Sentiment analysis using averaged weighted word vector features.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Machine Learning

Background:

• Online reviews are crucial for consumer decisions and measuring product/service satisfaction.
• Sentiment analysis aims to identify opinions in text data.
• Existing methods for sentiment analysis can be improved through advanced techniques.

Purpose of the Study:

• To develop and evaluate two novel methods for sentiment analysis using combined word vectors.
• To estimate the polarity of reviews by learning from weighted word frequencies.
• To compare the proposed methods against existing sentiment analysis benchmarks and literature approaches.

Main Methods:

• Combining different types of word vectors to create average review vectors.
• Weighting review vectors using word frequencies from sensitivity-tagged reviews.
• Ensembling the developed techniques with each other and with existing sentiment analysis methods.

Main Results:

• The proposed methods were applied to multiple standard sentiment analysis benchmark datasets.
• Performance was evaluated by comparing against state-of-the-art approaches.
• The developed sentiment analysis techniques demonstrated superior performance over existing methods.

Conclusions:

• The novel methods for sentiment analysis show significant improvements in estimating review polarity.
• Combining word vectors and frequency-based weighting offers a powerful approach for opinion mining.
• This research contributes to more accurate analysis of online consumer feedback.