Neural Networks with Emotion Associations, Topic Modeling and Supervised Term Weighting for Sentiment Analysis

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Artificial Intelligence

Background:

• Automated sentiment analysis is crucial for social media and e-commerce.
• Deep neural networks offer dense document representations but struggle with topical semantics and sentiment polarity.
• Existing methods require improvement for comprehensive sentiment understanding.

Purpose of the Study:

• To propose a novel sentiment analysis model with richer document representations.
• To improve the capture of topical semantics and sentiment polarity.
• To enhance the performance of sentiment analysis in review datasets.

Main Methods:

• Integrating word embeddings with lexicon-based sentiment and emotion indicators (negations, emoticons).
• Utilizing word-emotion associations and topic models for richer document representation.
• Employing a bag-of-words model with a supervised term weighting scheme.

Main Results:

• The proposed model demonstrates valid document representation for sentiment analysis.
• Effectiveness is proven on large datasets of Amazon product and hotel reviews.
• The model performs well irrespective of class imbalance in datasets.
• Experimental results show improvement over existing machine learning methods.

Conclusions:

• The novel document representation effectively captures topical semantics and sentiment polarity.
• The proposed sentiment analysis model offers superior performance in review analysis.
• This approach advances automated sentiment analysis for e-commerce and social media platforms.