1. Scaling the Wall: Experiences adapting a Semantic Web application to utilize social networks on mobile devices Evan W. Patton 1 ( pattoe@cs.rpi.edu ) & Deborah L. McGuinness 1 (dlm@cs.rpi.edu) pattoe@cs.rpi.edudlm@cs.rpi.edu ( 1 Rensselaer Polytechnic Institute 110 8 th St., Troy, NY, 12180 United States) Glossary: RPI – Rensselaer Polytechnic Institute TWC – Tetherless World Constellation at Rensselaer Polytechnic Institute OWL – Web Ontology Language for describing concepts and relationships between them RDF – Resource Description Framework, a layer for encoding information as subject- predicate-object statements (also known as triples). JSON – JavaScript Object Notation format used for representing objects in web browsers Acknowledgments: Thanks to our many friends and colleagues for their comments on usability and design issues that have contributed to the evolution of the Wine Agent. Sponsors: National Science Foundation Problem. Social interaction between individuals on mobile devices is becoming more and more pervasive and acceptable. The Semantic Web community has taken notice and new applications are coming out that enable social interaction over semantic technologies. While applications aimed for mobile devices often take into consideration the many needs of the users of these devices, existing application built before the advent of the mobile device as a general purpose computing device require more resources. They also might not be designed to take advantage of innovative data sources that mobile devices provide. Ontologies provide flexibility and extensibility, both to developers and users. The Wine Agent provides a rich vocabulary that individuals can use to describe instances in the real world (see Fig. 1). The system uses logic over the ontology to control what information the user can enter in order to prevent inconsistencies. For example, the user cannot claim that an object is both a food and a wine. Since the ontologies powering the Wine Agent are modeled using the Web Ontology Language, other communities on the web can provide supplemental files that extend the concepts in the Wine Agent, providing an opportunity for external entities to customize the user experience. For example, restaurants can extend the ontology to support local dishes or wine varietals. Ontology Figure 1. Users can use the Wine Agent’s vocabulary to describe a rich set of instances. The wine described here is a medium-bodied Cabernet from Napa Valley. Wine is a very social field, so this mobile version of the Wine Agent supports interaction with a user’s Facebook and Twitter communities. Users can access event information from their Facebook profile, and opt-in to share personal preferences with the event creator, provided that the creator also uses the Wine Agent. When the agent publishes content to Facebook, it includes a JSON-encoded representation [3] of the description, so other Semantic Web applications interacting with Facebook can retrieve and interact with the underlying semantic content. Friend can also share links to semantic content. When accessed via Mobile Safari, these links will launch the Wine Agent application and direct the user to the appropriate resource. Figure 2. A Facebook event with sharing options (left) and a sample Twitter message where the user can insert links to content such as the current wine, dish, or restaurant. Social Networking Future Work The Wine Agent has evolved as the web has evolved. This work has shown how to take an existing Semantic Web application not designed for mobile devices and has adapted that application to provide a socially accessible, semantically powered tool for mobile device users. Lastly, we have begun to address the issue of power management in Semantic Web applications natively running on mobile devices. Geolocation and Context Conclusion The Wine Agent continues to be a test bed for combining many different technologies together using Semantic Web. As part of this work, more investigation is being done into parallelizing OWL reasoning across mobile devices to enable group problem solving based on many different individual constraints. More work needs to be done on the reasoner so the agent can provide explanations of recommendations to users. References [1] Hsu, E., & McGuinness, D.L. (2003). KSL Wine Agent: Semantic Web Testbed Application. In: Proceedings of the 16 th International Workshop on Description Logics. [2] Michaelis, J., Ding, L., & McGuinness, D.L. (2008). The TW Wine Agent: A Social Semantic Web Demo. In: Poster and Demo session of the 7 th International Semantic Web Conference [3] Alexander, K. (2008). RDF/JSON: A Specification for Serializing RDF in JSON. In: Proceedings of the 4 th Workshop for Scripting on the Semantic Web. Today’s mobile device provide developers with the opportunity to obtain a user’s location in order to use that information to provide a better experience that can be tailored to a user’s location. The Tetherless World’s Mobile Wine Agent takes advantage of this information in order to locate nearby restaurants with RDF-encoded menus that the Wine Agent can understand. For demonstration purposes, we provide a number of such menus. The agent presents a map view (see right) that shows the user’s current location and the available restaurants. If the phone’s location system returns a point within error of a restaurant’s location, it assumes the individual is present at that restaurant loads its menu automatically. At first, general information about a restaurant is presented, and tapping items will provide the user with opportunities to get directions, visit websites, or share content with others. Users can also load the RDF- based menus into the agent to view their contents. Since all items are described using the Wine Agent’s ontology, they can then be submitted to the system to obtain recommendations, rather than the user having to enter a custom object. Power Consumption. The biggest difference between developing semantic applications for the desktop and semantic applications for mobile devices is power consumption. Users have a limited amount of power available for tasks, so care must be taken to perform heavily computational processes elsewhere. Running Time. One way to increase power consumption is by reducing the running time of certain operations. Additionally, steps can be taken to amortize the power usage by performing reasoning up front to provide better response time to the user. Available Memory. Today’s desktop machines often come with at least 2 Gigabytes of Random Access Memory (RAM). However, mobile devices are often limited on the amount of memory available to applications, either due to power requirements or physical space. Therefore, reasoners for mobile devices need to be aware of when they are running low on memory and clear intermediate caches or flush them