Knowledge-aware Graph Neural Networks with Label Smoothness Regularization for Recommender Systems
About
Knowledge graphs capture structured information and relations between a set of entities or items. As such knowledge graphs represent an attractive source of information that could help improve recommender systems. However, existing approaches in this domain rely on manual feature engineering and do not allow for an end-to-end training. Here we propose Knowledge-aware Graph Neural Networks with Label Smoothness regularization (KGNN-LS) to provide better recommendations. Conceptually, our approach computes user-specific item embeddings by first applying a trainable function that identifies important knowledge graph relationships for a given user. This way we transform the knowledge graph into a user-specific weighted graph and then apply a graph neural network to compute personalized item embeddings. To provide better inductive bias, we rely on label smoothness assumption, which posits that adjacent items in the knowledge graph are likely to have similar user relevance labels/scores. Label smoothness provides regularization over the edge weights and we prove that it is equivalent to a label propagation scheme on a graph. We also develop an efficient implementation that shows strong scalability with respect to the knowledge graph size. Experiments on four datasets show that our method outperforms state of the art baselines. KGNN-LS also achieves strong performance in cold-start scenarios where user-item interactions are sparse.
Key insights
What problem does KGNN-LS address?
Recommender systems using knowledge graphs face cold-start issues and rely on manual feature engineering. KGNN-LS transforms the KG into a user-specific weighted graph via a trainable relation scoring function, then applies a GNN for personalized item embeddings.
How does KGNN-LS differ from prior work?
Unlike prior KG-aware methods relying on manual feature engineering, KGNN-LS trains end-to-end. Unlike prior GNNs designed for homogeneous bipartite graphs, it handles heterogeneous KGs. Label smoothness regularization, proven equivalent to label propagation, regularizes learnable edge weights.
What are the experimental findings?
On four datasets (MovieLens-20M, Book-Crossing, Last.FM, Dianping-Food), KGNN-LS outperformed baselines including SVD, LibFM, PER, CKE, and RippleNet in Recall@K and AUC. AUC surpassed baselines by 5.1%, 6.9%, 8.3%, and 4.3% on average respectively.
What limitations and future work remain?
The model over-smoothes entity representations when using too many GNN layers. LS regularization was proposed only for KG-based recommendation; the authors note examining it on link prediction and node classification, and investigating the theory between feature and label propagation as future work.