Identification of opinions in Arabic newspapers

Abstract

Identification of opinions is a set of techniques which is a part of the natural language processing, especially in the information research area. This consists in developing systems able to extract and explore the opinions existing in corpuses. The presence of important textual mass of Arabic newspapers in an electronic format requires a particular exploration technique. We intend to present in this paper a system of opinions identification, based on the model of Aila Rosà [1], representing the opinion as an object composed of four elements : predicate, source, topic and content. Two properties: polarity and intensity which are inspired from the work of Plantié Mathieu [2] and are added to this model to establish relationships between the different opinions present in the text according to their different degrees of intensity and polarity. In presenting its general architecture, our system uses several techniques such as: XML representation of opinions, semantic expansion of opinions as explained by Nicolas B [3] and finally a statistical representation of the opinions in occurrences matrix format to facilitate the calculation of the similarity between the opinions in the classification phase.