머신러닝 알고리즘을 결합한 협업필터링 개인화 추천시스템 응용연구

This study proposes a new personalized recommendation system by combining collaborative filtering model with machine learning algorithms from the Random Forest and K-means clustering methods. It presents a solution to the traditional issues of collaborative filtering such as data sparsity, cold start problem, increased computational complexity, and difficulties in applying big data. We reduce the dimension of the data-matrix greatly, by first utilizing methods from the Random Forest model to predict user preferences for product categories, and in turn using these predictions as a basis for forming groups of users with similar preferences through k-means clustering. We demonstrate the practical applicability and effectiveness of the recommendation system with real-world big data from an actual e-commerce company. With our proposed method we were able to create a user specific recommendation system based on the entire dataset of over 7 million transactions with 214,638 unique items, all the while greatly improving computational efficiency by only using 221MB of memory. The results of our study suggest that our proposed collaborative filtering approach combined with machine learning algorithms offers a different perspective in the development of collaborative filtering based recommendation systems, providing new groundwork for future research.