A Plug-in Critiquing Approach for Knowledge Graph Recommendation Systems via Representative Sampling

What is it about?

Recommendation systems often suggest movies, music, or products based on what users like. But sometimes, users feel the suggestions are “not right” and give feedback, such as saying “I don’t like this type of film.” Traditional systems struggle to handle this feedback—they require complex model adjustments and may even forget the user’s past preferences. Our research introduces a new approach called the RISC framework. The key idea is that when users critique recommendations, the system samples “representative items” from a knowledge graph to better understand what the user really wants. This allows the system to quickly and accurately update recommendations without retraining the entire model. We also designed a method called WER (Weighted Experience Replay), which helps the system remember long-term preferences and avoid forgetting past likes. In simple terms, our work makes recommendation systems smarter at listening to user feedback and better at balancing new adjustments with old preferences. As a result, users receive recommendations that are more personalized and satisfying.

Featured Image

Photo by Testeur de CBD on Unsplash

Photo by Testeur de CBD on Unsplash