Hyperparamaters Fine Tuning for Bidirectional Long Short Term Memory on Food Delivery

Abstract

Food delivery is growing rapidly in Indonesia. Every food delivery order holds big promotions to attract users’ attention, so it has advantages and disadvantages. However, users only focus on evaluating drivers and restaurants, so the company does not get feedback on its services. This research aimed to understand user sentiment and maximize model accuracy with hyperparameters and fine-tuning. Sentiment analysis can be used to determine user sentiment based on reviews, and the results of this analysis can provide suggestions for companies. The bidirectional long short-term memory method was used for sentiment analysis to understand a word’s meaning better. The Bidirectional Short-Term Memory model andWord2Vec extraction features were proven to be better than several other extraction modelsand features. The dataset was balanced, and the hyperparameters in the model and optimization could also improve accuracy. So, the Gofood and Shopeefood research results had an accuracy of 98.1%, and Grabfood’s was 97.4%.