Abstract. Location is the most essential presence information for mobile users. In this paper, we present an improved time-based clustering technique for extracting significant locations from GPS data stream. This new location extraction mechanism is incorporated with Google Maps for realizing a cooperative place annotation service on mobile instant messenger (MIM). We also design an ontology-based MIM presence model for inferring the location clues of IM buddies, to support context-aware presence management in our MIM system. The GPS-based location extraction algorithm has been implemented on a Smartphone and evaluated using a real-life GPS trace. We show that the proposed clustering algorithm can achieve more accurate location extraction as it considers the time interval of intermittent location revisits. The incorporation of location information with the high-level contexts, such as mobile user's current activity and their social relationship, can achieve more efficient presence management and context-aware communication.

Abstract-This paper presents the design of the BetterLife 2.0 framework, which facilitates implementation of large-scale social intelligence application in cloud environment. We argued that more and more mobile social applications in pervasive computing need to be implemented this way, with a lot of user generated activities in social networking websites. We adopted the Case-based Reasoning technique to provide logical reasoning and outlined design considerations when porting a typical CBR framework jCOLIBRI2 to cloud, using Hadoop's various services (HDFS, HBase). These services allow efficient case base management (e.g. case insertion) and distribution of computational intensive jobs to speed up reasoning process more than 5 times. With the scalability merit of MapReduce, we can improve recommendation service with social network analysis that needs to handle millions of users' social activities.

We

Abstract-In the ever-changing pervasive computing paradigm, applications, especially those running on resource-scarce mobile devices, have to adapt to the runtime environment as the users are roaming around. Various adaptation techniques, relying on dynamic composition of components, have been proposed by a number of researchers. Nevertheless, most existing approaches only support component selection based on predefined rules and strategies. Because of the limitation of pure rule-based approach, context-awareness can not be well supported. In this paper, we propose a software component selection framework for mobile pervasive computing. Our approach adopts the case-based reasoning technique to provide proactive component selection. Context-awareness and personalization are embodied in the reasoning and selection process. As a proof of concept, we developed and evaluated a context-aware personal communicator (CAPC) application using adaptive component selection, with a synthesized execution trace obtained from real-life E-mail softwares ported to CAPC. Our results show that the adaptive component selection can reduce maximum memory consumption by at least 20%, and the context-guided reasoning technique can improve reasoning accuracy by nearly 10% within acceptable reasoning time.

Cost-effective localization for large-scale Geosocial networking service is a challenging issue in urban environment. This paper studies an ad-hoc localization technique which takes advantages of short-range interchanged location information for calibrating the location of mobile users carrying non-GPS mobile phones. We demonstrate by simulation that a small percentage of GPS-enabled mobile phones can greatly enable the localization of other non-GPS pedestrians in the urban environment. Based on the proposed localization technique, we implement a location-aware social networking tool called Mobile Twitter, similar to the microblogging service of Twitter, for fast propagation of social events happening in surroundings. Evaluation shows the our localization algorithm can achieve better accuracy of the location estimation and wider coverage as compared with the Amorphous algorithm and the Monte Carlo Localization (MCL) method. Moreover, we show that the Mobile Twitter implemented on an Android mobile phone is power-efficient in real-life usage scenarios.