Name
Abstract | ||
|---|---|---|
Incorporating a knowledge graph (KG) into recommender systems has been widely studied by researchers. Existing methods mostly extract the paths connecting user-item pairs and model these paths or iteratively propagate the user preference over the KG. These methods can capture the semantics of entities and relations well and help to comprehend users’ interests. However, these methods ignore the implicit features between the KG’s external items and thus cannot fully capture the users’ preferences. To solve this problem, we propose a novel model named E nhanced K nowledge-aware P ath N etwork (EKPN) to exploit the KG and capture implicit features between items outside the KG for recommendation. EKPN consists of two neural network modules: one module captures explicit features between items in the knowledge graph by automatically generating paths from users to candidate items for recommendation; the other module explores implicit features between items outside the knowledge graph by utilizing users’ historical interactions. To better capture the implicit features between items outside the knowledge graph, we propose the activation gate mechanism. Finally, we use a fusion mechanism to combine the two modules to enhance each other and achieve higher performance. Extensive validation on two real-world datasets shows the superiority of EKPN over baselines. |
Year | DOI | Venue |
|---|---|---|
2022 | 10.1007/s10489-021-02758-9 | Applied Intelligence |
Keywords | DocType | Volume |
Recommendation, Knowledge graph, Implicit features, Collaborative filtering | Journal | 52 |
Issue | ISSN | Citations |
8 | 0924-669X | 0 |
PageRank | References | Authors |
0.34 | 2 | 4 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Yang Peng | 1 | 13 | 4.18 |
| Ai, Chengming | 2 | 0 | 0.34 |
| Yao, Yu | 3 | 0 | 0.34 |
| Li, Bing | 4 | 0 | 0.34 |