ET4EL

Fine-Grained Entity Typing for Domain Independent Entity Linking
Yasumasa Onoe and Greg Durrett
AAAI 2020

Prerequisites

• The code is developed with python 3.7 and pytorch 1.0.0 or newer versions (we've tested our code on pytorch 1.4.0).

Training Entity Typing Models

• Download our training data (entity_typing_data) from here (12GB) and put under data.
  • entity_typing_data/train/et_conll_60k uses the type set data/onotology/conll_categories.txt.
• Check entity_typing/constant.py to make sure paths are correct.
• Run the training function in entity_typing/main.py. Please see example commands in entity_typing/scripts.

Evaluating Models on Entity Linking

• Put entity linking evaluation data in the appropriate folder.
• Run the evaluation function in entity_typing/main.py. Please see example commands in entity_typing/scripts.

Data

Training Data

• We train our entity typing model on data derived from March 2019 English Wikipedia dump. This data can be downloaded from here (12GB).

Entity Linking Data for Evaluation

CoNLL-YAGO

• This data is not publicly available. You can find more information here.

WikilinksNED Unseen-Mentions

• This data is created by splitting the WikilinksNED training set (Eshel et al. 2017) into train, development, and test sets by unique mentions (15.5k for train, 1k for dev, and 1k for test). There are no common mentions between the train, dev, and test sets. The dataset can be downloaded from here. Note that the training set is used for baselines only.

Questions

Contact us at yasumasa@cs.utexas.edu if you have any questions!

Acknowledgements

Code for entity typing model is based on Eunsol Choi's pytorch implementation.

GitHub: https://github.com/uwnlp/open_type
Paper : https://homes.cs.washington.edu/~eunsol/papers/acl_18.pdf