Abstract
To alleviate the data sparsity and cold start problems for collaborative filtering in recommendation systems, side information is usually leveraged by researchers to improve the recommendation performance. The utility of knowledge graph regards the side information as part of the graph structure and gives an explanation for recommendation results. In this paper, we propose an enhanced multi-task neighborhood interaction (MNI) model for recommendation on knowledge graphs. MNI explores not only the user-item interaction but also the neighbor-neighbor interactions, capturing a more sophisticated local structure. Besides, the entities and relations are also semantically embedded. And with the cross&compress unit, items in the recommendation system and entities in the knowledge graph can share latent features, and thus high-order interactions can be investigated. Through extensive experiments on real-world datasets, we demonstrate that MNI outperforms some of the state-of-the-art baselines both for CTR prediction and top-N recommendation.
Highlights
Nowadays, we are in the era of data explosion as the internet develops rapidly, which raises a lot of obstacles for users to find their interested information
The knowledge graph embedding module shown on the upper part of Fig 2, uses a multi-layer to extract the features of head and relation from the triples, which can preserve the semantic of relations and the graph structure information
On Book-Crossing, our framework increases the Area Under Curve (AUC) by 1.17% compared to knowledge-enhanced neighbor interaction (KNI) and 10.9% compared to MKR
Summary
We are in the era of data explosion as the internet develops rapidly, which raises a lot of obstacles for users to find their interested information. Inspired by graph attention network [20], KGAT [21] was proposed to model a high-order relation with attention mechanism in knowledge graphs These methods take into consideration graph structural information as to improve the recommendation precision. This method completes the user-item interaction with entities in the knowledge graph, which improves the recommendation precision It neglects some useful information by the scoring matrix. To address the limitations of the existing methods mentioned above, we propose an enhanced multi-task neighborhood interaction (MNI) model for recommendation on knowledge graph, which utilizes the multi-task learning and extends the user-item interaction to their neighbors. We adopt the idea of KNI to compress the knowledge graph into a local structure where both users’ and items’ neighbors are collected for prediction tasks This process avoids the early summarization problem.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
