We extract the underlying hidden relations from the given location-based datasets (e.g. GPS data) and we estimate (approximate) the hidden a social network in the data by using a particular data reduction algorithm (i.e., by using coresets).

The goal in this project is to extract and represent the activity summaries of users from underlying data exchanges in a compact way. We develop and analyze a coreset (data reduction) algorithm for minimizing the memory size required to compute the activity summaries of users from location-based datasets. The streaming coreset algorithm approximates the weighted sum of vectors sparsely and we study guaranteed error bounds for the sparse approximation. We evaluate the algorithm on several large data sets.