Sentiment Analysis of Product Reviews: Deep Learning Approach

The proposed work explores the diverse impacts of customer feedback in the e-commerce sector, with a particular focus on online product reviews. Sentiment analysis helps businesses understand consumer satisfaction by gaining insights into customer preferences. The study provides five sentiment categories-Extremely Negative, Negative, Neutral, Positive, and Extremely Positive. The approach integrates feature engineering and deep learning techniques, including tokenization, stop word removal, and lemmatization followed by feature extraction using Word2Vec and sentiment analysis through VADER for sentiment scoring. Convolutional Neural Networks (CNNs) such as Text CNN, GRU, Bi-LSTM and Naïve Bayes are utilized as to effectively harness these features, generating strong representations and capturing contextual information. The results demonstrated that the Text CNN model achieved an overall precision of 0.88 for extremely negative, 0.91 for negative, 0.97 for neutral, 0.96 for positive, and 0.98 for extremely positive sentiments. The accuracy achieved is 96%. The test cases were also compared with other architectures across all sentiments. The work uses visualizations, to display the sentiment distribution of 75 different products. A word cloud is created to highlight the most frequently used terms in reviews categorized as positive or negative. Additionally, a list of products with the highest number of reviews is generated, reflecting their popularity and customer engagement.