Opinion Mining in Social Networks for Algerian Dialect

There has been a significant increase in the volume of Arabic dialect messages on social networks, providing a rich source for opinion mining research. Most research works done on Arabic dialect focus on messages written in Arabic script, with very limited scope on Latin script. In this paper, we are interested in the classification of social networks messages retrieved from Twitter, Facebook and YouTube written in Algerian dialect in Latin script into positive or negative classes using existing opinion mining approaches (lexical-based, machine learning, and hybrid). Also, we apply a regrouping process in the preprocessing step to overcome the issues related to the Algerian dialect such as the orthographic varieties to express the same word. Furthermore, we focus on the hybrid approach which consists in automatically annotating the training corpus with the lexical-based approach and then use the machine learning approach on this corpus for creating the classification model. This approach allows classifying the messages into positive or negative classes, without having to annotate manually a training corpus.