File size: 3,028 Bytes
f89caa2 a5263ee |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 |
---
name: WordNetNoun (Disambiguated)
description: >
Disambiguated version of WordNet's noun hierarchy where entity names are formatted
as "name: definition" to resolve polysemy issues. This prevents training signal
conflicts when the same word has multiple meanings.
license: apache-2.0
language:
- en
multilinguality:
- monolingual
size_categories:
- 1M<n<10M
task_categories:
- feature-extraction
- sentence-similarity
pretty_name: WordNetNoun (Disambiguated)
tags:
- hierarchy-transformers
- disambiguation
- wordnet
---
# WordNetNoun (Disambiguated Version)
This is a **disambiguated version** of the WordNet Noun hierarchy dataset, where entity names are formatted as `name: definition` to resolve polysemy issues.
## Disambiguation
**Original format (ambiguous):**
```
child: "bank"
parent: "slope"
```
**New format (disambiguated):**
```
child: "bank: sloping land (especially the slope beside a body of water)"
parent: "slope: an elevated geological formation"
```
## Problem Solved
In the original dataset, the word "bank" appears with multiple meanings:
- bank.n.01: "sloping land" → parent: slope
- bank.n.09: "a building in which banking transacted" → parent: depository
- bank.n.10: "flight maneuver" → parent: flight maneuver
- ... (8 different senses total)
This caused **training signal conflicts** where the same text "bank" needed to be embedded close to multiple different parents simultaneously.
## Dataset Structure
Following the same structure as [Hierarchy-Transformers/WordNetNoun](https://huggingface.co/datasets/Hierarchy-Transformers/WordNetNoun):
- `MixedHop-RandomNegatives-Pairs/`: (child, parent, label) format for evaluation
- `MixedHop-RandomNegatives-Triplets/`: (child, parent, negative) format for training
Each contains train/val/test splits in parquet format.
## Statistics
- **Train**: 750,915 pairs / 682,650 triplets
- **Val**: 364,925 pairs / 331,750 triplets
- **Test**: 364,936 pairs / 331,760 triplets
- **Total entities**: 74,401 (with definitions)
## Source
- **Original data**: [Zenodo 10511042](https://doi.org/10.5281/zenodo.10511042)
- **Modification**: Added WordNet definitions to entity names for disambiguation
- **Code**: Modified `hierarchy_transformers.datasets.load` module
## Usage
```python
from datasets import load_dataset
# Load disambiguated dataset
ds = load_dataset("Jinrui/WordNetNoun", "MixedHop-RandomNegatives-Pairs")
# Example
print(ds['train'][0])
# {
# 'child': 'boarhound: large hound used in hunting wild boars',
# 'parent': 'hound: any of several breeds of dog used for hunting...',
# 'label': 1
# }
```
## Citation
If you use this dataset, please cite the original HierarchyTransformers paper:
```bibtex
@inproceedings{he2024language,
title={Language Models as Hierarchy Encoders},
author={He, Yuan and Yuan, Zhangdie and Chen, Jiaoyan and Horrocks, Ian},
booktitle={Advances in Neural Information Processing Systems},
year={2024}
}
```
## License
Same as the original dataset (Apache 2.0).
|