Research On Aspect Based Sentiment Analysis

With the development of E-commerce, more and more people choose online shopping. A massive of online comments become the main reference for user consumption. The comments can provide both objective description of products and the preference of users. Extracting the attribute information from reviews and structured organized can not only save the consumers' time, but also allow the electronic businesses understand the preference of users, which has great practical significance.In this paper, we extract aspects based commodity features and analyze the sentiment of these aspects. We define the aspect based sentiment analysis task as text classification task and propose the extraction algorithm based on characters. These model are based on characters, which not only save the preprocessing, such as word segmentation and part-of-speech tagging, but also avoid the noise according to the preprocessing. We use character based deep neural network to extract aspect. The method use the raw text as input and labels as supervised information to learn the latent features for classification. We propose the character based sequential models for our task. The sequential models retain both the original sequence and syntax information. The hybrid sequential model combine the convolutional neural network and long short term memory, which can obtain the richer expression of text.For aspect based sentiment analysis task, we use tag based convolutional neural network to judge the emotion of users. We introduce the relative position of each character to make the sentiment classification task more accurate.In this paper, we use character level deep neural network for aspect based sentiment analysis task. Experiments show that our models outperform state-of-art methods in real datasets.