基于长短期兴趣的深度强化学习推荐模型

A Deep Reinforcement Learning Recommendation Model Based on Long and Short Term Interest

现有的基于深度学习的推荐模型将推荐过程视为静态过程,在一段时间内使用固定策略进行推荐,难以动态捕捉用户兴趣变化,影响推荐结果的准确性。本文提出了一个利用深度强化学习动态地对推荐过程进行建模的推荐模型,模型以最大化长远收益为目标,通过分别提取长短期序列中的特征信息对用户兴趣进行描述,根据兴趣变化不断改变推荐策略。在Movielens-1m数据集上的实验结果表明,相较于其他基线模型,本文模型可在precision@10和recall@10上分别提升1.7%~7.6%和1.5%~3.8%。

The existing deep learning based recommendation models treat the recommendation process as a static process and use fixed strategies for recommendation over a period of time, which makes it difficult to dynamically capture changes in user interests and affects the accuracy of recommendation results. This article proposes a recommendation model that utilizes deep reinforcement learning to dynamically model the recommendation process. The model aims to maximize long-term benefits and describes user interests by extracting feature information from long and short term sequences. The recommendation strategy is constantly changed according to changes in interest. The experimental results on the Movielens-1m dataset indicate that compared to other baseline models, our model can be applied in precision@10 and recall@10 Increased by 1.7%~7.6% and 1.5%~3.8% respectively.