Context-based sentiment analysis on customer reviews using machine learning linear models.

Area of Science:

• Natural Language Processing
• Machine Learning

Background:

• Analyzing high volumes of customer reviews for sentiment analysis is challenging and time-consuming.
• Existing methods like deep learning (DL), Artificial Neural Network (ANN), and bag-of-word (BOW) models face issues such as polarity incoherence, overfitting, and high data processing costs.

Purpose of the Study:

• To develop a high-performance, cost-effective model for accurate sentiment prediction from large customer review datasets.
• To address the limitations of existing sentiment analysis models.

Main Methods:

• Utilized the fastText library for text classification and word embedding.
• Employed a traditional Linear Support Vector Machine (LSVM) for comparison.
• Developed and compared against a custom multi-layer Sentiment Analysis (SA) Bi-directional Long Short-Term Memory (SA-BLSTM) model.

Main Results:

• The proposed fastText model achieved an accuracy of 90.71%.
• The fastText model demonstrated a 20% performance improvement compared to LSVM and SA-BLSTM models.
• Indicated superior accuracy and performance over traditional and custom deep learning models.

Conclusions:

• The fastText model offers a highly accurate and cost-effective solution for customer sentiment analysis.
• This approach effectively overcomes the challenges associated with processing large volumes of customer reviews.
• The study validates fastText as a superior alternative for sentiment prediction tasks.