Extracting Co-locator context Swaroop Kalasapur, Henry Song and Doreen Cheng Abstract- Having reliable context sources is very important for context-aware applications and the devices around a user can be a useful context for many applications. While the importance of 'devices around' as a context has been highlighted many times, to the best of our knowledge, there is no systematic mechanism reported to identify and capture this context. In this paper we report our mechanism for extracting patterns that we call co-locators from the surrounding Bluetooth devices. The co-locator context can then be used as an indication about the user's surroundings. The context can be directly associated with user behavior in the surroundings, which in turn can be used for recommendation purposes. The context can also be used as a supplemental location indicator in the absence of other location mechanisms such as GPS. We then show that our experimental results support the value of using extracted co-locator patterns as good supplemental location indicators for several applications. Index Terms-Location Context, Co-locators, Semantic Location, Bluetooth Location. I. INTRODUCTION T HE present day personal devices have become powerful platforms capable of performing considerable amount of computation and communication. Context aware applications are fast becoming a reality on such devices. To enable such applications, availability of reliable context information is necessary. One of the most precious contexts that have proven to be extremely valuable is the location of the user. While there are a number of existing mechanisms for location identification, each providing information at different levels of granularity, each of the existing mechanisms has its own practical impediments. For example, GPS based solutions do not work indoors and amidst high-rise buildings. WiFi based solutions require databases that are kept up-to-date. Since the aim of existing mechanisms is to provide location information to the masses, they do not differentiate their results for individual users. The richness in the association between user actions and Manuscript received March 27, 2009. Swaroop Kalasapur is a Senior Engineer at Samsung Electronics R&D Center in San Jose, CA, USA (e-mail: s.kalasapur@sisa.samsung.com) Henry Song is a Staff Engineer at Samsung Electronics R&D Center in San Jose, CA, USA (e-mail: hsong@sisa.samsung.com) Doreen Cheng is a Principal Engineer at Samsung Electronics R&D Center in San Jose, CA, USA (e-mail: doreen.c@sisa.samsung.com) Digital Object Identifier: 10.4108/ICST.MOBIQUITOUS2009.6857 http://dx.doi.org/10.4108/ICST.MOBIQUITOUS2009.6857 their location is the Holy Grail which Location Based Services are going after. In many instances, the user behavior is also dependant on their surroundings, even within the same location. For example, the user might be willing to take a phone call when they are in their office, but might not be as receptive to the phone call if they are in a meeting. By using just the location of the user, it is quite difficult to distinguish between the above scenarios. It is possible however, to distinguish the above based on the presence of other devices in the user's vicinity. For example, if there is good evidence that the user does not take any phone calls when surrounded by a particular set of devices, the decision can be automated on behalf of the user. Many of the users spend a significant amount of time in familiar surroundings, such as their work place or home. Under such familiar surroundings, the user's device will witness a similar set of devices repeatedly, over a period of time. With many of the personal devices being equipped with Bluetooth interface, it is common to see many Bluetooth devices in our everyday surroundings. These Bluetooth devices can be discovered by another Bluetooth capable device on demand. By extracting patterns within the discovered Bluetooth devices, it is possible to use such extracted patterns as a reliable context. The Familiar Stranger [11] project from Intel's lab at UC Berkeley has studied this phenomenon of people encountering same people repeatedly from a social sciences perspective. We attempt to take advantage of the presence of such familiar surroundings to model the familiar devices as a context. In this paper, we focus on discovering such patterns among the devices around the user. We call this context as 'co- locators'. The co-locator context can be used as a supplemental location identifier, in the absence of a reliable positioning mechanism such as GPS. Also, the co-locator context can be directly consumed by context aware applications. If the users are willing to associate the discovered co-locator patterns with meaningful names, such as 'in a meeting', 'in the cafeteria', 'with design team', etc, such names can be directly consumed by a wide range of applications such as social networks that require presence information, auto-tagging applications for photographs, life streaming applications, etc. We believe that our approach can be very valuable even in the presence of other location identification mechanisms. We have constructed a hierarchical location store that keeps track of the user locations based on identified granularity. At the highest level of the store, the zip-code based location as identified from the base station associated with the device is stored. At the next level, the street address retrieved from