Feature extraction from customer reviews using enhanced rules

Area of Science:

• Natural Language Processing
• Sentiment Analysis
• Information Extraction

Background:

• Opinion mining is crucial for understanding customer sentiment and purchasing decisions.
• Extracting opinion features from unstructured reviews is challenging due to language errors and limitations of existing pattern rules.
• Current methods often miss relevant features not strictly nouns or adjectives.

Purpose of the Study:

• To enhance the performance of explicit feature extraction from product review documents.
• To identify and extract features with associated opinions using sequential pattern rules.
• To address limitations in existing rules for feature extraction.

Main Methods:

• Proposed an approach employing sequential pattern rules for feature and opinion extraction.
• Developed 16 new pattern rules, combined with 25 existing rules, totaling 41 rules.
• Evaluated the approach on five datasets.

Main Results:

• The new set of 16 rules significantly improved feature extraction.
• Achieved average precision of 0.91, recall of 0.88, and F-measure of 0.89.
• Demonstrated effectiveness in extracting previously overlooked features.

Conclusions:

• The study successfully enhanced explicit feature extraction in opinion mining.
• The proposed sequential pattern rules effectively address gaps in existing methods.
• The improved approach provides a more accurate understanding of customer expectations from reviews.