Generating Story Reviews Using Phrases Expressing Emotion

Generating Story Reviews Using Phrases Expressing Emotion Hiroshi Ota and Kazuhide Yamamoto Nagaoka University of Technology, 1603-1, Kamitomioka, Nagaoka, Niigata 940-2188 Japan {ota, ykaz}@nlp.nagaokaut.ac.jp Abstract. This paper presents a method for generating reviews of stories. In this work, we focus on generating sentences that include subjective expressions. First, we constructed lexicons using emotion-emerged expressions that are thought to be the origin of the emotional content. The lexicon consists of syntactic pieces that are proposed as units of syntactic structure to represent suitable units of expression for estimating emotions. We confirmed the effectiveness of the lexicons by estimating the emotional content of blog text. Second, we confirmed the relationship between the naturalness of the sentences and the coherence of the emotion it expresses. Finally, we proposed a method of review generation using the lexicons. Keywords: emotion, syntactic piece, emotion-emerged expression(E3), review, language generation 1. Introduction In many situations, we use subjective expressions that convey emotions, feelings and evaluations about someone or something. It is essential to be able to understand such expressions, and in particular, to be able to use them in language generation tasks. One kind of text that involves subjective expressions is a review of consumer goods and services. In this work, we focus on the task of generating review sentences. This study considers emotional expressions as a subset of subjective expressions. Reviews of stories may include many emotional expressions that emerge in response to the story being reviewed. In this sense, we call the expressions of these emerging emotions ‘emotion-emerged expressions’ (hereafter, E3) in this paper. We first built lexicons for E3, and then made a prototype system to generate story reviews. 2. Related Studies Sentence generation technology has been approached from the viewpoint of the semantic network (Ozaki et al., 1997) or lexical relations (Seki et al., 1999). However, subjective expressions have not been considered. Sentences should be coherent, as emotional coherence affects the naturalness of a sentence. Consider these three sentences: ‧ She receives a present that feels joyous. ‧ She receives a present that is sad. ‧ She receives a present that is afraid. We can select the first sentence ‘She receives a present that feels joyous.’ as the most natural. In this study, we confirm the relationship between the naturalness of a sentence and the coherence of the emotion it expresses. 302 22nd Pacific Asia Conference on Language, Information and Computation, pages 302–310 Related studies on emotion analysis include those of Kozareva et al. (2007), Yang et al. (2007) and Mihalcea and Liu (2006). Kozareva et al. propose an emotional classification approach based on frequency and co-occurrence. They calculate an emotional score for each word using mutual information. A word can express several types of emotions, and we aim to compile these expressions for text generation. Some methods are available for extracting sentences and paragraphs that have emotional content from the blog corpus. Yang et al. used an emoticon as a marker for emotional expression and extracted emotion expressing sentences. Mihalcea and Liu used …

Content truncated to 500 words