Hyperparameter Optimization of Machine Learning Models Using Grid Search for Amazon Review Sentiment Analysis

In this study, we utilize machine learning algorithms with a range of hyperparameters, which are tuned using grid search. The primary impediment is identifying the proper set of hyperparameter values, which directly influence the model’s performance. Additionally, we have employed four techniques for feature extraction, namely bag-of-words (BoW), term frequency-inverse document frequency (TF-IDF), average word2vec (AvgW2V), and TF-IDF word2vec (TF-IDF W2V). The vectors obtained are fed as input to the decision tree (DT), logistic regression (LR), support vector machine (SVM), naive Bayes (NB), XGBoost (XG), and random forest (RF) models whose hyperparameters have been fine-tuned for optimal performance using grid search. The contribution of this work is to determine the best parameter values that result in the best model while dealing with high training costs and uncertain prediction outcomes when using different parameter values. To address this, we proposed a method called Grid Search Hyperparameter Optimization (GSHO) on Amazon women’s e-commerce clothing review dataset. It was observed from our research that BoW and TF-IDF are the best feature extraction techniques compared to others. According to our findings, adjusting the hyperparameters of the model had a noteworthy beneficial effect on its overall performance.