Penerapan Data Mining pada Sistem Rekomendasi Destinasi Wisata Berbasis Term Frequency-Inverse Document Frequency dan Vector Space Model

Abstract

Digital tourism transitions cause information overload, complicating tourist decisions with unstructured data. Collaborative filtering systems also encounter cold-start issues for new destinations that lack sufficient popularity metrics. To address this, this research develops a recommendation system for Bali tourist destinations using content-based filtering. The method integrates Term Frequency-Inverse Document Frequency (TF-IDF) to extract text weights and the Vector Space Model (VSM) via cosine similarity to measure user query and destination relevance. Built on a Model-View-Controller (MVC) architecture, the system dynamically links PHP Laravel and Python Flask via a REST API. Text preprocessing involves case folding, tokenizing, filtering, and stemming using the Sastrawi library. Evaluation results without Top-N cutoff restrictions show an average recall rate of 90.20%, an average accuracy of 88.18%, and a mean average precision (MAP) of 72.82%. Although the absence of cutoff limits suppressed average precision to 34.43%, setting a 0.05 relevance threshold efficiently filtered out irrelevant noise. The system comprehensively recommends potential semantic destinations without relying on popularity, while maintaining high-ranking quality.