Sentiment Classification at Discourse Segment Level: Experiments on multi-domain Arabic corpus

Abstract

Sentiment classification aims to determine whether the semantic orientation of a text is positive, negative or neutral. It can be tackled at several levels of granularity: expression or phrase level, sentence level, and document level. In the scope of this research, we are interested in the sentence and sub-sentential level classification which can provide very useful trends for information retrieval and extraction applications, Question Answering systems and summarization tasks. In the context of our work, we address the problem of Arabic sentiment classification at sub-sentential level by (i) building a high coverage sentiment lexicon with semi-automatic approach; (ii) creating a large multi-domain annotated sentiment corpus segmented into discourse segments in order to evaluate our sentiment approach; and (iii) applying a lexicon-based approach with an aggregation model taking into account advanced linguistic phenomena such as negation and intensification. The results that we obtained are considered good and close to state of the art results in English language.