Aspect Term Extraction via the Fusion of Domain-Specific and Implicit Aspect Information

EasyChair Preprint 11694

Abstract

Aspect term extraction is primarily focused on extracting the main topics described in a text, and it has been widely applied in areas such as e-commerce and sentiment analysis. Most existing methods mainly focus on extracting features from one aspect of ambiguous information or professional domain information, which can result in loss of information from the input sentence sequence. Therefore, in this paper, we propose the Fusion of Domain-specific and Implicit Aspect Information (FDIA) model to improve the accuracy of aspect term prediction. In the FDIA model, firstly, in the representation layer, LDA and VAE-encoder are used to capture the ambiguous feature information of the input sentence sequence and concatenate them to obtain the ambiguous representation of the input sequence. At the same time, professional domain corpus is utilized to obtain the professional information representation of the input sequence. Next, attention mechanism is used to calculate the ambiguous and professional domain feature representations of the input sentence sequence. Finally, through deep mining by BILSTM and Convolution, the output label sequence is obtained.

Keyphrases: BiLSTM, Convolution Neural Network, FDIA Model, LDA, VAE-encoder, aspect term extraction

@booklet{EasyChair:11694,
  author    = {Xiao Wang and Wen Rui Lu},
  title     = {Aspect Term Extraction via the Fusion of Domain-Specific and Implicit Aspect Information},
  howpublished = {EasyChair Preprint 11694},
  year      = {EasyChair, 2024}}