Academic Commons

Presentations (Communicative Events)

Finding Emotion in Image Descriptions

Ulinski, Morgan Elizabeth; Soto Martinez, Victor; Hirschberg, Julia Bell

In this paper, we approach the problem of classifying emotion in image descriptions. A method is proposed to perform 6-way emotion classification and is tested against two labeled datasets: a corpus of blog posts mined from LiveJournal and a corpus of descriptive texts of computer generated scenes. We perform feature selection using the mRMR technique and then use a multi-class linear predictor to classify posts among the Ekman Big Six emotions (happiness, sadness, anger, surprise, fear, and disgust). We find that TFIDF scores on lexical features and LIWC scores are much more helpful in emotion classification than using scores calculated from existing sentiment dictionaries, and that our proposed method performs significantly better than a baseline classifier that chooses the majority class. On the blog posts, we achieve 40% accuracy, and on the corpus of image descriptions, we achieve up to 63% accuracy.

Files

More About This Work

Academic Units
Computer Science
Publisher
WISDOM '12
Published Here
August 2, 2013
Academic Commons provides global access to research and scholarship produced at Columbia University, Barnard College, Teachers College, Union Theological Seminary and Jewish Theological Seminary. Academic Commons is managed by the Columbia University Libraries.