Arabic Domain-Oriented Sentiment Lexicon Construction Using Latent Dirichlet Allocation

Abstract

Sentiment lexicon is crucial in the process of sentiment analysis. The efficient lexicon is the one that is able to provide the classifier with the right tokens of each class, positive and negative. In this paper, we have built a domain-oriented Arabic sentiment lexicon automatically using a generative statistical model called Latent Dirichlet Allocation (LDA). We tested our lexicon by doing documents classification and compare it with a classification done based on a manual lexicon created in previous study. We achieved good results from both accuracy and recall perspectives.