Sentiment Clasisifcation using varous Text Clasiifcation Techniques using Hybrid XGboost and Machine Learning Techniques

Abstract

In recent years, with the advancement of social media, e-commerce, and online reviews, users keep generating massive amounts of text data continuously. In this sense, sentiment analysis has a vital role in text generation and as a mechanism of understanding a user’s opinion. In marketing research, text analysis serves to capture and monitor consumer sentiment and report social occurrences. Strategies based on a data collection machine (classifier) are a popular and established way to predict consumer attitudes, although they require a significant data collection effort. In addition, most text collection strategies based on machine learning comprise a multiple set of components. In this sense, unidimensional semantic similarity, a measure of the degree of meaning between a set of data are particularly useful in this form. These shortcomings drive the development of adaptive frameworks for sentiment analysis in the future, and the objective of this research is to determine how sentiment analysis can be improved using a “Machine Learning driven XGBoost classifier combined with SVM based similarity mapping. The proposed SVM based Sentiment Analysis (SA) system is inclusive and relies on the analysis of texts that have been pre-processed to a significant degree in which irrelevant and redundant elements have been removed and their volume (dimensionality) has been reduced with the use of TF-IDF matrix construction. In this manner, in cases of data from different sentiment classes, SVM attempts to construct a unique and discrete data point. XGBoost is suggested as the predictive machine of choice in this case. This is because it can capture variables that are non-linear as well as being the least in terms of dimensionality in terms of data variables. The proposed model analyzes the performance on benchmark sentiment analysis datasets with more than two sentiment categories. The experimental results show that the integrated XGBoost-SVM model consistently outperformed the traditional classifiers: Naïve Bayes, standalone SVM, Random Forest, and Logistic Regression. The hybrid model improved overall classification accuracy, F1 score, and generalization, especially for the more challenging and close-meaning sentiment instances. The findings confirm that the addition of similarity-aware learning considerably increased the discriminatory ability of the gradient boosted decision trees. The current study concluded that combining XGBoost with SVM based on similarity techniques is both powerful and highly effective for sentiment classification. The described framework applies state-of-the-art techniques in achieving high accuracy, efficiency, and flexibility, which is ideal for sentiment analysis in real-world applications at scale.