Repository information

This repository contains data and code for the paper below:

Learning to Ask Good Questions: Ranking Clarification Questions using Neural Expected Value of Perfect Information
Sudha Rao ([email protected]) and Hal Daumé III ([email protected])
Proceedings of The 2018 Association of Computational Lingusitics (ACL 2018)

Downloading data

• Download the clarification questions dataset from google drive here: https://go.umd.edu/clarification_questions_dataset

• cp clarification_questions_dataset/data ranking_clarification_questions/data

• Download word embeddings trained on stackexchange datadump here: https://go.umd.edu/stackexchange_embeddings

• cp stackexchange_embeddings/embeddings ranking_clarification_questions/embeddings

The above dataset contains clarification questions for these three sites of stackexchange:

1. askubuntu.com
2. unix.stackexchange.com
3. superuser.com

Running model on data above

To run models on a combination of the three sites above, check ranking_clarification_questions/src/models/README

Generating data for other sites

To generate clarification questions for a different site of stackexchange, check ranking_clarification_questions/src/data_generation/README

Retrain stackexchange word embeddings

To retrain word embeddings on a newer version of stackexchange datadump, check ranking_clarification_questions/src/embedding_generation/README

Contact information

Please contact Sudha Rao ([email protected]) if you have any questions or any feedback.