Sistem Rekomendasi Hybrid untuk Pemesanan Hidangan Berdasarkan Karakteristik dan Rating Hidangan

Abstract

The method that was often applied in recommender systems was content-based filtering or collaborative filtering which had several drawbacks if applied singly so that its accuracy was not too high. This study intended to solve the drawbacks of both by combining these two methods into a hybrid method. Apriori algorithm was used to provided recommendations based on dishes’s category and price range in customer order history or wishlist. The similarity between dishes was calculated using adjusted-cosine similarity algorithm while customer’s rating for dishes prediction was calculated using weighted sum algorithm. The values generated by these two methods were then averaged for recommendation process. The proposed hybrid recommender system successfully combines content-based with collaborative filtering methods where its precision and recall values when measured by confusion matrix are 80.73% and 76.52%. By considering the characteristics of dishes that have been ordered by customer, the recommender system is able to recommend new dishes or dishes that have not been ordered as long as their characteristics are similar to the dishes the customer has ordered.