Sentiment Classification on Customer Reviews using Variants of Deep Learning Frameworks and Embedding Techniques

Abstract

Sentiment analysis has become an essential task in natural language processing (NLP), particularly in understanding customer opinions and feedback. This paper presents a deep learning-based approach for sentiment analysis on review datasets, leveraging NLP techniques and word embeddings. We utilize Amazon review dataset to evaluate the performance of three deep learning models: Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and a Multichannel CNN model. The data preprocessing phase includes stop words removal, tokenization, stemming, and lemmatization, ensuring refined textual input. For word representation, we employ Word2Vec, GloVe, and FastText embeddings to capture contextual meaning effectively. Experimental results indicate that the Multichannel model with FastText embeddings achieves the highest accuracy, outperforming other models in sentiment classification. The findings highlight the significance of word embedding selection and model architecture in improving classification performance.