基于近邻用户评论的推荐辅助网络

评论数据存在稀疏问题,不足以支撑学习出更全面的用户偏好。针对评论稀疏问题进行了研究,并提出一种应对评论稀疏的“即插即用”辅助网络（NRSN）,其能与不同的模型进行结合,以添加辅助信息的方式,来重新调整当前模型输出的用户偏好向量。首先根据目标用户,使用aspect-attention机制从其近邻用户评论中学习出近邻用户的偏好,然后采用co-attention机制将近邻用户和目标用户进行契合度匹配,调整出目标用户新的偏好向量。在三组公开数据集下的实验结果表明,NRSN不仅能提高所结合模型的推荐性能,且能有效应对“冷启动”场景下的评论稀疏问题。

Recommendation supplemental network of neighbor reviews

Review data has sparse problems that are insufficient to support learning more comprehensive user preferences. Focusing on this issue, this paper proposed a "plug and play" auxiliary network(NRSN), which can be combined with different models to improve their performance. The network mainly re-adjusts the user preference vector of the current model output by adding auxiliary information. Firstly, according to the target user, it used the aspect-attention mechanism to learn the preference of the neighboring users from the reviews of their neighbors. Then, it used the co-attention mechanism to match the neighboring users and the target users, and adjusted the new preference vector of the target users. The experiment on three public datasets result show that NRSN can not only improve the performance of existing models, but also can effectively alleviate the impact of review sparseness in "cold start" scenario.