Kgat

Abstract

To provide more accurate, diverse, and explainable recommendation, it is compulsory to go beyond modeling user-item interactions and take side information into account. Traditional methods like factorization machine (FM) cast it as a supervised learning problem, which assumes each interaction as an independent instance with side information encoded. Due to the overlook of the relations among instances or items (e.g., the director of a movie is also an actor of another movie), these methods are insufficient to distill the collaborative signal from the collective behaviors of users.In this work, we investigate the utility of knowledge graph (KG), which breaks down the independent interaction assumption by linking items with their attributes. We argue that in such a hybrid structure of KG and user-item graph, high-order relations -which connect two items with one or multiple linked attributes -are an essential factor for successful recommendation. We propose a new method named Knowledge Graph Attention Network (KGAT) which explicitly models the high-order connectivities in KG in an end-to-end fashion. It recursively propagates the embeddings from a node's neighbors (which can be users, items, or attributes) to refine the node's embedding, and employs an attention mechanism to discriminate the importance of the neighbors. Our KGAT is conceptually advantageous to existing KG-based recommendation methods, which either exploit highorder relations by extracting paths or implicitly modeling them with regularization. Empirical results on three public benchmarks show that KGAT significantly outperforms state-of-the-art methods like Neural FM [11] and RippleNet [29]. Further studies verify the efficacy of embedding propagation for high-order relation modeling and the interpretability benefits brought by the attention mechanism. We release the codes and datasets at https://github. com

Content‐based and knowledge graph‐based paper recommendation: Exploring user preferences with the knowledge graphs for scientific paper recommendation

“…This paper uses the same inputs as BPRMF. CML: 35 A metric learning algorithm that simultaneously encodes user preferences and user‐user, item‐item similarity. This paper uses the same input as BPRMF. KGAT: 20 It explicitly models the high‐order associations between users and items on the knowledge graph, using an aggregation method of attention. DKN: 9 A content‐based deep learning recommendation framework that integrates multichannel CNNs to represent the semantic and knowledge layers of a paper. In this paper, the features of content C are taken as the features of the semantic layer, and the features of the knowledge graph G are taken as the features of the knowledge layer.…”

“…However, GCN generally treats the knowledge graph as an undirected graph and ignores the distinction between relation types. Therefore, this article considers a user's implicit preference distribution for all relations in advance 18,20 . The following is the general form of computing a node v embedding in a single GCN layer: where ,represents the neighborhood aggregation function used to aggregate information from the neighborhoods.…”

“…For data sparsity and cold start problems, Wang proposed KGCN 18 and KGCN‐LS 31 to automatically mine high‐order structural information between users and items on KGs with GCNs to obtain implicit user preferences. To make full use of the relations between items or instances, KGAT 20 builds a collaborative knowledge graph based on the user‐item bipartite graph and the item knowledge graph for high‐order associations. MKR 32 design a general cross‐compression unit as an information interaction channel for KG representation learning task and recommendation task.…”

“…With an external knowledge database, CGPRec extracts concepts from paper texts to construct a concept‐level KG with user nodes, paper nodes, and other metadata nodes. Finally, CGPRec uses an improved graph convolutional network (GCN) to learn high‐order associations between users and papers on the KG because it can effectively propagate high‐order node embeddings 18‐21 . The main contributions of this article are as follows: This article proposes a new paper recommendation model, CGPRec, based on both knowledge graph features and text features.…”

A Comprehensive Survey of Knowledge Graph-Based Recommender Systems: Technologies, Development, and Contributions

“…Graph-based models have shown potential as the technologies for next-generation recommender systems. Recent efforts in graph representation learning such as KGAT [100], KGCN [101], and KGNN-LS [102] use GNN to synthesize information from such connectivity, strengthening the representation capability and enriching the relationships between a user and an item Wang2020c. These advances show the importance of exploring the potential of neural networks for KG-based recommender systems.…”