| dc.description.abstract | For KTV, virtual storefronts and many other industries, the recommendation systems have to be interactive, adaptive and accurate enough since customers make series of decisions quickly. A system slowly adapt to customers need may find customers make all decisions before the system can react. Therefore, an ideal recommendation system for customers who make a set or series of decisions quickly should have following characteristics: interactive, adaptive, accurate enough, bulk recommendations. However, most recommender systems can’t meet all conditions. Because of the lack of interacting with customers, current recommender systems can not adapt to customers in real time. Once, customers can not obtain useful information when making decision and they would never be satisfied.
In this paper, we want to introduce an ideal recommender system applied to KTV server. We propose a new method to produce recommendations based on a context hierarchy for association rules which are discovered from picking historical data. By rolling up and drilling down the context level, we are able to make bulk recommendations. After recommending, we measure the accuracy of suggestion for quickly adaptive to customers. | en_US |