earticle

영어

Sentiment classification task has attracted considerable interest as sentiment information is crucial for many natural language processing (NLP) applications. The goal of sentiment classification is to predict the overall emotional polarity of a given text. Previous work has demonstrate the remarkable performance of Convolutional Neural Network (CNN). However, nearly all this work assumes a single word embedding for each word type, ignoring polysemy and thus inevitably casting negative impact on the downstream tasks. We extend the Skip-gram model to learn multiple sense embeddings for the word types, catering to introduce sense-based embeddings for CNN during sentiment classification. Instead of using the pipeline method to learn multiple sense embeddings of a word type, the sense discrimination and sense embedding learning for each word type are performed jointly based upon the semantics of its contextual words. We validate the effectiveness of the method on the commonly used datasets. Experiment results show that our method are able to improve the quality of sentiment classification when comparing with several competitive baselines.