Process-Specific Information for Learning Electronic Negotiation Outcomes

Article type: Research Article

Authors: Shah, Mohak | Sokolova, Marina | Szpakowicz, Stan

Affiliations: Department of Computer Science, Laval University, Quebec, Canada. E-mail: Mohak.Shah.1@ulaval.ca | Department of Computer Science, University of Montreal, Montreal, Canada. E-mail: sokolovm@iro.umontreal.ca | School of Information Technology and Engineering, University of Ottawa, Ottawa, Canada. Institute of Computer Science, Polish Academy of Sciences, Warsaw, Poland. E-mail: szpak@site.uottawa.ca

Abstract: We introduce Process-Specific Feature Selection, an innovative procedure of feature selection for textual data. The procedure applies to data gathered in person-to-person communication. The procedure relies on the knowledge of the processes that govern such communication. It is general enough to represent data in a wide variety of domains. We present a case study of electronic negotiation, in which participants exchange text messages. We present the empirical results of classifying the outcomes of electronic negotiations based on such texts. The results achieved using process-specific feature selection are marginally better than those afforded by several traditional feature selection methods. We show that this tendency is consistent across several learning paradigms.

Keywords: textual data, feature selection, electronic negotiations, machine learning

Journal: Fundamenta Informaticae, vol. 74, no. 2-3, pp. 351-373, 2006