Cross-Topic Opinion Mining for Real-Time Human-Computer Interaction

Abstract

With the recent growth and expansion of the Web 2.0, there has been an important development of new textual genres, such as blogs posts or forum entries, etc. that are employed to share opinions about a topic of interest. To the best of our knowledge, previous approaches focused on corpus annotation mostly concentrated on subjectivity versus objectivity classification and did not annotate emotions on a fine-grained scale. The scheme we propose in this arti-cle allows for both coarse and fine-grained annotation boundaries, as well as to distinguish among polarities and a large set of emotion classes. We used the an-notated elements to train our real-time opinion mining system, which we subse-quently employ for the classification of new sentences on a closely related topic - “recycling”. We obtain promising results in all the test scenarios, proving, on the one hand, that the corpus is a valid and useful resource, and, on the other hand, that our method used for opinion mining, is adequate.