Trust Recommendation Based on Deep Deterministic Strategy Gradient Algorithm

Trust recommendation is a vital system application based on social networks. It can recommend items the trust between users, which alleviate data sparseness and enhance interpretability of results. A large number algorithms have been proposed, but most them believe that fixed, ignoring changes in process interaction. In addition, deep learning models are good at solving complex tasks processing high-dimensional data, they model algorithms, insufficient capturing user preferences timely. Therefore, given shortcomings existing researches, we propose DDPG-TR algorithm reinforcement to capture update users. The uses deterministic policy gradient DDPG user-item interaction process. Firstly, improve state representative structure express user’s state, convenient preferences. Then, when accepts recommendation, combines similar information predict item score, as well calculates difference score. Finally, Agent gets score feedback trust. Experiments conducted three datasets, verified provide more accurate results, compared with other algorithms.