A Recommender System for Cold-start Items: A Case Study in the Real Estate Industry

The recommender systems provide users with what they prefer and filter unnecessary information. In the fierce marketing environment, it is crucial to recommend items to users in an early stage to keep user’s interests and loyalty. With the fast product renewal, classical recommendation methods such as collaborative filtering cannot handle the cold-start item problem. In many real-world applications, content information of items or users is available and can be used to assist recommendation. Besides, user may interact with the items in different behaviors such as view, click or subscribe. How to use the complex content information and multiple user behaviors are real problems that are not well solved in applications. In this paper, we propose a content-based recommender system to deal with the practical problem. Boosting tree model also added to the system to avoid potential Spam. We applied our developed method to real estate application to recommend new property which just landed into the market to users. Experimental results with three data subsets and three recommendation scenarios demonstrate that the proposed method can outperform the baseline on recommendation accuracy. The results indicate that our method can effectively reduce potential Spam to users, so that user experience will be improved.