PREDICT: Persian Reverse Dictionary

A reverse dictionary takes a description of a concept as input, and produces the most appropriate word matching that description. Over the past decades, researchers have designed reverse dictionaries for many languages including English, Turkish, French, and Japanese. This repository contains the implementation details and source codes used for the development of the first intelligent Persian reverse dictionary (called "PREDICT").

Project Files

All of the files needed to run the jupyter notebooks are uploaded on Google Drive, and are accessible through this link.

Dataset

We have introduced a novel dataset which can be used to develop Persian reverse dictionaries in the future. The dataset is a collection of 855217 words along with the phrases describing them. These phrases were extracted from the top three most well-known Persian dictionaries (including Amid, Moeen, and Dehkhoda), Persian Wikipedia, and a Persian Wordnet (called Farsnet). Our new dataset can be downloaded through Kaggle.

Cite

Please cite the following paper in your publication if you are using PREDICT in your research:

@article{malekzadeh2021predict,
  title={PREDICT: Persian Reverse Dictionary},
  author={Malekzadeh, Arman and Gheibi, Amin and Mohades, Ali},
  journal={arXiv preprint arXiv:2105.00309},
  year={2021}
}

Acknowledgements

This work was supported by the Mathematics and Computer Science Department of Amirkabir University of Technology (Tehran Polytechnic).