Recherche de sentiments opposés par une approche floue à partir de textes libres A fuzzy approach for determining opposite sentiments from free text

Most research in opinion mining has been focused on the classification of opinion orientations expressed in text content. Although this classification is useful and important, the sentiments opposite to the opinion expressed in text content are more and more interesting for the decision makers. On the other hand, such opposite sentiments often contain imprecise information and cannot be described by crisp methods. In this paper we introduce the problem of determining opposite sentiments in free format text content with fuzzy measures. We propose two fuzzy methods, called Local-Fuzzy andGlobal-Fuzzy, for determining the semantic strength of opposite sentiments within different fuzzy membership partitions. The experimental results on movie review data show the usefulness and accuracy of our approach.