Content-Based Filtering in Recommendation Systems Culinary Tourism Based on Twitter (X) Using Bidirectional Gated Recurrent Unit (Bi-GRU)

Abstract

To address the challenge of information overload in the rapidly expanding culinary sector, a recommendation system using Content-Based Filtering (CBF) and the Bidirectional Gated Recurrent Unit (Bi-GRU) algorithm was developed. This system can help users to suggest culinary options based on user profiles and preferences. Twitter (X) is frequently used to gather culinary reviews in Bandung, forming the foundation for developing recommendation systems. This research contributes to integrating CBF and Bi-GRU to enhance the relevance of culinary recommendations. The system uses Term Frequency-Inverse Document Frequency (TF-IDF) for feature extraction and Cosine Similarity for item matching. Research adapting CBF and Bi-GRU methods specifically for culinary recommendations, especially in Bandung, remains limited. This study focuses on evaluating the performance of a culinary recommendation system. Data collected from Twitter (X) and PergiKuliner includes 2,645 reviews from 44 Twitter (X) accounts and on 200 culinary places. The culinary recommendation model, using CBF with TF-IDF and Cosine Similarity, achieved a Mean Absolute Error (MAE) of 0.254 and Root Mean Square Error (RMSE) of 0.425, indicating high accuracy in rating predictions compared to previous studies. From the experiments conducted, the third experiment using Bi-GRU, SMOTE, and the Nadam algorithm showed the best improvement with a learning rate of 0.014563484775012459, achieving an accuracy of 86.8%, precision of 86.3%, recall of 85.2%, and an F1-Score of 85.5%, with a 16.2% increase in accuracy from the baseline. Thus, this system effectively helps users with culinary recommendations in Bandung, providing good performance based on user preferences.