Classification of Negotiation Thought Units using Supervised Learning

In this paper we apply natural language processing and machine learning techniques to a corpus of transcripts from Imai and Gelfand [4] in order to develop classifiers that can be used to automate coding of negotiation thought units. The models we used include multinomial Naive Bayes, maximum entropy, and a first-order conditional random field. We demonstrate that such classifiers perform significantly better than a baseline that samples from the observed probability distribution over labels. However, our results suggest that the existing corpus suffers from sparseness and that training a classifier with accuracy high enough to eliminate the need for manual coding requires a much larger corpus.