Reproduce “A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dai Quoc Nguyen, Tu Dinh Nguyen, Dat Quoc Nguyen, Dinh Phung
ConvKB, an embedding model for knowledge base completion based on convolutional neural networks, captures global relationships across entity and relation embeddings within a triple. It achieved state-of-the-art results on WN18RR and FB15k-237, including a Mean Reciprocal Rank of 0.396 and Hits@10 of 51.7% on FB15k-237.
Key insights
- The `1 x 3` filter design, convolved row-wise over the `k x 3` input matrix, allows ConvKB to explicitly model 'global relationships among same dimensional entries' across the entire triple.
- ConvKB's architecture inherently generalizes the 'transitional characteristic' (v_h + v_r ≈ v_t), demonstrating that it can reduce to TransE under specific filter and weight conditions.
- By designing the CNN to perceive relationships across the entire triple's embeddings, ConvKB effectively combines the feature learning power of CNNs with the valuable intuitions from transition-based models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.