Bipolar Sentiment Analysis of Japanese Social Media Posts: A Semantic Similarity Based Approach

Abstract

Social media generates a colossal amount of data round the clock. Many social entities including governments, business organizations, and researchers harness insights from these social-media data. With the recent advancement in algorithms based on machine learning, deep learning, and language models, automatic extraction of sentiments from textual social-media posts has become highly effective. These algorithms, however, are often need huge amount of labeled data for training, making them largely inapplicable when such data datasets do not exist-as is the case with Japanese social-media posts. Hence, in this paper, we propose an alternative approach that combines machine-learning-based word embedding and polarity dictionaries for classifying a social-media post written in Japanese as either positive or negative. Our experiments have demonstrated promising results with this approach which does not require any labeled dataset.