PROJECT TITLE :

PGeoTopic: A Distributed Solution for Mining Geographical Topic Models

ABSTRACT:

The mining of geo-tagged documents for topical regions and geographical topics is one application of geographical topic models. These models also have applications in recommendation systems, user mobility modeling, event detection, and other areas. Studies that have already been done concentrate on learning geographical topic models that are effective but ignore the issue of efficiency. However, training geographical topic models is a very costly endeavor; on a collection of documents containing millions of word tokens, it may take several days to train a geographical topic model of a small scale. In this article, we present the first distributed solution, which we refer to as PGeoTopic, for the process of training geographical topic models. The proposed solution includes a number of innovative technical components in order to increase parallelism, decrease the amount of memory that is required, and lower the cost of communication. Experiments have shown that our methodology for mining geographical topic models is scalable with regard to both the size of the model and the amount of data that is being mined on distributed systems.