Auto-Parse: Automated Preprocessing And Sentiment Analysis Engine for Automobile Reviews

Abstract

The current sentiment analysis approaches struggle with noisy, unstructured text due to insufficient preprocessing and sentiment classification, resulting in incorrect predictions.  To address these issues, this paper introduces AUTO-PARSE (Automated Preprocessing and Sentiment Analysis Engine), a framework that improves text preprocessing and classification for better sentiment prediction. AUTO-PARSE incorporates contraction handling, slang replacement, duplicate review removal utilizing Word Mover's Distance (WMD), and effective text normalization to improve input data. It utilizes a boosted hybrid classification model, combining IB1, BayesNet, ADTree, REPTree, AdaBoostM1, and Bagging classifiers to optimize sentiment polarity detection. Utilizing the English automobile dataset, AUTO-PARSE is assessed through word-frequency-based prediction (Prediction_1) and a hybrid model-based technique (Prediction_2), incorporating both for final classification. Results show its superiority over traditional SA models, attaining 89.2% accuracy, 90.4% precision, 89.2% recall, 89.0% F1-score, and 77.7% MCC. By incorporating sophisticated text preprocessing with a resilient classification framework, AUTO-PARSE provides a scalable and high-accuracy SA solution, significantly enhancing sentiment classification in the automobile industry.