Accuracy-diversity trade-off in recommender systems via graph convolutions

Highlights：

• We propose a novel accuracy-diversity trade-off for recommender systems with graph convolutional filters and neural networks.

• The approach leverages joint information from the nearest and furthest neighbours to balance accuracy with diversity.

• We develop a parameter design framework applicable both in a learning-to-rate and learning to-rank setting.

• We provide a graph special analysis of the proposed method to explain how ratings are exploited for the trade-off.

• Experimental evaluation demonstrates the proposed approach outperforms competing alternatives for establishing an accuracy-diversity trade-off.