SUNNYNLP

This repository contains the code for our paper: SUNNYNLP at SemEval-2018 Task 10: A Support-Vector-Machine-Based Method for Detecting Semantic Difference using Taxonomy and Word Embedding Features

Task description: SemEval 2018 Task 10 -- Capturing Discriminative Attributes

Our Support-Vector-Machine(SVM)-based system combines features extracted from pre-trained embeddings and statistical information from Probase to detect semantic difference of concepts pairs.

Requirements

Python and packages

We recommend using a separate Python 3.6 environment to install packages. All packages required are listed in requirements.txt. You can install them using pip:

pip install -r requirements.txt

As our system is using the English model in spaCy, run

python -m spacy download en

to install the language model required

Data

• Pre-trained word vectors:

  • Word2Vec
  • GloVe
  • FastText

• Our spaCy parsed version of Probase or the original Probase

Usage

• Edit the path for Probase, pre-trained vectors, output path, etc. according to the instructions in ./config/configuration-sample.yml.

• Run the main program in root directory to generate predictions based on your configuration:

$ python src/main.py config/configuration-sample.yml

• Run the official script to evaluate the predictions in the directory you have specified and save scores in ./score/

$ ./official-evaluation.sh ./prediction/configuration-sample
./prediction/configuration-sample/dev-5folds-FastText-dtc.txt
3 ./score/all-score.txt
./prediction/configuration-sample/dev-5folds-FastText-LinearSVC.txt
5 ./score/all-score.txt

References

If you find our work useful, please cite our work.

@inproceedings{lai2018sunnynlp,
  title={SUNNYNLP at SemEval-2018 Task 10: A Support-Vector-Machine-Based Method for Detecting Semantic Difference using Taxonomy and Word Embedding Features},
  author={Lai, Sunny and Leung, Kwong Sak and Leung, Yee},
  booktitle={Proceedings of The 12th International Workshop on Semantic Evaluation},
  pages={741--746},
  year={2018}
}

If you use the code, please cite according to the hyperwords repository

If you have used our spaCy parsed version of Probase, please cite according to the Probase official website