A Sentiment analysis classification for texts in social media:application to the Tunisian dialect

Currently, our life is based on information and its analysis. More and more, people communicate, differ in content and express their opinions on the Internet about a large number of subjects, in newsgroups, blogs, forums and other sites regarding product reviews. These data play an important role in decision-making for many people and organizations. For example, it is important for a government to know the views and concerns of these citizens and to reflect their moods and opinions about the news. Similarly, the example of companies wishing to know consumer reactions to their products. It might help them change strategy to improve the quality of their products and that of their services. For this reason, the expansion of digital communication through social media presented an opportunity for the research in data science to analyses the big data for artificial information extraction and it is useful to analyze the content of social networks that have gained great popularity around the world Sentiment analysis (Sentiment Analysis or Opinion Mining) is one of the most popular information extraction task for automatic processing of natural language (Natural Language Processing) which attempts to identify the presence of feelings or emotions expressed in a text, or in a sentence (positive, negative, or neutral) and it help organisations to understand the public’s opinion towards news, public , events, productsand , figures. However, sentiment analysis has advanced English has the most significant number studies, while research is more limited for other languages, including Arabic. While sentiment analysis for Arabic is developing in the literature, a popular variety of Arabic, Dialect , has been overlooked. Dialect is an the first form of Arabic spoked by people in their daily life and social texting but it still overlooked and this can be explained by two major factors. First, the lack of additional resources for this type of languages. Second, the complexity of treatment of this language, namely the variety of dialectal Arabic spoken in Tunisia. The lack of resources is seen as a serious problem which hinders, in a way decisive, the development of natural language processing tools for Arabic dialects in general and in particular for the Moroccan Dialect. It could be considered the main reason that motivated the creation of a lexicon Tunisian dialect in our work. Indeed, our study aims to develop a sentiment analysis system for data extracted from social network. This system will make it possible to collect, process and classify the polarity of comments written in Tunisian dialect according to the category (positive, negative, neutral). In our work, we conduct a rule-based approach by modifying a popular English tool VADER to support Tunisian sentiment polarity identification. We have compiled a Tunisian polarity lexicon from an English polarity lexicon of SO-CAL. Furthermore, we have modified the functionalities of English VADER, so that it can directly classify Tunisian text sentiments without the requirement of Arabic to English translation. This report consists of four chapters, In the first chapter, we focus on sentiment analysis, its Tasks and its approaches. The second chapter is devoted to the Arabic language and more specifically to the Tunisian dialect, emphasizing the works which have been done on Arabic sentiment analysis and Tunisian dialect, In the third chapter, we present in detail the general contribution of this memory . This chapter presents the overall design as well as the general architecture of the proposed approach. The fourth chapter presents the experiments and the discussion of the results obtained. Finally, we put a conclusion summarising the presented work and some perspectives representing our future works. p. 53-57