| Abstract: |
Recommender systems are an answer to information overload on the web. They filter and present to customers, small subsets of items that they are most likely to be interested in. Users’ interests may change over time, and accurately capturing this dynamics in such systems is important. Sugiyama, Hatano and Yoshikawa proposed to take into account the user’s browsing history. Ding and Li were among the first to address this problem, by assigning weights that decrease with the age of the data. Others authors such as Billsus and Pazzani, Li, Yang, Wang and Kitsuregawa proposed to capture long- and short- terms preferences and combine them for personalized search or news access. The Session-based Temporal Graph (STG) is a general model proposed by Xiang et al. to provide temporal recommendations by combining long- and short-term preferences. Later, Yu, Shen and Yang have introduced Topic-STG, which takes into account topics information extracted from data. In this paper, we propose Time Weight Content-based STG that generalizes Topic STG. Experiments show that, using Time-Averaged Hit Ratio as measure, this time weight content-based extension of STG leads to performance increases of 4%, 6% and 9% for CiteUlike, Delicious and Last.fm datasets respectively, in comparison to STG. |