Trust Recommended Algorithm Based on DDPG and Attention Mechanism

In view of the insufficient researches on the dynamic change of user trust in existing trust-based recommendation algorithms, a DDPG-TRA algorithm based on deep reinforcement learning and attention mechanism is proposed, which can catch the user interests and dynamically update user trust. The algorithm firstly uses the state module combined with the attention mechanism to model the user's state, and then recommends item. When the user adopts the item, it integrates social trust relationships and similarity relationships to predict item score, and computes the error between the predicted score and the actual score. Finally, the algorithm gets the recommendation reward and updates the trust with the score difference. We conduct experiments on three datasets and compared with other algorithms, and the proposed algorithm can get more accurate results.