4.5 Article

Modeling Multiple Coexisting Category-Level Intentions for Next Item Recommendation

期刊

出版社

ASSOC COMPUTING MACHINERY
DOI: 10.1145/3441642

关键词

Recurrent neural networks; recommender system

向作者/读者索取更多资源

Purchase intentions have a significant impact on future purchases, but they are typically complex and subject to change. Empirical study shows that user behaviors of multiple types can indicate intentions, and users may have multiple coexisting category-level intentions that evolve over time. The proposed Intention-Aware Recommender System (TARS) effectively mines complex intentions from diverse user behaviors and outperforms state-of-the-art recommendation methods.
Purchase intentions have a great impact on future purchases and thus can be exploited for making recommendations. However, purchase intentions are typically complex and may change from time to time. Through empirical study with two e-commerce datasets, we observe that behaviors of multiple types can indicate user intentions and a user may have multiple coexisting category-level intentions that evolve over time. In this article, we propose a novel Intention-Aware Recommender System (TARS) which consists of four components for mining such complex intentions from user behaviors of multiple types. In the first component, we utilize several Recurrent Neural Networks (RNNs) and an attention layer to model diverse user intentions simultaneously and design two kinds of Multi-behavior GRU (MGRU) cells to deal with heterogeneous behaviors. To reveal user intentions, we carefully design three tasks that share representations from MGRUs. The next-item recommendation is the main task and leverages attention to select user intentions according to candidate items. The remaining two (item prediction and sequence comparison) are auxiliary tasks and can reveal user intentions. Extensive experiments on the two real-world datasets demonstrate the effectiveness of our models compared with several state-of-the-art recommendation methods in terms of hit ratio and NDCG.

作者

我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。

评论

主要评分

4.5
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

暂无数据
暂无数据