HYBRID FILTERING FOR MOVIE RECOMMENDATION SYSTEM USING DEMOGRAPHIC INFORMATION

• Main Author: Alif Salman Alfarisy, Rizki
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/54186
• Institution: Institut Teknologi Bandung

A recommendation system is a tool to overcome information overload on various domains including movie domain. One of the commonly used recommendation systems today is hybrid filtering recommendation system that can provide fairly accurate recommendations. However, it is still experiencing cold start problem. Research by Laila and Akram (2013) attempts to solve the problem by creating a recommendation system based on demographic information. The research was able to overcome the problem of cold start, but the performance of the recommendation was not really accurate. Meanwhile, research by Gupta and Gadge (2015) that utilizes demographic information to improve the performance of collaborative filtering can provide good results but has not been able to solve all cold start issues. In this final project, a recommendation system is built using hybrid filtering and also utilizing demographic information to solve cold start problem. The system is built using collaborative filtering that uses rating history and demographic information and content-based filtering that are modified to be able to handle user cold start. The two components are then combined using mixed hybridization and switch hybridization. Experiments results showed that a hybrid filtering recommendation system that uses demographic information gave better results than conventional hybrid filtering recommendation systems in cold start situations. The best recommendation results are achieved by adding weight to mixed hybridization. The addition of weights to the proposed recommendation system in user cold start situations increased the average hit rate by 213.3% from 0.1041 to 0.3263 compared to without using weight and the average ARHR increased 42.3 % from 0.0169 to 0.0295. As for the cold start item situation, the average hit rate increased by 340% from 0.0347 to 0.1388 and average ARHR increase by 58.9% from 0.0047 to 0.0122 compared to without using weight.