Abstract

Recommendation system has been used more and more frequently in many applications recent years. With the increasing information avail- able, not only in quantities but also in types, how to leverage these rich information to build a better recommendation system becomes a natu- ral problem. Most traditional approaches try to design a specic model for each scenario, which demands great eorts in developing and modi- fying models. In this technical report, we describe our implementation of feature-based matrix factorization. This model is an abstract of many variants of matrix factorization models, and new types of information can be utilized by simply dening new features, without modifying any line of code.

Introduction

Recommendation systems that recommends items according to users in- terest has become more and more popular among many websites. Collab- orative Filtering(CF) techniques that behind the recommendation system have been developed for many years and keep to be hot area in both academic and industry. Currently CF problems face two kinds of major challenges: how to handle large-scale dataset and how to leverage the rich information of data collected.

Traditional approach to solve these problems is to design specic model for each problem. This means write a code for each model, which demands great eorts in engineering. Matrix factorization(MF) technique is one of the most popular style of CF model, and extensive study had been made in dierent variants of matrix factorization model, such as [2][3] and [4]. However, we can nd many matrix factorization models share common patterns, which allows us to put them into one model. We call this model feature-based matrix factorization. We write a toolkit for solving the general feature-based matrix factorization problem, and save the eort for engineering for detailed kinds of model.