Detecting Implicit References in Chats Using Semantics

Traian Rebedea, Gabriel-Marius Gutu

Abstract: Chat conversations with multiple participants are widely used in solving a wide range of CSCL tasks. One of the reasons for their success is that they encourage multiple conversation threads to exist in parallel, thus allowing multiple topics and ideas to be debated at the same time. These threads may be detected more easily if we would be able to identify the links that exist between the utterances of a conversation. This paper tries to explain whether semantic similarity measures from Natural Language Processing (NLP) may be successfully used to detect the links between utterances in CSCL chat conversations.