Identification of significant features and machine learning technique in predicting helpful reviews

Area of Science:

• Natural Language Processing
• Machine Learning
• Consumer Behavior

Background:

• Consumers heavily rely on online reviews for purchase decisions.
• The overwhelming volume of online reviews necessitates identifying the most helpful ones.
• Existing research has not fully utilized available features to predict review helpfulness.

Purpose of the Study:

• To identify significant features for predicting helpful online reviews.
• To compare the performance of different machine learning models in predicting review helpfulness.
• To enhance the consumer decision-making process by filtering valuable reviews.

Main Methods:

• Utilized linguistic, metadata, readability, subjectivity, and polarity features.
• Applied five machine learning models to two large Amazon open datasets.
• Employed the Random Forest technique for its predictive capabilities.

Main Results:

• The Random Forest model, using the identified significant features, achieved an accuracy of 89.36%.
• This performance surpassed other machine learning techniques evaluated in the study.
• Key features contributing to helpfulness were identified across linguistic, metadata, readability, subjectivity, and polarity categories.

Conclusions:

• The proposed feature set and Random Forest model effectively predict helpful online reviews.
• This approach can significantly aid consumers in navigating product review information.
• Further research can explore additional features and models for even greater accuracy in helpfulness prediction.