아이템 특성을 기반으로 한 상품의 추천 기법

Abstract

Now users have to pour more endeavor to find information they want for the drastic increase of information and the expansion of various information communication devices. Companies are using a recommendation system which recommends an item to the user in order to settle the problem of such information overload and secure more users. Collaborative filtering is the most widely used method among the recommendation systems and is based on user or item evaluation. However, collaborative filtering exhibits problems such as sparsity, scalability, and cold start which reduce the accuracy of recommendation. To get rid of the problems of collaborative filtering, this dissertation suggests a method to use item characteristics as the weighted value. The proposed method consists of steps for item characteristic extraction and item recommendation. The item characteristic extraction step classifies items by genres and uses only the data of which rating is 4 or higher to analyze the user information over the item. The item recommendation step computes similarity by making use of item characteristic. After computing similarity, only the data which satisfies the condition is designated as nearest neighbor. It computes the predicted value of the targeted item using the rating of k-nearest neighbors and designates the rating of the unrated cell using only the predicted value of Top-N. When it recommends the item to the user, only the items of which the predicted value is 3 or higher are recommended. The technique suggested through this method not only reduces sparsity problem but enhances accuracy as well. And when a new item and user are registered, it is possible to conduct fast classification and recommendation based on analyzed information and also reduce the cold start problem. The experiment result using MovieLens data set showed that the suggested technique has been more enhanced the accuracy, appropriacy, and efficiency than item-based and genre-based method.