基于异构信息网络多关系嵌入的层次化意图和偏好进行推荐  

Hierarchical Intention and Preference with Heterogeneous Information Network Multi-Relational Embedding for Recommendation

作　　者：陈丽华 CHEN Lihua(College of Computer Science,Sichuan University,Chengdu 610065)

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《现代计算机》2021年第14期21-27,共7页Modern Computer

摘　　要：现有推荐方法通常只利用用户-商品交互二元关系并假设用户对商品只有扁平化的偏好,却忽略了用户层次化的意图和偏好和用户交互的自然过程,即用户首先具有购买某一类型商品的意图,然后在这种意图驱动下,再基于用户偏好选择特定商品。在推荐系统中存在多种不同类型的数据我们称之异构信息,高效地整合这些异构信息有助于提升推荐性能。所以本文提出基于异构信息网络多关系嵌入的层次化意图和偏好进行推荐,本文构建异构信息网络的多关系语义空间学习节点基于特定关系的嵌入。然后利用树来建模层次化意图和偏好,充分利用了用户的结构化决策模式学习用户偏好并进行推荐。我们通过在真实的两个数据集上进行大量的实验来验证所提出的算法的有效性。The existing recommendation methods typically use user-item binary relationship and assume that each user has only flat preference distribution over items,which ignores the hierarchical characteristic of users’preference and the natural interaction process,that is,the user first has an intention to purchase one type of items,followed by selecting a specific item based on the user's preference driven by this intention.There are many different types of data called heterogeneous information in recommender systems.Efficient integration of these heterogeneous information helps to improve the recommendation performance.Therefore,we propose a Hierarchical Intention and Preference with Heterogeneous Information Network Multi-Relational Embedding for Recommendation(HIP-HINMRE).We embed nodes of heterogeneous information network in distinct relational semantic space.Then,we hierarchical model users’intention and preference by a hierarchical tree,which users structured decision pattern is fully utilized to learn users preference and make recommendations.The experimental results on two real-world data demonstrate the advantages of our approach over several widely adopted state-of-the-art methods.

关 键 词：异构信息网络 推荐系统 层次树 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]