Implicit mood computing via LSTM and semantic mapping
- PDF / 5,473,882 Bytes
- 15 Pages / 595.276 x 790.866 pts Page_size
- 31 Downloads / 237 Views
METHODOLOGIES AND APPLICATION
Implicit mood computing via LSTM and semantic mapping Chang Su1 · Junchao Li1 · Ying Peng1 · Yijiang Chen1
© Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract This article proposes an implicit mood computing system. The implicit mood computing task is a part of affective computing. Previous works in affective computing mostly focus on twitters, blogs, movie interviews, and news corpus. These works detect sentiment polarity (positive/negative), emotion types (joy, sadness, anger, etc.), or mood types (boring, tired, happy, etc.) of the text. Different from previous studies, our work focuses on the literature texts and detects the implicit mood of them. The implicit mood is sometimes discussed as the tone or the atmosphere of the text. The implicit mood is an important affective feature in the literature such as poetry, prose, and drama. Our work regards the implicit mood as a semantic phenomenon. We capture the feature of implicit mood via a semantic mapping approach and the long short-term memory neural network. The proposed system is capable of identifying 12 kinds of implicit moods with a promising result. Keywords Affective computing · Mood · LSTM · Semantic mapping
1 Introduction Affects play a critical role in human society (Frijda and Mesquita 1994; Mesquita and Leu 2007; Cambria 2011). Affective computing is a way to endow computers the ability to detect, understand, and express affects as humans. Previous affective computing systems in nature language processing (NLP) area tend to narrow affect categories to basic emotions (Ekman 1992; Plutchik 2012), a list of psychological mood labels (Mishne and Rijke 2006), or sentiment polarities. The basic emotions always include Joy, Sadness, Anger, Disgust, Surprise, Fear, Trust, and Anticipation; the psychological mood labels contains Amused, Boring, Tired, etc.; the sentiment polarities can be divided into positive, negative, and neutral. These settings make it convenient to discuss affective issues in the twitter, blog, and product review-related studies. However, it is not enough in facing literature texts. Literature texts, such as poetry, prose, novel, and drama, usually arouse the feelings of the readers by creating a certain kind of background and setting. For instance, Communicated by V. Loia.
B
Chang Su [email protected] Yijiang Chen [email protected]
1
the sad stories usually happen in the cold winter or hazy rainy days, while the happy love story tends to have a rosy romantic background. The total of the setting and background in the literature texts are called as “mood” of the text (contributors 2017a). Our work believes the mood is an important affective feature in the literature affective computing. We call the mood as “implicit mood” and develop an LSTM and semantic mapping-based method to detect the implicit mood of the literature texts. Our work calls the literature sense of mood as “implicit mood“ for two reasons: firstly, the function of the mood implies that it is related to affect but the re
Data Loading...
