Ontology Adaptation for Opinion Mining in French Corpora

Abstract

This paper presents the development of an ontology for opinion mining in French corpora. Since ontology construction from scratch is expensive and time consuming, the model is built by adapting an existing ontology modeling appraisal categories in English. The construction method consists of two main steps: first, the ontology of appraisals is translated in French by using a concept-to-concept approach; then different adaptation strategies are implemented to improve the result of this translation. Adaptation strategies are based on text mining, weakly supervised methods and the use of WordNet to refine the model. The goal of the adaptation phase is to cope with limitations of concept-to-concept translation, to integrate concepts and relations from external corpora and resources, and to build a model that captures various features of opinions. The ontology was designed and build to incorporate linguistic and extra linguistic information on the description of opinions in French. The model was validated by using both expert insights to validate the relevance of concepts and relations and formal criteria to describe the ontology qualities for practical use.