Sentiment Thesaurus, Synset and Word2Vec Based Improvement in Bigram Model for Classifying Product Reviews

Area of Science:

• Natural Language Processing (NLP)
• Computational Linguistics
• Sentiment Analysis

Background:

• Product review classification is crucial for business growth via customer satisfaction.
• Traditional bigram models in NLP for sentiment analysis overlook semantic similarity.
• Customer reviews often use varied vocabulary to express similar sentiments.

Purpose of the Study:

• To propose and evaluate improved bigram models for product review sentiment classification.
• To address the limitation of traditional bigram models by incorporating semantic word relationships.
• To enhance the accuracy of sentiment identification in product reviews.

Main Methods:

• Developed improved bigram models utilizing semantically similar words.
• Constructed a sentiment polarity thesaurus with sentiment words and synonyms.
• Employed Synset and Word2Vec for synonym extraction to enrich bigram features.
• Compared proposed models against traditional bigram models and state-of-the-art methods.

Main Results:

• The proposed improved bigram models demonstrated superior performance in sentiment classification.
• Incorporating semantically similar words led to more accurate sentiment identification.
• The enhanced models outperformed both traditional bigram approaches and existing advanced methods.

Conclusions:

• Improved bigram models incorporating semantic similarity offer a more effective approach to product review sentiment analysis.
• The methodology enhances the ability to capture nuanced customer sentiment expressed through diverse vocabulary.
• This research contributes to advancing NLP techniques for better business intelligence and product development.