Using Roget’s Thesaurus for Fine-grained Emotion Recognition

Using Roget’s Thesaurus for Fine-grained Emotion Recognition Saima Aman School of Information Technology and Engineering University of Ottawa, Ottawa, Canada saman071@site.uottawa.ca Stan Szpakowicz School of Information Technology and Engineering University of Ottawa, Ottawa, Canada ICS, Polish Academy of Sciences Warszawa, Poland szpak@site.uottawa.ca Abstract Recognizing the emotive meaning of text can add another dimension to the under- standing of text. We study the task of automatically categorizing sentences in a text into Ekman’s six basic emotion cate- gories. We experiment with corpus-based features as well as features derived from two emotion lexicons. One lexicon is automatically built using the classification system of Roget’s Thesaurus, while the other consists of words extracted from WordNet-Affect. Experiments on the data obtained from blogs show that a combina- tion of corpus-based unigram features with emotion-related features provides superior classification performance. We achieve F- measure values that outperform the rule- based baseline method for all emotion classes. 1 Introduction Recognizing emotions conveyed by a text can pro- vide an insight into the author’s intent and senti- ment, and can lead to better understanding of the text’s content. Emotion recognition in text has re- cently attracted increased attention of the NLP community (Alm et al., 2005; Liu et al, 2003; Mi- halcea and Liu, 2006); it is also one of the tasks at Semeval-20071. Automatic recognition of emotions can be ap- plied in the development of affective interfaces for 1 Affective Text: Semeval Task at the 4th International Work- shop on Semantic Evaluations, 2007, Prague (nlp.cs.swarthmore.edu/semeval/tasks/task14/summary.shtml). Computer-Mediated Communication and Human- Computer Interaction. Other areas that can poten- tially benefit from automatic emotion analysis are personality modeling and profiling (Liu and Maes, 2004), affective interfaces and communication sys- tems (Liu et al, 2003; Neviarouskaya et al., 2007a) consumer feedback analysis, affective tutoring in e-learning systems (Zhang et al., 2006), and text- to-speech synthesis (Alm et al., 2005). In this study, we address the task of automati- cally assigning an emotion label to each sentence in the given dataset, indicating the predominant emotion type expressed in the sentence. The possi- ble labels are happiness, sadness, anger, disgust, surprise, fear and no-emotion. Those are Ekman’s (1992) six basic emotion categories, and an addi- tional label to account for the absence of a clearly discernible emotion. We experiment with two types of features for representing text in emotion classification based on machine learning (ML). Features of the first type are a corpus-based unigram representation of text. Features of the second type comprise words that appear in emotion lexicons. One such lexicon con- sists of words that we automatically extracted from Roget’s Thesaurus (1852). We chose words for their semantic similarity to a basic set of terms that represent each emotion category. Another lexicon builds on lists of words for each emotion category, extracted from WordNet-Affect (Strapparava and Valitutti, 2004). We compare the classification results for groups of features of these two types. We get good results when the features are combined in a series of ML experiments. 312 2 Related Work …

Content truncated to 500 words