An enhancement of fuzzy support vector machine using natural language processing model applied on online product reviews

ABSTRACT: The increasing complexity and volume of online product reviews pose significant challenges to traditional sentiment analysis methods, especially in multilingual and informal language contexts. Conventional Fuzzy Support Vector Machine (CFSVM) models demonstrate limitations in analyzing n on-English texts, understanding modern slang, and recognizing cultural nuances such as sarcasm and ambiguous expressions. This study addresses these shortcomings by enhancing FSVM through the integration of advanced Natural Language Processing (NLP) techniques, specifically BERT, DistilBERT, and Multilingual BERT (mBERT). The primary objective is to develop a sentiment analysis model capable of handling Filipino language input, neologisms, and culturally nuanced expression within product reviews. The researchers employed a hybrid methodology combining TF-IDF feature extraction, NLP-based language modeling, and a fuzzy inference system (FIS) that utilizes Gaussian membership functions and Bayesian rule sets to refine sentiment prediction. The model architecture was implemented using Python, PyCharm, and Streamlit, with testing performed on e-commerce review datasets. The results showed that the enhanced FSVM model outperformed traditional FSVM and other baseline approaches in terms of accuracy, adaptability, and interpretability. Specifically, it achieved better classification performance on multilingual and informal text data, effectively handling linguistic ambiguity and evolving language use. This enhancement contributes to the broader field of sentiment analysis by offering a robust, interpretable, and context-aware solution suitable for diverse linguistic environments, with practical implications for e-commerce platforms, marketing analysis, and natural language research.

Filipiniana