Location-Based Services (LBS) use portable devices to deliver information to users based on their geographical position. Growth in this area is supported by the success of the mobile web. In the United States alone there have been 20 million unique users of the mobile internet over the past two years. Mapping is among the top ten activities conducted using the internet on a mobile device. The number of mobile broadband subscriptions worldwide has surpassed fixed ones, highlighting the potential of LBS, which is expected to increase to a USD13 billion business by 2013. This makes it imperative that information overload is resolved quickly.

Information Overload
Web search engines represent a good example of information overload in the non-spatial domain. A simple text-based search can return millions of results, impossible for a human user to process. Similarly, in terms of map data, a single search can return a multitude of spatial data points, as demonstrated in Figure 1. This makes it difficult for users to find information relevant to them. The problem is intensified when dealing with LBS, because mobile devices have a smaller screen and limited processing power. There is a real need for the introduction of intelligent techniques to adapt map data so that a relevant subset of that which is available can be presented to the user, as seen in Figure 2.

Research Project
In the Spatial Information Systems Group at UCD we have developed techniques for displaying and rendering map data using web-based technologies. Cross-platform operation was traditionally difficult, requiring individual implementation for each target device. Utilising the power of open-source web-based technologies these issues can be overcome, allowing any web-enabled device, such as that shown in Figure 3, to access the LBS. In addition to open-source technologies we use data from the OpenStreetMap project 1 to provide rich and up-to-date maps:
a suitable platform for developing and testing algorithms for inferring user interests and personalising map data to resolve the issues of information overload.

Monitoring Interaction
The first step towards providing appropriate adapted map data involves ascertaining the interests of the user. Once user interests are known, personalised and targeted data can be provided. Determining interests can be achieved using techniques that are either explicit, by asking user, or implicit. While the former can provide an instant view of the user's current interests, it may be distracting and distorted by subjectivity. Often it is more advantageous to obtain this information using unobtrusive techniques which do not distract the user from the task in hand. Part of our project involves monitoring user interactions with objects displayed on the map, underlying geographical data and interactions with the available toolset. The user's context, including their geographical position and the device being used, are also monitored and recorded. While existing approaches in this area mainly monitor a user's usual behaviour (for example, which layers on a map are never used), our approach records interaction with individual objects and features on the map in order to generate very precise insight into user interests.

Generating Profiles
We are studying ways to use implicitly recorded user data to accurately infer user interests. This involves designing efficient algorithms which can handle the complexity and volume of recorded interaction data and combine it with contextual information. The premise behind this approach is that the more often a user interacts with a map object, the more interested they are in it. Similarly, the more often a user is in proximity to a map object (in the real world), the more interested they are in that object. A balance must be found between the two to ensure generation of the most accurate profile.

Personalisation
Several approaches may be adopted for purposes of map personalisation. For example, details of map features deemed of no interest to the user can be removed or completely excluded from the map while map objects relevant to the user can be highlighted. Another part of our project concerns visualising user map inter­action, as depicted in Figure 4. Such a visual analysis tool provides a mechanism for observing behavioural patterns which can then be used to identify key user actions and activities. This contributes to an understanding of their interests; knowledge that supports design of algorithms for determining these.

The visualisation tool also helps research in the field of group profiling by enabling identification of clusters of similar users. Group profiling and collaborative filtering are already known as components of the non-spatial web, where they are used for recommending items based on others making similar purchases or having similar browsing habits. In LBS, similarity between individuals can be determined using profiles or interactional behaviour patterns.
This approach could help improve the quality of recommendations presented to new users whose individual profile may be immature.

Concluding Remarks
By using personalisation techniques for LBS via a web-based platform, we hope to improve end-user experience. As the need to search through an abundance of information for relevant material is removed, time spent actually using services will be rendered more productive.

Acknowledgements
Our thanks to Science Foundation Ireland, under the National Development Plan, for the Strategic Research Cluster grant (07/SRC/I1168) that provided funding for this research.

Strategic Research
The research carried out by the Spatial Information Systems Group at UCD is part of a wider research initiative called Strategic Research Cluster on Advanced Geotechnologies (StratAG). Led by Professor Stewart Fotheringham of the National Centre for Geocomputation (NCG), UK, this represents collaboration between University College Dublin, the National University of Ireland, Maynooth, Dublin Institute of Technology, Trinity College Dublin and the NCG. Funded by Science Foundation Ireland, researchers are investigating several key areas of the Geographic Information System (GIS) domain, including data acquisition, algorithms, visualisation, and LBS.