在基于会话的推荐中捕捉用户意向
首发时间:2018-12-21
摘要:近年来,基于会话的推荐方法被广泛应用于推荐任务,在该类任务中,循环神经网络 (RNNs)与传统方法相比有明显的改进。以前的工作只考虑用户在当前会话中的顺序行为,而不强调用户在当前会话中的主要意向,本文提出了一种注意力神经网络框架来解决这一问题。具体来说,本文探索了一个具有注意力机制的双向RNNs(BiRNNs)编码器,用以建模用户顺序行为,同时捕捉用户意向。然后,通过匹配嵌入向量获得各候选项的得分并根据得分给出推荐结果。在RecSys2015数据集的实验结果表明,本文提出的基于会话的注意力推荐模型相对于其他被广泛使用的方法有明显改进。
关键词: 基于会话的推荐 循环神经网络 注意力机制 用户顺序行为
For information in English, please click here
Capturing User\'s Intention in Session-based Recommendations
Abstract:In recent years, session-based recommendation methods have been widely used in recommendation tasks, compared with traditional methods, recurrent neural networks (RNNs) have shown significant improvements in session-based recommendation tasks. Previous work only considered the user\'s sequential behavior, but did not emphasize the user\'s intention in the current session. This paper proposes an attention neural network to solve this problem. Specifically, we explore a bidirectional RNNs (BiRNNs) encoder with attention mechanism to model user\'s sequential behavior while capturing user\'s intention. Then, we get the scores of each candidate item by matching the embedded vectors and get the recommended results according to the scores. We conducted experiments on the RecSys2015 dataset. The result shows that the session-based attention recommendation model proposed in this paper is significantly improved compared to other widely used methods.
Keywords: session-based recommendation recurrent neural networks attention mechanism user\'s sequential behavior
引用
No.****
同行评议
勘误表
在基于会话的推荐中捕捉用户意向
评论
全部评论0/1000