Integrating time forgetting mechanisms into topic- based user interest profiling Xiaoyu Tang, Yue Xu, Shlomo Geva School of Electrical Engineering and Computer Science, Queensland University of Technology, Brisbane, Australia e-mail: xiaoyu.tang@student.qut.edu.au, yue.xu@qut.edu.au, s.geva@qut.edu.au Abstract—The rapid development of the World Wide Web has created massive information leading to the information overload problem. Under this circumstance, personalization techniques have been brought out to help users in finding content which meet their personalized interests or needs out of massively increasing information. User profiling techniques have performed the core role in this research. Traditionally, most user profiling techniques create user representations in a static way. However, changes of user interests may occur with time in real world applications. In this research we develop algorithms for mining user interests by integrating time decay mechanisms into topic-based user interest profiling. Time forgetting functions will be integrated into the calculation of topic interest measurements on in-depth level. The experimental study shows that, considering temporal effects of user interests by integrating time forgetting mechanisms shows better performance of recommendation. Keywords—recommender systems; user profiling; time forgetting; temporal dynamics I. INTRODUCTION The rapidly developing Internet provides enormous amount of data to users leading to the information overload problem, which makes users find it hard to obtain useful content of their interests. Personalization is the ability to provide content and services tailored to individuals based on knowledge about their preferences and behaviors [1]. To assist people in seeking the information meeting their interests or needs, recommender systems have been designed to support decision making process by providing personalized contents, services and information items to potential consumers [2]. Through making suggestions regarding which information is most relevant to people, recommender systems are one of the effective tools to deal with information overload issue and play an important role in people’s life. As the core component of most recommender systems, user profiles describe users’ interests, preferences or needs. Most existing user profiling techniques generate user profiles without considering when this data was generated. However, changes of user interests may occur with time. Some old interests may slowly fade out of users’ preferences, while new interests may arise or user interest drifts occur. Under this background, this paper explores the influence of temporal information in user profiling techniques. In order to achieve this target, we focus on using time forgetting mechanisms in topic-based user profiling techniques on an in- depth level. We have designed time decay mechanisms for an updated topic-based user profiling and collaborative filtering recommender system. The remainder of the paper is organized as follows. Section II reviews some recent works regarding recommender systems and user profiling. Section III presents the proposed user profiling algorithm. Section IV describes our experimental studies and the last section concludes the paper and points out possible directions for future work. II. BACKGROUND In this section, we review recent relevant works regarding collaborative recommendation and user profiling techniques. A. Recommender Systems Based on the recommendation approaches, recommender systems can be classified into three categories: collaborative recommender systems, content-based recommender systems and hybrid recommender systems [3]. Typically, collaborative recommender systems like the Grundy system, Grouplens system [4], Ringo system [5] etc., utilize the ratings of users for items to calculate the similarities between users and between items. Content-based recommenders predict values of utility function for each item to decide whether or not to put it into recommendation or to determine its order in recommendation for each user [3]. Hybrid recommender techniques integrate the two techniques to abate or solve some key problems in collaborative or content based recommenders like sparsity, cold start, serendipity and new users and so on. B. User Interest Profiling User profiles play a significant role in recommender systems [6]. User profiling is the process of acquiring and maintaining the knowledge related to the interests or needs of a specific user. There are various methods for creating user profiles [7, 8]. According to representation methods, user profiles can be classified as rating-based, keyword-based, graph-based, taxonomy-based and rule-based user profiles, etc. The most common type of user profiles is the keyword-based user model [2]. As the popularity of the social tagging systems increases, the tags which are created by users subjectively become useful information for improving recommendation [9- 12]. In Huizhi Liang’s work [13, 14], they try to eliminate the noises in tags, and propose to use the multiple relations among 2013 IEEE/WIC/ACM International Conferences on Web Intelligence (WI) and Intelligent Agent Technology (IAT) 978-1-4799-2902-3/13 $31.00 © 2013 IEEE DOI 10.1109/WI-IAT.2013.132 1