Sentiment analysis has been an interesting research topic in recent years. Previous work has focused on generic problem settings where only textual description is available, which is essentially a natural language processing task. In terms of movie review sentiment analysis, however, linguistic approaches suffer from the fact that sentiment words in these movie reviews might indicate the characteristics of the movie rather than those of the reviewer’s opinions.

In this work, the authors attempt to fuse audio and video features with linguistic features to help classify the sentiment of the review. Acoustic features include low-level descriptors (LLD) and their derivatives, and video features mainly focus on the smile expression, scaled by intensity from 0 to 100.

The experimental results on different classification fusion settings show that audio and video features are useful in adding discriminating information, although it is not clear whether the proposed fusion approach depends on a specific choice of audio and video features. It would be interesting to see if other facial expressions might be useful.