Aspect-Based and Multi-Level Sentiment Information Applying Contrast Dictionary

摘要

Customer feedback provides alternative and important sources to discover knowledge supporting the marketers and customers to make better decisions. However, the manual process to extract useful information depends on domain experts. This paper focuses on improving the performance of the automatic sentiment information extraction from customer feedback. The article proposes a new extraction method that consider multiple dimensions of feedback information, aspect, word, contrast, sentence or phrase, and document levels. The aspect-based sentiment extraction uses a named entity recognition technique to extract the desired aspects of a target product. The aspect-based sentiment combines with sentiment information from multiple levels of feedback contexts resulting in the fused sentiment information improving the extraction performance. The authors validate the effectiveness by measuring the accuracy of the sentiment and aspect recognition methods comparing with SentiStrength and Word-Count. This information gives some insights on customer satisfaction and can be applied in an alarming tool.