In Search of Knowledge about Mobile Users

by Ee-Peng Lim, Yida Wang, Kok-Leong Ong and San-Yih Hwang

At the Center for Advanced Information Systems at Nanyang Technological University, Singapore, research is carried out on using knowledge discovery techniques such as frequent pattern mining, classification, and clustering to create new mobile applications or to enhance existing mobile applications.

Mobile phones and other mobile devices are fast becoming indispensible in our modern society. According to a recent survey by Frank N. Magid Associates and Upoc.com, 59 percent of Americans age 12 and over (about 140 millions of them) own mobile phones, and that almost a quarter of non-owners plan to buy a mobile phone in the near future. The sales of mobile phones worldwide was 385 millions in 2001 and it has been predicted to reach 675 million in 2006. In tandem to this growth trend, we also witness the emergence of many new applications and businesses that exploit mobile phone technologies in different ways. Before other wearable computing gadgets become more feasible and popular, mobile phones are likely to remain as the more dominant wearable devices in the coming years.

Mobile phones, unlike computers connected to wired networks, are highly personalizable. While it is common for a user to own a few mobile phones, it is very unlikely for different users to share a mobile phone. Also unlike other personalized accessories such as watches, walkmans, etc., many of the mobile phones are trackable. They are trackable because they have to maintain regular contacts with the mobile telecommunication networks in order to receive and make calls. With these trackability and personalization features, one can conceive many unique and interesting mobile applications for end users.

At the Center for Advanced Information Systems, we conduct extensive research on using knowledge discovery techniques such as frequent pattern mining, classification, and clustering to create new mobile applications or to enhance existing mobile applications. Examples of such mobile applications include e-commerce systems, databases, email systems, search engines and web browsers. We investigate both the functionalities and operational efficiencies of mobile applications, and determine the kinds of knowledge required for enhancing them. We also develop algorithms for discovering knowledge from mobile user data and evaluate their performance. We further study how the discovered knowledge and mining algorithms can be integrated with the operational systems turning the knowledge into actions.

In the e-commerce application domain, we envisage the importance of using mobile user movement data to derive knowledge about mobile users. As users’ purchase behaviors are often highly correlated to their group affiliations, knowing the latter well will allow e-commerce vendors to develop group-specific pricing models and marketing strategies to better meet the buying needs of the user groups. While there have been several existing methods developed to address the problem of discovering user groups, they are usually based on transaction histories and user profiles, and the accuracies may not be satisfactory.

In our research, we introduce a new approach to mine user groups, known as group pattern mining. In group pattern mining, we determine the customer grouping information based on the spatio-temporal distances among the customers. We assume that the user movement data are first collected by logging location data emitted from the mobile phones and similar devices. Mobile users that are always close to one another for significant amount of time are modeled as a group pattern represented by 4-tuple <G, max_dist, min_dur> where G denotes the set of mobile users, max_dist denotes the maximum distance between any pair of users for them to be considered close, and min_dur denotes the minimum duration in which the users in G are close to one another. For example, for {Tom, Mary} to be a valid pattern, Tom and Mary must be less than max_dist apart for at least min_dur continuous time period.

Compared to traditional frequent pattern mining, group pattern mining incurs much more computation and storage. To discover group patterns from a set of mobile users, we have proposed a few efficient algorithms that are based on frequent pattern mining. As part of this research, we also investigate new data summarization techniques and data structures to reduce the computational and storage overheads. In brief, our research framework for mobile data mining is divided into three core areas, which we shall highlight some of the issues that should be further investigated. They are:

• Infrastructure for mobile data mining - we explore the design issues involving a warehouse for mobile data. This may include the algorithms for efficient aggregation and transformation of mobile data. The challenge in this case is to deal with the heterogeneous data format from different mobile devices at different locations with different bandwidth and computing resources.
• Algorithms for mobile data mining - once data is available, the next challenge is to make sense of the data. Hence, algorithms are needed to find knowledge to improve the efficiency of mobile applications/queries, and to enhance the user experiences of the phone.
• Incorporating mobile mining results into operational systems - knowledge obtained from data mining must be integrated with the operational systems. The challenge is to develop algorithms that evaluate which are the 'actionable' data mining results, and then apply them in a timely fashion to ensure the effectiveness of the mobile data mining system.

Mobile data mining, as a very new area of research, has created a wide range of opportunities for researchers, engineers and developers to create new interesting applications for both the end users and businesses. We believe that our group pattern mining work is only one of these many examples. As many more knowledge about mobile users can be mined in the near future, we therefore expect the upcoming applications and systems to be able to adapt more seamlessly into our daily lives.

Further Reading:

• Upoc.com. “More Mobile Owners Turning to Text Messaging”, http://www.upoc.com/corp/news/news-emarketer.html, Feb. 2003.
• Reed Electronics Research, “RER- The Mobile Phone Industry - A Strategic Overview,” Oct. 2002.
• U. Varshney, R. Vetter and R. Kalakota, “Mobile commerce: A New Frontier,” IEEE Computer: Special Issue on E-commerce, Oct. 2000.
• Ee-Peng Lim, Keng Siau, Advances in Mobile Commerce Technologies, Idea Group Publishing, Jan. 2003.
• MIT Media Lab. The Context Aware Cell Phone Project. (under the MIThril wearable computing initiative). http://www.media.mit.edu/wearables/mithril/phone.html
• Yida Wang, Ee-Peng Lim, San-Yih Hwang, “On Mining Group Patterns of Mobile Users,” 14th International Conference on Database and Expert Systems Applications (DEXA2003), Prague, Czech Republic, September, 2003.

Please contact:
Ee-Peng Lim, Yida Wang and Kok-Leong Ong Nanyang Technological University, Singapore
E-mail: aseplim@ntu.edu.sg, ongkl@pmail.ntu.edu.sg, wyd66@pmail.ntu.edu.sg

San-Yih Hwang, National Sun Yat-Sen University, Taiwan
E-mail: syhwang@mis.nsysu.edu.tw