PENGGUNAAN METODE CONTENT-BASED FILTERING DALAM REKOMENDASI GAME DENGAN ALGORITMA SUPPORT VECTOR MACHINE

This research addresses the challenge of finding suitable games in the rapidly expanding gaming industry, where many platforms, such as Steam, currently host 15,422 titles and recorded 33.6 million online players in January 2024. The Content-Based Filtering method is proposed, leveraging a game dataset from Kaggle. The methodology involves data preprocessing steps like removing duplicates, performing comma-separated tokenization, and converting text into
numerical vectors using TF-IDF, with data divided into various ratios for testing. Three primary classification algorithms—Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest (RF)—were implemented, trained, and evaluated using accuracy, precision, recall, and F1-score metrics. Testing results show that Content-Based Filtering successfully recommends relevant games based on Genre, Tag, and Category. SVM consistently outperformed other models across several analyses. For the genre column, SVM's accuracy ranged from 98%, precision 98%, recall 97%, and F1-score 97%. For the tag column, SVM's accuracy results were approximately 98%, precision 98%, recall 99%, and F1-score 98%. Even in the category column, where both SVM and RF were consistent, SVM showed an accuracy of 99%, precision 99%, recall 99%, and F1-score 99%. Overall, SVM emerges as the most efficient algorithm for improving the accuracy of game recommendation systems.
Keywords: recommendation system, content-based filtering, support vector machine, random forest, k-nearest neighbor.