A neural network model for semi-supervised review aspect identification

• Main Authors: DING, Ying, YU, Changlong, JIANG, Jing
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2017
• Subjects: Entropy; coherence; opinion mining; Aspect identifications; Different domains; Multinomials; Neural network model; Databases and Information Systems; Numerical Analysis and Scientific Computing
• Online Access: https://ink.library.smu.edu.sg/sis_research/3724; https://ink.library.smu.edu.sg/context/sis_research/article/4726/viewcontent/101007_2F978_3_319_57529_2_52.pdf
• Institution: Singapore Management University

Aspect identification is an important problem in opinion mining. It is usually solved in an unsupervised manner, and topic models have been widely used for the task. In this work, we propose a neural network model to identify aspects from reviews by learning their distributional vectors. A key difference of our neural network model from topic models is that we do not use multinomial word distributions but instead embedding vectors to generate words. Furthermore, to leverage review sentences labeled with aspect words, a sequence labeler based on Recurrent Neural Networks (RNNs) is incorporated into our neural network. The resulting model can therefore learn better aspect representations. Experimental results on two datasets from different domains show that our proposed model can outperform a few baselines in terms of aspect quality, perplexity and sentence clustering results.