Construction of a Multilingual Annotated Corpus for Deeper Sentiment Understanding in Social Media

Abstract

The surge of social media makes it possible to understand people’s emotion in different cultures. In this paper, we construct an annotated corpus for multilingual sentiment understanding. The annotation is developed in a multilingual setting including English/Japanese/Chinese, and on a representative dataset including 4 topics (spanning 3 genres, which are product, people, and event).To deep understand expression mechanism of feeling entailed in the text, we labelled sentimental signal words and rhetoric phenomenon in addition to overall polarity. This innovative corpus can be a helpful resource for the improvement of sentiment classification, cross-cultural comparison etc.