Research map / Label-efficient and robust learning / Recognition and long-tail
Modelling Semantic Categories Using Conceptual Neighborhood
Zied Bouraoui, Jose Camacho-Collados, Luis Espinosa-Anke, Steven Schockaert
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 |
|---|---|---|---|
| Incorporate predicted conceptual neighbors into category representations | SECOND-WEA₄ vs. Gauss · BabelNet category induction | F1: 22.3 -> 54.2 | Improved |
| Select predicted conceptual neighbors rather than assuming all siblings are neighbors | SECOND-WEA₄ vs. Multi · BabelNet category induction | F1: 44.2 -> 54.2 | Improved |
| Select siblings by predicted conceptual neighborhood rather than cosine similarity | SECOND-WEA₄ vs. Similarity₄ · BabelNet category induction | F1: 38.2 -> 54.2 | Improved |
| Use four rather than one predicted neighbor (ablation) | SECOND-WEA · BabelNet category induction | F1: 41.9 -> 54.2 | Improved |
| Use four rather than two predicted neighbors (ablation) | SECOND-WEA · BabelNet category induction | F1: 49.3 -> 54.2 | Improved |
| Use four rather than three predicted neighbors (ablation) | SECOND-WEA · BabelNet category induction | F1: 50.4 -> 54.2 | Improved |
| Use five rather than four predicted neighbors (ablation) | SECOND-WEA · BabelNet category induction | F1: 54.2 -> 51.1 | Hurt |
| Use BERT rather than averaged word embeddings to identify neighbors when selecting one neighbor (ablation) | SECOND · BabelNet category induction | F1: 41.9 -> 47.0 | Improved |
| Use averaged word embeddings rather than BERT representations for text classification (ablation) | Text classifier · Text classifier cross-validation | F1: 65.8 -> 69.0 | Improved |
| Use a text representation classifier rather than co-occurrence sentence counts | Text classifier · Text classifier cross-validation | F1: 46.6 -> 69.0 | Improved |
| Use four rather than one predicted neighbor (ablation) | SECOND-BERT · BabelNet category induction | F1: 47.0 -> 53.3 | Improved |
| Use five rather than four predicted neighbors (ablation) | SECOND-BERT · BabelNet category induction | F1: 53.3 -> 51.3 | Hurt |
Closest papers by meaning
- Imposing Category Trees Onto Word-Embeddings Using A Geometric Construction (ICLR 2019 · similarity 0.86)
- The Geometry of Categorical and Hierarchical Concepts in Large Language Models (ICLR 2025 · similarity 0.86)
- Learning Semantic Representations for Novel Words: Leveraging Both Form and Context (AAAI 2019 · similarity 0.85)
- A Semantic Space is Worth 256 Language Descriptions: Make Stronger Segmentation Models with Descriptive Properties (ECCV 2024 · similarity 0.85)
- By Tying Embeddings You Are Assuming the Distributional Hypothesis (ICML 2024 · similarity 0.83)
- From Causal to Concept-Based Representation Learning (NeurIPS 2024 · similarity 0.83)
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.
