User Occupation Aware Conditional Restricted Boltzmann Machine Based Recommendation
Weizhu Xie, Yuanxin Ouyang, Jingshuai Ouyang, Wenge Rong, Zhang Xiong · 2016
Nowadays, the explosive growth and variety of information available on the Web frequently overwhelms users and leads users to make poor decisions. Consequently, recommender systems have become more and more important to assist people to make decisions faster. Among all related techniques, collaborative filtering approach is currently one of the effective and widely used techniques to build recommender systems. However, there are major challenges like data sparsity and scalability. Meanwhile it is hard to integrate demographic statistical information (Age, gender and occupation etc.) to collaborative filtering model. Unfortunately, it is significant to take account into these information, especially user occupation when making recommendation. As we all know, people with different occupations may have totally different tastes. It has been proved that restricted Boltzmann machines(RBM) model can infer lower-dimensional representations automatically and is potential in handling large and sparse dataset. In this paper, we propose an improved User Occupation aware Conditional Restricted Boltzmann Machine Frame(UO-CRBMF) model, which employs an improved RBM and takes full use of user occupation information by adding a conditional layer with user occupation information. Experimental studies on the standard benchmark datasets of MovieLens 100k and MovieLens 1M have shown its potential and advantages beyond baseline methods.