Product Related Information Sentiment-Content Analysis Based on Convolutional Neural Networks for the Chinese Micro-Blog

Abstract

Sentiment analysis is a hot topic in natural language processing. People usually concern the sentiments in a specific domain. By focusing on specific domain such as micro-blogs of a specific product, we can discover the sentiment distribution of customs about this product. Unlike traditional sentiment analysis which divides the sentiments into positive/negative and neutral categories, we divided the product related information into three types: positive/negative about the product and product advertisements. The reason for this kind division is that ordinary positive category in sentiment analysis may contains product advertisements which do not reflect custom's opinions. This phenomenon is very common in micro-blog where there are lots of product promotions and propagandas. In this paper, we propose a sentiment analysis method based on Convolutional Neural Network (CNN). The proposed CNN model combines Chinese word-level embedding and Chinese character-level embedding as input features of the sentiment analysis. In the experiment, Chinese micro-blog datasets of weight loss products and mobile phones are used to show that our sentiment analysis method has better performance than the lexicon based method.