Utilizing the Bidirectional Effect of Evolutive Trust-Rating for Recommendation in E-Commerce

Existing recommendation systems are not efficient due to the insufficient rate of accuracy and coverage. It is difficult to recommend suitable items for users, especially for long-tail items and cold start users. Unlike previous trust-based recommendation algorithms, which usually focus on the impact of trust on scores and the calculation method of trust values used in trust propagation, we propose a personalized items recommendation algorithm based on the bidirectional effect of trust and rating considering their time evolution. The model takes the mutual influence of evolutive trust-rating over time into account. When recommending an item for a given user, we can update the ratings and trust values on the current time dynamically, and then perform random walk to rank Top-N items. Our experimental results demonstrate that the proposed model achieves a substantial increase in recommendation accuracy and coverage, particularly in cold start situation.