Identifying Expressions of Emotion in Text

Related Work

Some researchers have studied emotion in a wider framework of private states [12]. Wiebe et al. [20] worked on the manual annotation of private states including emotions, opinions, and sentiment in a 10,000-sentence corpus (the MPQA corpus) of news articles. Expressions of emotions in text have also been studied within the Appraisal Framework [5], a functional theory of the language used for conveying attitudes, judgments and emotions [15, 19]. Neither of these frameworks deals exclusively with emotion, the focus of this paper.
In a work focused on learning specific emotions from text, Alm et al. [1] have explored automatic classification of sentences in children's fairy tales according to the basic emotions identified by Ekman [3]. The data used in their experiments was manually annotated with emotion information, and is targeted for use in a text-to- speech synthesis system for expressive rendering of stories. Read [14] has used a corpus of short stories, manually annotated with sentiment tags, in automatic emotion-based classification of sentences. These projects focus on the genre of fiction, with only sentence-level emotion annotations; they do not identify emotion indicators within a sentence, as we do in our work.
In other related work, Liu et al. [4] have utilized real-world knowledge about affect drawn from a common-sense knowledge base. They aim to understand the semantics of text to identify emotions at the sentence level. They begin with extracting from the knowledge base those sentences that contain some affective information. This information is utilized in building affective models of text, which are used to label each sentence with a six-tuple that corresponds to Ekman's six basic emotions [3]. Neviarouskaya et al. [8] have also used a rule-based method for determining Ekman’s basic emotions in the sentences in blog posts.
Mihalcea and Liu [6] have focused in their work on two particular emotions – happiness and sadness. They work on blog posts which are self-annotated by the blog writers with happy and sad mood labels. Our work differs in the aim and scope from those projects: we have prepared a corpus annotated with rich emotion information that can be further used in a variety of automatic emotion analysis experiments.

3.