Aspect-Specific Oriented Fine-Grained Sentiment Classification

With the popularity of the Internet and the rapid development of e-commerce platforms,online shopping has become the most popular way of shopping,the number of user product reviews is increasing every day on the electric business platform.It is not enough to rely on artificial methods to deal with the huge amount of online reviews,so review mining emerges as an automated review analysis method to solve this problem.However,the traditional coarse-grained sentiment analysis methods only identifies the consumers ' sentiment polarity towards the product as a whole,but ignores the important attribute information,which leads to the inability to refine consumer preferences and clarify the advantages and disadvantages of commodity attributes.To solve the above problems,this paper proposes aspect-specific oriented fine-grained sentiment classification,and the following works are carried out:(1)In view of the lack of the resources of the sentiment classification corpus for the aspect-specific,a reasonable annotation system of the aspect-specific sentiment analysis corpus is explored,which is based on the labeling standard of English corpus,and then start a large-scale manual annotation.(2)Most of the existing online reviews contain multiple aspects,while the traditional sentiment analysis ignores the important aspect information in the sentiment classification,an aspect-specific sentiment classification method based on high-dimensional representation is proposed,which constructs a multi-level and high-dimensional deep neural network model from three different dimensions of words,clauses and sentences,using the reviews text and its aspect-specific information.Experimental results show that our high-dimensional representation can achieve better classification performance than other baseline methods.(3)The contribution of each part of the review text to the classification result is different for the given aspect words.Based on the high-dimensional representation method,we also add attention represent learning,and propose an aspect-specific sentiment classification method based on attention mechanism.This method focuses on the general text and the question and answer text using the attention mechanism to capture the part of the review text that is more closely related to a specific aspect.Experimental results show that the proposed method can further improve the classification performance than the high-dimensional representation method.Besides,the application of attention mechanism to task of question and answer text sentiment classification can also effectively improve classification performance.