A Temporal Collaborative Filtering Algorithm Based on Purchase Cycle

Existing Temporal Collaborative Filtering (TCF) recommendation algorithms exploit the time context to capture the user-interest drift. They have been used in the movie and music recommendation domain successfully. In online wholesale domain, e.g. Alibaba B2B online trading platform, most of the customers are wholesalers. Unlike individual customers, the wholesaler's demand dominates their purchase intentions rather than their interest. Hence, detecting the user-interest drift is not appropriate in the online wholesale domain. In order to capture the user-demand drift, we make use of customer's historical purchased records to predict the next purchase cycle (from the last purchase date to the next purchase date). We assume that the user's demand for the target product will reach the peak in the next purchase date, so the product should have a highest probability to be recommended in this date. Our proposed algorithm uses a deep neural network to predict the next purchase cycle, then incorporating next purchase cycle to the TCF recommender by a time-demand function. We evaluate our method on an online wholesale dataset. The experimental results demonstrate that our approach significantly improves the recommendation accuracy and ensures an acceptable novelty effect on the recommendation result.