Review-based hierarchical attention cooperative neural networks for recommendation
Review-based hierarchical attention cooperative neural networks for recommendation
复制标题
DOI：
10.1016/j.neucom.2021.03.098
复制
发表时间：
2021-04
期刊：
影响因子：
6
通讯作者：
Yongping Du;Lulin Wang;Zhi Peng;Wenyang Guo
中科院分区：
文献类型：
--
作者：
Yongping Du;Lulin Wang;Zhi Peng;Wenyang Guo
在电子商务平台中，用户进行购买行为并为购买的物品撰写评论。这些评论通常包含了大量有价值的推荐信息，这些信息可以反映用户的购买偏好和商品的特性。本文提出了层次注意力协作神经网络（HACN）推荐模型。采用层次注意机制，从评论文本中丰富用户和项目的特征表示。使用两个基于评论文本的并行网络分别对用户和项目进行建模，使得生成的特征更有针对性。此外，引入目标ID嵌入来捕获数据集中的全局实体关系。实验在来自Amazon的五个不同领域的真实世界数据集上进行，我们提出的HACN模型取得了比现有的最先进的方法更好的结果。
In e-commerce platform, users conduct purchase behavior and write reviews for the purchased items. These reviews usually contain a lot of valuable information for recommendation, which can reflect the purchase preference of the user and the characteristic of the item. We propose the Hierarchical Attention Cooperative Neural Networks (HACN) model for recommendation. Hierarchical attention mechanism is adopted to enrich user’s and item’s feature representation from review texts. Two parallel networks based on review texts are used to model users and items respectively, which makes the generated features more purposeful. Further, the target ID embedding is introduced to capture the global entity relationship in the dataset. The experiments are performed on five real-world datasets of different domains from Amazon, and our proposed HACN model has achieved better results than the existing state-of-the-art methods.