Performance of Known Ratings-Based Multi-Criteria Recommender System for Housing Selection

Housing developments are increasingly massive, and the lack of available information makes prospective customers experience difficulties in choosing a housing. These conditions resulted in the need for a recommendation system to assist consumers in choosing a place to live. In this study, we propose using the Multi-Criteria Recommender System (MCRS) to produce the most recommended housing selection recommendations in a case study of five housing complexes in Malang Raya. The system generates recommendations based on known user rating of 14 criteria and an overall rating (R0) stored in the database. In the experimental stage, the MCRS system in this study used four different methods: cosine, adjust cosine, Pearson correlation, and spearman rank-order correlation coefficient. The test results show that the recommendation system with each similarity method can produce housing recommendations by displaying the three most relevant housing recommendations to the user. Next, we use a confusion matrix to analyze the accuracy of the recommendations generated by the four similarity methods. The results of the confusion matrix calculation show that the average accuracy value for cosine-based similarity is 63.8%, the adjusted-cosine similarity is 70.4%, the Pearson correlation is 88.7%, and the Spearman rank-order correlation coefficient is 75.57%.