Research map / Label-efficient and robust learning / Recognition and long-tail
Improving Entity Linking by Modeling Latent Entity Type Information
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add BERT-based entity similarity to the local model | DeepED on AIDA-B test · AIDA-B test | F1: 92.22 $\pm 0.14$ -> 93.54 $\pm 0.12$ | Helped |
| Add BERT-based entity similarity to the local model, without global modeling | DeepED local model on AIDA-B test · AIDA-B test | F1: 88.75 ± 0.30 -> 90.06 ± 0.22 | Helped |
| Build entity embeddings by averaging BERT representations of masked anchor contexts instead of using DeepED embeddings | Entity type prediction on FIGER, BBN, and OntoNotesFine · FIGER entity typing test | Micro F1: 80.38 -> 88.69 | Helped |
| Replace BERT entity embeddings with a learned projection into Ganea and Hofmann's embedding space (ablation) | BERT+G&H on AIDA-B test · AIDA-B test | F1: 92.22 $\pm 0.14$ -> 91.00 $\pm 0.72$ | Hurt |
| Replace Ganea et al.'s global model with DCA (ablation) | BERT-Entity-Sim on AIDA-B test · AIDA-B test | F1: 93.54 ± 0.12 -> 93.66 $\pm 0.17$ | Helped |
| Replace Ganea et al.'s global model with DCA (ablation) | BERT-Entity-Sim on five out-domain test sets · Five out-domain test sets (Avg) | F1: 86.02 -> 86.72 | Helped |
| Add Jaccard similarity using predicted Ultra-fine types | DeepED on AIDA-B test · AIDA-B test | F1: 92.22 $\pm 0.14$ -> 91.35 $\pm 0.18$ | Hurt |
| Add Jaccard similarity using predicted ZOE types | DeepED on AIDA-B test · AIDA-B test | F1: 92.22 $\pm 0.14$ -> 91.42 $\pm 0.09$ | Hurt |
| Add Jaccard similarity using oracle mention types | DeepED on AIDA-B test · AIDA-B test | F1: 93.07 $\pm 0.27$ -> 96.35 $\pm 0.14$ | Helped |
| Use a parameterized similarity formula instead of cosine similarity (ablation) | BERT-Entity-Sim local model · Entity linking | F1 | No clear effect |
| Use BERT-Entity-Sim instead of the attention-based bag-of-words local context representation | DeepED type-error cases on AIDA-CoNLL development · AIDA-CoNLL development | Type errors corrected: 124 out of 185 (67.03\%) | Helped |
Closest papers by meaning
- Fine-Grained Entity Typing for Domain Independent Entity Linking (AAAI 2020 · similarity 0.89)
- Neural Cross-Lingual Entity Linking (AAAI 2018 · similarity 0.87)
- LATTE: Latent Type Modeling for Biomedical Entity Linking (AAAI 2020 · similarity 0.86)
- A Bidirectional Multi-paragraph Reading Model for Zero-shot Entity Linking (AAAI 2021 · similarity 0.86)
- Fine-Grained Named Entity Typing over Distantly Supervised Data Based on Refined Representations (AAAI 2020 · similarity 0.86)
- A Dual-Way Enhanced Framework from Text Matching Point of View for Multimodal Entity Linking (AAAI 2024 · similarity 0.85)
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.