User Dynamic Preference Construction Method Based on Behavior Sequence

People’s needs are constantly changing, and the performance of traditional recommendation algorithms is no longer enough to meet demand. Considering that users’ preferences change with time, behavior sequence hides evolution law preferences, so mining dependence extremely important predict dynamic preferences. From perspective constructing this paper proposes a preference model based on sequences. Firstly, user’s interest divided into short-term long-term models. The reflects current preference, refers from all his historical behaviors, representing consistent stable preference. Users’ obtained by integrating interest, which solves problem cannot reflect in real-time. We use public Amazon review dataset test we propose paper. Our achieves best performance, maximum improvement 15.21% compared basic (BPR, NCF) 2.04% (GRU4REC, Caser, etc.), proves can effectively helpful predicting real-time especially field recommendation.