WORD EMBEDDING FEATURE DESIGN FOR PRODUCT REVIEW CLASSIFICATION

Abstract

Sentiment analysis has emerged as an important research area in Natural Language Processing (NLP) due to the massive growth of user-generated content on e-commerce platforms and social media. Online product reviews significantly influence consumer purchasing decisions and business strategies, making automated review classification an essential task. Traditional text representation techniques such as Bag-of-Words and TF-IDF rely on sparse vectors and fail to capture semantic relationships between words, which often leads to limited classification performance. To address these challenges, this work presents a system for product review classification using word embedding feature design combined with deep learning techniques. In the proposed approach, textual product reviews are preprocessed through tokenization, stop-word removal, and normalization to prepare the data for feature extraction. Word embeddings are generated using FastText, which captures semantic similarity and contextual meaning by representing words in dense vector space while also handling subword information. The embedded representations are then fed into a Bidirectional Long Short-Term Memory (Bi-LSTM) neural network integrated with an attention mechanism to capture contextual dependencies within sentences and emphasize important sentiment-bearing words. The system classifies reviews into three sentiment categories: positive, negative, and neutral. A web-based interface allows users to submit product reviews and obtain real-time sentiment predictions. Experimental results demonstrate that the proposed word embedding-based architecture improves classification accuracy and contextual understanding compared with traditional feature engineering methods. The system provides an efficient, scalable, and interpretable solution for automated product review sentiment analysis and can assist businesses in understanding customer feedback and improving decision-making processes