融合长短期记忆网络与广义矩阵分解的神经协同过滤模型

Neural Collaborative Filtering Model Integrating LSTM and Generalized Matrix Factorization

在实现推荐的过程中,用户对项目的浏览和关注的时间顺序是推荐算法中重要的数据信息,同一用户在不同时间对项目的喜好不同对推荐结果也有着一定的影响.本文在神经协同过滤模型的框架下,提出将长短期记忆网络和广义矩阵分解进行融合,同时捕捉用户的短期偏好和长期偏好.利用长短期记忆网络对时序数据的强拟合能力,学习用户的短期偏好信息,捕捉序列的长依赖关系,通过广义矩阵分解学习用户的长期偏好信息,从而优化推荐算法,提高推荐性能.通过MovieLens-1M数据集进行试验后,结果表明,本文提出的新模型在收敛速度和推荐性能方面都有提升.

In the process of implementing recommendations, the browsing order of users is important information for the recommendation algorithm. The same user’s different preferences for items at different times also affect the recommendation results. Under the framework of the neural collaborative filtering model, this study proposes to integrate long short-term memory networks with generalized matrix factorization and capture both the user’s long-term and shortterm preferences. The new model utilizes the strong fitting ability of long short-term memory networks to time series data to learn the user’s short-term preference and capture the long dependence relationship of the sequence. The user’s longterm preference is learned through generalized matrix factorization. The recommendation algorithm is thereby optimized,and the recommendation performance is improved. Experiments are carried out on the Movielens-1 M dataset and the results show that the new model has a higher convergence rate and better recommendation performance.