based collaborative filtering algorithm from Netflix, etc. In this article, we hope to combine traditional model with behaviour pattern extraction method. We use desensitized mobile transaction record provided by T-mall, Alibaba to build a hybrid dynamic recommender system. The sequential pattern mining aims to find frequent sequential pattern in sequence database and is applied in this hybrid model to predict customers' payment behaviour thus contributing to the accuracy of the model. INTRODUCTION Recommender systems are ubiquitous. There perform quite well in recommending products, movies, music, etc. Traditional methods of recommender systems include content based recommender system which make recommendations that is similar to what the user has purchased in the past. While the key part of the content-based recommender system is that this model requires a feature extraction process. Take NewsWeeder [19] an example, this system in 1995 solve a significant problem in information filtering systems that is the creation of a user profile that describes the user's interests. The NewsWeeder's way of creating such profiles have been applied in content-based recommender system, which recommend a product to a customer based on a description of the item and a profile of the user’s interests. But the shortcoming of this system is that in most cases only a very shallow analysis of certain kinds of content can be supplied [18], especially in some e-commerce sites because of the privacy-preserving policy. As for CF (collaborative filtering) recommendation, the system will not recommend items that is similar to what this user has purchased, but the items what other users have liked. Collaborative filtering requires no profile or content of both users or items, recommendations are make solely based on similarity matching to other users (the nearest neighbours’ searching). Traditional CF method include memory based and model based Collaborative Filtering algorithm. Matrix factorization based Collaborative Filtering algorithm are proposed by Yehuda Koren [8] in 2009, allowing the incorporation of additional information such as implicit feedback, temporal effect and etc. Most of those recommendation models are static, in the sense they use static features to describe products and users. But nowadays people are using mobile devices to browse products everywhere, generating a lot more behavioural data. More importantly, as the density of behavioural data growth, there may be a potential chance to make use of the behavioural pattern of users. We proposed a hybrid recommender system that combine the prefix span algorithm with traditional matrix factorization.