A Dual Perspective Framework of Knowledge-correlation for Cross-domain Recommendation

What is it about?

This study investigates cross-domain recommendation systems (CDRS), focusing on the useful mining of inter-domain correlation. A series of qualitative and quantitative experiments verify that our design can not only accurately predict users’ expressed preferences, but also explore their potential diverse interests.

Featured Image

Photo by Carlos Muza on Unsplash

Photo by Carlos Muza on Unsplash

Why is it important?

Since different domains usually have inconsistent data structures and heterogeneous knowledge content, it is difficult for CDRS to generalize the source data directly to the target domain. The current literature concentrates more on cross-domain relationships through common knowledge (e.g., tags, association rules), but they have some limitations due to human efforts and prerequisites. Our paper proposes a knowledge-correlated cross-domain recommendation method, which employs an encyclopedic Knowledge Graph (KG) as common knowledge, and innovatively examines the contribution of KG from dual perspectives of content semantic and structural connectivity.