Sentiment Analysis Of Short Text With Deep Neural Networks

Sentiment Analysis is classifying the emotional polarity of text into several emotional categorys. In this paper, we focus on sentiment analysis of Chinese short text. First we use bag of words to do text modeling for Chinese short text. Then we train SoftMax Regression model for classifying the emotional polarity. To achieve impressive performance for such task, we also use deep learning method to represent short text. Based on bag of words representation, we train deep neural network to learn a new representation of short text. Deep learning can discover multiple levels of distributed representations, with higher levels representing more abstract concepts.Besides, the paper proposes some new methods to more effectively represent the short text. More specifically, two major approaches have been exploited to enrich the representation of short text. One is to derive latent topics from existing large corpus, which are used as features to enrich the representation of short text.The other is to learn word vector from large corpus and then uses the word vector to get the representation of short text. The first method derives latent topics of certain granularity through well-known topic models such as Latent Dirichlet Allocation. The second method gets text representation based on neural network language model and k-means algorithm. After getting the representation of short text, we train SoftMax regression model to predict the emotional polarity.In order to verify the effectiveness of the proposed algorithms, we use large-scale product reviews in our experiments. We design a spider algorithm to crawl data through a famous web site and then use these data to test our algorithm results. The experiments results show that our algorithms have a better performance and the imporved algorithm can get better representation.