Emotion Models Used in Linguistic Analysis

Kim, S. M., Valitutti, A., & Calvo, R. A. (2010). Evaluation of unsupervised emotion models to textual affect recognition. In Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text (pp. 62–70). Association for Computational Linguistics. Retrieved from http://anthology.aclweb.org/W/W10/W10-0208.pdf  

After providing useful background information on various emotion models used in linguistic analysis, this paper details the findings of a study evaluating techniques for automatically detecting emotions in text. Such automatic recognition aids in reducing the time and expense of utilizing annotators to interpret text, while reducing subjectivity. The authors evaluated two models to recognize four affective states (i.e., anger, fear, joy, and sadness) within three datasets (i.e., SEmEval2007, ISEAR, and children’s fairy tales). The first model was the linguistic lexical resource WordNet-Affect, and the second model was the normative database Affective Norm for English Words. For each model, the authors evaluated three dimensionality reduction techniques including latent semantic analysis (LSA), probabilistic latent semantic analysis (PLSA), and non-negative matrix factorization (NMF) with the goal of affect classification, or predicting a single emotional label given an input sentence. The authors concluded that “NMF-based categorical classification performs the best among categorical approaches to classification. When comparing categorical against dimensional classification, the categorical NMF model and the dimensional model have better performances” (p. 69). The authors recommended research research to identify more effective strategies and improve methodology to work with a generic dataset.  

‍