Sentiment Analysis and Prediction of the Engagement of Users of the Algerian Dialect on Social Networks

     Two problems related to the use of social networks have turned out to be very important. First, Sentiment Analysis focuses on the study and analysis of peoples’ opinions, sentiments and emotions based on written language. Since social media platforms are increasingly used by many companies as a major channel to advertise and sell their products, tools are clearly needed to analyse peoples’ opinions on and reviews of the various products, feedback on events, etc. Second, for the purposes of online marketing, the various social networks provide advertising platforms that allow the sponsoring of advertising content to reach targeted users. However, anticipating the effectiveness of a content is very important to optimise the return on investment. The performance of an advertising content is usually measured by a metric called the Engagement Rate often used in the field of social media marketing to measure the extent to which the users will show “interest” for and interact with the advertised content. Thus, being able to predict the engagement rate of a publication is of utmost importance to social marketers. This talk will start by stressing the golden opportunities that the current flux of data represents, especially from the perspective of Natural Language Processing. Then the challenges of processing the Arabic language are exposed, followed by the even more challenging processing of the Algerian Dialect. Next, solutions are presented to the problems of Sentiment Analysis and Engagement Prediction in the context of users of the Facebook social network in the Algerian dialect. These solutions are based on a painstaking pre- processing of the corpus of Algerian dialect posts and comments, and then the use of Deep Learning architectures which are presented and compared. In the case of Sentiment Analysis, two neural network models, MLP and CNN, are trained to classify comments as negative, neutral or positive opinions. On the other hand, for Engagement Prediction, two neural network models are proposed, one based on an MLP architecture and the other on a hybrid Convolutional-LSTM and MLP architecture.