Mobile Recommendations for Leisure Activities Bo Begole, Victoria Bellotti, Ed H. Chi, Nicolas Ducheneaut, Mark Newman, Kurt Partridge Palo Alto Research Center 3333 Coyote Hill Road, Palo Alto, CA magitti@parc.com ABSTRACT We demonstrate a context-aware mobile system for recommending information about leisure activities (Shopping, Eating, Doing, Seeing, and Reading), codenamed Magitti, which infers the user’s leisure activity from context and patterns of behavior. Magitti filters a database of city-guide-style leisure information to find the most relevant items based on the user’s profile, history, context, and predicted activity. Users can also customize the profile or dynamically adjust the current preferences if they wish to improve the recommendations further. Author Keywords context-aware, mobile recommendation systems. ACM Classification Keywords H5.m. Information interfaces and presentation: Misc. INTRODUCTION We demonstrate a system, codenamed Magitti, which uses context filtering to narrow down the inevitable overload of leisure time offerings in dense urban areas. It can do so without the user having to explicitly define her profile or preferences. The system infers interests and activities from models that are learned over time, based on individual and aggregate user behavior, such as places visited, web browsing, and communications with friends. More details about the motivation and fieldwork that led to the system design can be found in prior work [2]. USER INTERFACE Magitti’s Main Screen (Figure 1) shows a scrollable list of up to 20 recommended items that match the user’s current situation and profile. As the user walks around, the list updates automatically to show items relevant to new locations. Each recommendation is presented in a summary form on the result list, but the user can tap each one to view its Detail Screen on the touch screen (Figure 1). This screen shows the initial text of a summary, a formal review, and user comments, and the user can view the full text of each component on separate screens. The Detail Screen also allows the user to rate the item on a 5-star scale. To locate recommended items on the Main Screen, users can tap the Map tab to see the partial map (Figure 2), which shows the four items currently visible in the list on the map. A second tap slides the map out to full screen. The minimal size and one-handed operation requirements have a clear impact on the UI. As can be seen from Figures 1 and 2, large buttons dominate the screen to enable the user to operate Magitti with a thumb while holding the device in one hand. Our design utilizes marking menus on touch-screens to operate the interface, as shown in the right side of Figure 2. The user taps on an item and holds for 400ms to view the menu; then drags her thumb from the center X and releases over the menu item. As the user learns commands and their gestures, she can simply sweep her thumb in that direction without waiting for the menu to appear. Over time, she learns to operate the device without the menus, although they are available whenever needed. Menu buttons at the bottom of the Main Screen allow the user to adjust the recommendation list if needed. By default, the system is in “Any” mode, meaning it will offer recommendations based on its predictions about the likelihood of each of five classes of user activity; Eat, Buy, See, Do, or Read (these will be explained later). But the user can ask to see recommendations from just one category Appeared as a demonstration at the International Workshop on Recommendation and Collaboration at IUI 2008, Jan 13, 2008 Copyright is retained by the authors. Figure 1. Magitti’s Main Screen (left) and Detail Screen (right).