Food and Beverage Product Review Sentiment Analysis on E-Commerce with Word Embedding and LSTM

Abstract

Sentiment analysis is a widely used method to understand customer opinions about a product. This study aims to analyze the sentiment of food and beverage product reviews on the Tokopedia marketplace using the Long Short-Term Memory (LSTM) approach and word embedding. The data used consisted of customer reviews that were categorized into three sentiment classes, namely positive, neutral, and negative. The model was developed through a series of stages of preprocessing, embedding, training with LSTM, as well as performance evaluation using accuracy and F1-score metrics. The results show that the developed model is able to classify sentiment with a fairly high level of accuracy. Based on the results of the final test on 5,000 data, the model managed to classify 122 data as negative, 130 data as neutral, and 4,871 data as positive, although it still showed an imbalance in class classification. Further analysis through word cloud visualization showed that words like "delicious", "steady", and "good" dominated the positive sentiment, while words like "disappointed", "broken", and "slow" often appeared in negative sentiment. This study provides valuable insights for businesses in understanding customer opinions and improving the quality of products and services.