Repository | Book | Chapter

A central challenge in building sentiment classifiers using machine learning approach is the generation of discriminative features that allow sentiment to be implied. Researchers have made significant progress with various features such as n-grams, sentiment shifters, and lexicon features. However, the potential of semantics-based features in sentiment classification has not been fully explored. By integrating PropBank-based semantic parsing and class association rule (CAR) mining, this study aims to mine patterns of semantic labels from domain corpus for sentence-level sentiment analysis of product reviews. With the features generated from the semantic patterns, the F-score of the sentiment classifier was boosted to 82.31% at minimum confidence level of 0.75, which not only indicated a statistically significant improvement over the baseline classifier with unigram and negation features (F-score = 73.93%) but also surpassed the best performance obtained with other classifiers trained on generic lexicon features (F-score = 76.25%) and domain-specific lexicon features (F-score = 78.91%).

DOI: 10.1007/978-3-319-67008-9_30

Full citation: