Abstract
Knowledge graph (KG) is widely used as side information to improve recommendation performance. Among existing KG-based recommendation methods, propagation-based methods have the best performance. However, these methods are still problematic because they ignore the individual meaning of each item in the user's interaction history. So the representations of users and items learned by these methods are sub-optimal. In this paper, we proposed an end-to-end propagation-based recommendation model named Interactive Knowledge-aware Attention Network (IKAN). The proposed IKAN models each item in the user's interaction history separately, then utilize representations of these items to construct the corresponding user's representation. At the same time, our model uses knowledge-aware attention module and interaction-aware attention module to learn representations of users and items. We evaluate IKAN on three real-world datasets. Experimental results demonstrate that IKAN outperforms state-of-the-art KG-based recommendation methods.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have