Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE credits
Advances in mobile communication, computing and positioning technologies allow the real-time acquisition of continuously changing locations of moving objects, e.g., users carrying location-aware mobile devices. The geocontextual analysis of these locations of users can reveal valuable mobile consumer patterns and enable a number of promising Business Intelligence (BI) services and desirable intelligent Location-Based Services (LBSes). However, as some exact user locations can be extremely sensitive while others can be used to link the users to his or her real-world identity, location traces of users must be anonymized.To facilitate these promising BI services and desirable intelligent LBS in a privacy preserving manner, the present paper proposes an Anonymous Mobile Consumer Analysis Platform (AMCAP) that based on the geocontextual analysis of anonymized location traces, in four phases, derives mobile consumer characteristics of its users in terms of Anomymous Mobile Consumer Profiles (AMCPs) and information about the dynamically changing spatio-temporal distribution of the users that is recorded in a spatio- temporal data warehouse of Anonymous Mobile Consumer Types (AMCTs). In the first phase, based on a number of spatio-temporal criteria the platform probabilistically associates a user’s generalized location, that is represented by an anonymization region with activity types that activity centers in the anonymization region can facilitate. In the second phase, based on the activity type inferred in phase one, the platform, in an online and incremental fashion, summarizes an Anonymous Mobile Consumer Profile (AMCP) for each of the users. In the third phase, the platform periodically groups the users based on their AMCPs into a number Anonymous Mobile Consumer Types (AMCTs). Finally, in the fourth phase, in order to capture the dynamically changing spatio-temporal distribution of the consumers and their characteristics, the platform, in an online and incremental fashion, records the frequency and duration of the visits of different AMCTs to different spatio-temporal regions.The AMCAP is empirically evaluated on the simulated movements of a subset of the population of Copenhagen, Denmark. Experiments regarding the accuracy of the AMCP construction reveal that the proposed activity type inference from anomymized regions is effective and can predict the actual activity of the users with an accuracy of 0.75 and a κ- value of 0.73. AMCPs are clustered into 7 AMCTs. The quality of AMCTs is evaluated by comparing it with Actual Consumer Types (ACTs) which are extracted from Actual Consumer Profiles (ACPs). In both the AMCPs and ACPs, the optimal number of clusters is 7. Meaning thereby that although AMCPs are blurred, they do not loose their inherent property of being clustered into the same of number of types as ACPs. The Adjusted Rand Index (ARI) (a measurement for similarity between two grouping of objects) between the AMCTs and ACTs is 0.3 which is significantly higher than the random assignment where ARI value is -0.0000247. Computational performance evaluations show that a relational DBMS-based prototype implementation of the AMCAP on a single machine can process 50,000 anonymization rectangles in 20 seconds. Based on the assumption that during any given 5-minute interval no more than 10% of the users submit an anonymization region, it is extrapolated that the prototype is capable to perform in real-time the geocontextual analysis (i.e., phases 1, 2, and 4) of anonymized location traces of 7 million users. Finally, to illustrate the utility of the information that is derived by the AMCAP a number of derived BI services are discussed.