embedding-based methods can be divided into two classes
第一种:知识图谱由从数据集或外部知识库中提取的项目及其相关属性构成————项目图 item graph KGs are constructed with items and their related attributes, which are extracted from the dataset or external knowledge bases——————We name such a graph as the item graph.
用户不包含在此类项目图中。遵循该策略的论文利用知识图谱嵌入 (KGE) 算法对图进行编码以获得更全面的项目表示,然后将项目侧信息集成到推荐框架中。 Note that users are not included in such an item graph. Papers following this strategy leverage the knowledge graph embedding (KGE) algorithms to encode the graph for a more comprehensive representation of items, and then integrate the item side information into the recommendation framework.
每个Vj项目的最终表示(The latent vector vj of each item vj is obtained by aggregating information from multiple sources)
以上一系列说明了user's preference for candidate news vj can be calculated with Equation
另一种:基于嵌入的方法直接构建用户-项目图,其中用户、项目及其相关属性充当节点。 The other type of embedding-based method directly builds a user-item graph, where users, items, and their related attributes function as nodes. I、
the relation embedding r into a scalar——将 关系r 嵌入标量
2、BEM 用两种类型图来表示物品 the knowledge-related graph (containing item attributes information, like brand, category, etc.) and behavior graph (containing item interaction-related information, including co-buy, co-rate, co-add to cart) for recommendation
构建具有多种项目侧信息的KG,以丰富项目的表示,这些信息可以用于更精确地对用户表示进行建模。 build KGs with multiple types of item side information to enrich the representation of items, and such information can be used to model the user representation more precisely.
通过将用户引入图中来构建用户-项目图,可以直接对用户偏好进行建模。实体嵌入是基于嵌入的方法的核心 build user-item graphs by introducing users into the graph, which can directly model the user preference. Entity embedding is the core of embedding-based methods
一些论文使用 GAN [69] 或 BEM [74] 改进嵌入以获得更好的推荐。 some papers refine the embedding with GAN [69] or BEM [74] for better recommendation.
应用多任务学习的策略,将推荐模块与图相关任务联合训练,以提高推荐质量。 apply the strategy of multi-task learning to jointly train the recommendation module along with the graph-related task to improve the quality of recommendation.
Path-based Methods(基于路径的方法)
基于路径的方法构建用户项图,并利用图中实体的连接模式进行推荐.从2013年开始就开始使用了,传统papers称这种方法是HIN图中的推荐 。 Path-based methods build a user-item graph and leverage the connectivity patterns of the entity in the graph for recommendation. Path-based methods have been developed since 2013, and traditional papers call this type of method as recommendation in the HIN.
核心思想:利用users或items连接的相似性去提升推荐。 In general, these models take advantage of the connectivity similarity of users and/or items to enhance the recommendation.
定义了PathSim来衡量路径的相似性 To measure the connectivity similarity between entities in the graph,PathSim [107] is commonly used.
there is no comprehensive work to suggest under which circumstances, including data sources, recommendation scenarios, and model architectures, should a specific KGE method be adopted.
User Side Information 用户辅助信息
目前大多数的基于KG的推荐系统都是通过整合项目方的信息,而非考虑用户方的信息
综述三大问
论文介绍的研究究竟是什么
从两个方面总结了基于知识图谱的推荐系统
how the papers utilize the knowledge graph for accurate and explainable recommendation
introduce datasets used in these works
把知识图谱的使用归成了三类
embedding-based method
path-based method
unifified method
目前做得如何
A recent research trend is to unify the embedding-based method and the path-based method to fully exploit information from both sides. Moreover, unified methods also have the ability to explain the recommendation process.