Explainable and Accurate Top-N Recommendations in Heterogeneous Information Networks

Neural network-based recommendation algorithms have become the state-of-the-art in recommender systems and can achieve very high predictive accuracy. However, these models are usually considered as black boxes in terms of their interpretability due to the complex structure of their hidden layers. In this research work, we propose MP4Rec, a recommender system using heterogeneous information networks to provide both accurate and explainable recommendations. MP4Rec uses of user-user and item-item similarity matrices and applies a newly proposed pair-wise objective function to make top-N recommendations which are transparent and explainable. The similarity matrices are created from metapaths constructed with the PathSim algorithm, node embeddings with cosine similarity or their combinations. The proposed pair-wise objective function incorporates an additional soft constraint for pushing more explainable items into the top-N recommendations. We have performed several experiments that show the effectiveness of our model by outperforming the state-of-the-art and providing both accurate and explainable recommendations in three well-known datasets.