Preparing a Dataset for Extracting Decision Elements from a Meeting Transcript Corpus

This work describes the construction of a new dataset for the purpose of decision element extraction from a meeting corpus. Specifically, the corpus consists of annotated text spans of alternatives and criteria of decisions undertaken during spoken conversations. The annotations were conducted with the help of crowd sourcing and finally curated by a domain expert. Our experiments show that the curated dataset can lead to more consistent predictions in comparison to the crowdsourced one. The aim of the released dataset is to encourage further studies in automated information extraction for decision analysis, e.g. evaluating the effectiveness of supervised models for this task.

By: Tuan Tran, Francesca Bonin, Léa A. Deleris, Debasis Ganguly, Killian Levacher

Published in: RC25673 in 2018

rc25673.pdf

Questions about this service can be mailed to reports@us.ibm.com .