Valence-arousal ratings prediction of Chinese words using similarity measures based on Word2Vec

摘要

In this paper the systems submitted by the joint team of Dublin City University and National Taiwan University to the IALP 2016 Shared Task: Dimensional Sentiment Analysis for Chinese Words are presented. The systems learn the vector representation using Word2Vec algorithm for each Chinese word for sentiment analysis. The corpus used for the calculation of vector representation is 5 years (2006 to 2010) of the LDC Chinese Gigaword Fifth Edition corpus. The systems calculated similarities between a test Chinese word and each word in training corpus of the shared task with human annotation and took the valence-arousal ratings of the most similar words as the ratings of the test word. The performance of the submitted systems are around the same level of the shared task's baseline system. We will be looking at the performance gap with top-ranked systems in several aspects including corpus used for training and methodology.