Learning dimensional sentiment of traditional Chinese words with word embedding and support vector regression

Dimensional sentiment analysis approach, which represents affective states as continuous numerical values on multiple dimensions, such as valence-arousal (VA) space, allows for more fine-grained analysis than the traditional categorical approach. In recent years, it has been applied in applications such as antisocial behavior detection, mood analysis and product review ranking. In this approach, an affective lexicon with dimensional sentiment values is a key resource, but building such a lexicon costs much. To reduce the cost and speed up the process, we explore methods to expand a seed lexicon automatically in this study. A hybrid approach combining word embedding and support vector regression is proposed. Preliminary experiments on the dataset of the IALP-2016 shared task demonstrate its effectiveness.