Exploring the Effectiveness of Deep Learning in Analyzing Review Sentiment

Abstrak

This study aimed to analyze sentiment in office product reviews by using word embedding with three neural network modeling approaches: Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU). Office product review data is taken from Amazon's reviews of office products covering a wide range of sentiments. Word embedding converts text into a numerical vector representation for neural network processing. Experimental comparison of this model reveals that CNN achieves the highest accuracy, 77.99%. The CNN model effectively extracts significant features from review text, improving sentiment classification performance. Although the LSTM and GRU models show satisfactory results, they do not match CNN performance. These findings demonstrate the effectiveness of word embedding and neural networks for sentiment analysis in office product reviews. This provides valuable insights for companies to improve their products based on user feedback from online reviews. Additionally, this research serves as a foundation for further advances in sentiment analysis across a wide range of other products and services