Understanding dialogue: sentiment and topic analysis of dialogue transcripts

Abstract

The field of natural language processing has had success in analyzing sentiment and topics on written text, but similar analysis on dialogue is more difficult due to the fragmented and informal nature of speech. This work explores sentiment and topic analysis on data from the Switchboard dialogue corpus, as well as a dataset of recorded dialogues between parents and children while reading an interactive e-book. The goal was to be able to identify the emotion and mood of the dialogue in order to make inferences about what parents and children generally talk about when reading the book because conversations between an adult and child while reading a book can greatly contribute to the learning and development of young children.