Aspect-specific Sentiment Classification Method Based on High-dimensional Representation

Hanqian Wu, Lu Cheng, Jie Li, Jingjing Wang, Jue Xie

Abstract


Aspect-specific sentiment classification is a fine grained sentiment classification task. The traditional coarse-grained sentiment analysis methods only identify 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 this problem, we use the review text and its specific aspect information to construct a multi-level, high-dimensional deep neural network model. First, the clause segmentation algorithm is used to divide the review text into several clauses; secondly, the words in the clause are encoded by the bidirectional long short term memory neural network, and the vector representation of the clauses is obtained. Finally, the vector representation of the whole review text is obtained by using the bidirectional long short term memory neural network to code all the clauses. And then the sentiment polarity of review text is obtained by the softmax layer. The experimental result shows that the proposed method can effectively improve the performance of sentiment classification.

Keywords


Aspect-specific; Sentiment Classification; High-dimensional Text Representation; lstm


DOI
10.12783/dtcse/iciti2018/29097

Full Text:

PDF

Refbacks

  • There are currently no refbacks.