Building Domain Ontologies for Tunisian Dialect: Towards Aspect Sentiment Analysis from Social Media

Sentiment analysis from social media has received increasing attention during the last few years. Indeed, getting the opinion of customers, users, or people about a product or a service is paramount for many firms in different sectors, ranging from food to politics. Recently, Aspect Sentiment Analysis (ASA) constitutes a rapidly evolving research area. The idea is to determine for each comment, not only an overall polarity (i.e., positive, negative, or neutral) but a detailed polarity per aspect. This paper proposes an original work for building domain ontologies towards aspect sentiment analysis on Tunisian social media. These ontologies will be exploited to improve classification performance for both aspect detection and aspect sentiment analysis. To our best of knowledge, there is no Tunisian ontology extracted from social media content dedicated to sentiment analysis or any other close research domain. Our proposed method is based on an NLP pipeline that we apply on raw Tunisian dialect datasets scraped from social media. We focused on four domains which are Mobile phones, Food, Tunisian election and Radio/TV Programs and applied our proposed method to build Tunisian dialect domain ontologies.