개인화된 추천을 위한 하이브리드 협업필터링 알고리즘 및 인터넷 전자상거래에서의 활용

Recommender systems help customers find products they want to buy from a business. Hence, on one hand, recommender systems benefit customers by enabling them to find products they like and, on the other hand, they also help the business by generating more sales. Recommender systems are rapidly becoming a crucial tool in Electronic commerce on the Web because of the increasing volume of customer data available on the Web including the clickstream data. Collaborative filtering algorithms are the most successful and popular technologies used for product recommendation to date and is used in many of the most successful recommender systems on the Web. Recommender systems using collaborative filtering algorithms store the preferences and opinions/evaluations of many users of the system. These opinions may be recorded as ratings by users for items. When an active user would like a recommendation, the system finds users with similar taste and uses their opinions to generate a recommendation. However, we believe that current collaborative filtering algorithms which are being used popularly in many areas are not as effective as we expect. Therefore, we need to improve the performance of recommender systems by enhancing the accuracy of their recommendations. In this paper, we modify the current collaborative filtering algorithm (nearest user neighbor algorithm) to improve the performance of its recommendations and propose a hybrid collaborative filtering algorithm which considers not only the similarity between customers but also the similarity between products (items) purchased. It is shown that our hybrid collaborative filtering algorithm has better performance in the experiment with the MovieLens data set. We also discuss how our algorithm can be applied to the E-commerce on the Web. Particularly, it is suggested to use the Web log data for customer's preference on a product which may provide more information on the customer's preference. We also discuss how demographic data and traditional data mining techniques could be used with our algorithm to reduce some of the problems current collaborative filtering algorithms have and to improve the performance of recommendation.