A Word2vec Model for Sentiment Analysis of Weibo

Abstract

The booming of online social media has provided a platform for massive users to share viewpoints and emotional experiences. A huge volume of digital traces that accumulate and aggregate on social media provide a more efficient proxy for investigating users' behaviors, thoughts and emotions. How to precisely and effectively acquire the emotions and topic keywords from these short and colloquial texts is the key task in the analysis of social media. Through neural networks, Word2vec offers a unique contribution to embedding vector construction and expanding similar words. Based on a huge volume of texts on Sina Weibo, the most popular Twitter-like service in China, this paper presents a Word2vec model bringing extra semantic features to fit for short Chinese texts. By comparing with the model based on Internet contents of long texts, the experimental results illustrate that our model can effectively improve the performance of sentiment classification with six categories on Weibo. Furthermore, a series of application results demonstrate the usability and adaptability of our model for massive data on social media.