Dimensional Sentiment Analysis for Chinese Words Based on Synonym Lexicon and Word Embedding

This paper is mainly about the BIT group submitted system to the IALP-2016 Shared Task. This system is to automatically acquire the valence-arousal ratings of Chinese affective words. Two ways are designed to generate a given word's VA: one is based on Synonym Lexicons and the other is based on Word Embeddings. For the first way, we extend the annotated set based on synonym lexicon to improve coverage of unknown words, and then search test words or characters split from unknown words in extended annotated set. For the second way, we broaden the words coverage by building a local words segmentation lexicon in the vector space model. The cosine similarity is used to measure the distance between the test word and the annotated word. According to the experimental results, the strategy based on the synonym lexicons is better than the one based on the word embeddings, and makes our group in upper grades among 20 teams approximately.