Research map / Graph learning / Knowledge and temporal graphs
Knowledge Graph Completion via Complex Tensor Factorization
Théo Trouillon, Christopher R. Dance, Éric Gaussier, Johannes Welbl, Sebastian Riedel, Guillaume Bouchard
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use complex-valued embeddings (ComplEx) instead of real DistMult | knowledge graph embedding · WN18 | filtered MRR: 0.822 -> 0.941 | Improved |
| Use complex-valued embeddings (ComplEx) instead of real DistMult | knowledge graph embedding · FB15K | filtered MRR: 0.654 -> 0.692 | Improved |
| Use ComplEx with log-likelihood loss instead of HolE with max-margin loss | knowledge graph embedding · FB15K | filtered MRR: 0.524 -> 0.692 | Improved |
| Use shared entity embeddings (ComplEx) instead of separate subject/object embeddings (CP) | knowledge graph embedding · WN18 | filtered MRR: 0.075 -> 0.941 | Improved |
| Use ComplEx instead of RESCAL on a large relation set | knowledge graph embedding · FB15K | filtered MRR: 0.461 -> 0.692 | Improved |
| Use ComplEx instead of RESCAL on a fully observed data set (Kinships, UMLS) | knowledge graph embedding · Kinships and UMLS | average precision | No clear effect |
| Increase negatives per positive from 1 to 10 (ablation) | ComplEx · FB15K | filtered MRR: up to +0.08 | Improved |
| Increase negatives per positive from 1 to 100 (ablation) | ComplEx · FB15K | filtered MRR: 0.737 filtered MRR and 64.8% Hits@1 at 100 negatives | Improved |
| Increase negatives per positive from 100 to 200 (ablation) | ComplEx · FB15K | filtered MRR: decreases at 200 negatives | Hurt |
| Increase negatives per positive beyond one (ablation) | ComplEx · WN18 | filtered MRR | No clear effect |
| Add L2 regularization (optimal lambda instead of 0) (ablation) | ComplEx · WN18 and FB15K | filtered MRR: up to +0.05 | Improved |
| Increase rank from 50 to 200 (ablation) | ComplEx · WN18 and FB15K | filtered MRR: little performance gain for ranks 50-200 | No clear effect |
Closest papers by meaning
- SimplE Embedding for Link Prediction in Knowledge Graphs (NeurIPS 2018 · similarity 0.91)
- Complex Embeddings for Simple Link Prediction (ICML 2016 · similarity 0.91)
- Improved Knowledge Graph Embedding Using Background Taxonomic Information (AAAI 2019 · similarity 0.90)
- SPAC: Sparse Partitioning and Adaptive Core Tensor Pruning Model for Knowledge Graph Completion (AAAI 2025 · similarity 0.89)
- ER: Equivariance Regularizer for Knowledge Graph Completion (AAAI 2022 · similarity 0.89)
- Knowledge Graph Completion by Intermediate Variables Regularization (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
