Abstract
With the steadily increasing volume of e-commerce transactions, the amount of userprovided product reviews is increasing on the Web. Because many customers feel that they can purchase product based on the experiences of others that are obtainable through product reviews, the review summarization process has become important. In particular, feature-based product review summarization is needed in order to satisfy the detailed needs of some customers. To achieve such summarization, numerous techniques using natural language processing (NLP), machine learning, and statistical approaches that can evaluate product features within a collection of review documents have been studied. Many of these techniques require sentiment analysis or feature scoring methods. However, existing sentiment analysis methods are limited when determining the sentiment polarity of context-sensitive words, and existing feature scoring methods are limited when only the overall user score is used to evaluate individual product features. In our summarization approach, context-sensitive information is used to determine sentiment polarity while opinioned-feature frequency is used to determine feature scores. Based on experiments with actual review data, our method improved the accuracy of the calculated feature scores and outperformed existing methods.
Original language | English |
---|---|
Pages (from-to) | 1973-1990 |
Number of pages | 18 |
Journal | Journal of Information Science and Engineering |
Volume | 26 |
Issue number | 6 |
State | Published - Nov 2010 |
Keywords
- Feature scoring
- Opinion mining
- Product review summarization
- Sentiment analysis
- User score