Opinion target extraction for short comments

  • Authors:

  • Lin Shang;Haipeng Wang;Xinyu Dai;Mengjie Zhang

  • Affiliations:

  • State Key Laboratory of Novel Software Technology, Nanjing University, Nanjing, China;State Key Laboratory of Novel Software Technology, Nanjing University, Nanjing, China, Trend Micro China Develop Center, Nanjing, China;State Key Laboratory of Novel Software Technology, Nanjing University, Nanjing, China;School of Engineering and Computer Science, Victoria University of Wellington, Wellington, New Zealand

  • Venue:
  • PRICAI'12 Proceedings of the 12th Pacific Rim international conference on Trends in Artificial Intelligence
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Target extraction is an important task in sentiment analysis. Many existing methods have worked well in news and blogs. However, they are not effective for short product comments. In this paper, we firstly prove that a well-known method, Ku's method, cannot obtain good results for short comments. Then we propose a new method to extract opinion targets by developing a two-dimensional vector representation for words and a back propagation neural network for classification. The proposed method is examined and compared with two well-known opinion extraction methods (Ku's and LDA methods) on an crawled network mobile phone corpus from "Zhongguancun online" with 14408 comments. The strict evaluation and the lenient evaluation are used in the experiments to determine the goodness of the extracted opinion targets. Experimental results show that under the strict evaluation, the proposed method can achieve better precision by 8.33% improvement over Ku's and 16.67 % improvement over LDA. Under the lenient evaluation, the proposed method can achievableve a 28.33% improvement in precision over Ku's and 33.33% over LDA. In addition, the opinion targets extracted by our method are much closer to the true topics and much more meaningful than those extracted by the other two methods.