The Web 2.0 has changed the way that people communicate with each other. Web 2.0 platforms allow the creation and the exchange of user-generated contents (UGCs) on the Internet. The UGCs contain explicit and implicit geographic information. The quantity of geographic information continuously increases as number of Internet users expands. Since Volunteered Geographic Information (VGI) and social media have been considered as crowdsourcing data resources to harvest geographic information from the web, this thesis investigates two types of crowdsourcing data: VGI and social media data in three topics (i.e., researches). The first research addresses the bottlenecks of current traffic data collections and proposes a solution that utilizes VGI to solve the problems. To collect and use VGI to solve traffic problems, we have reviewed the existing transportation-related mobile applications and designed and developed a complete front-end and back-end GIS for transportation. This research may be beneficial for the design of modern transportation system. The second research aims to explore and extract local information from social media. To achieve the goal, this research proposes an algorithm to discover transient local communities (TLCs) in time evolving social media. The proposed model integrates a damping function to filter irrelevant connections over time, a graph-clustering algorithm to identify communities, and a geo-location proximity algorithm to gather geographically close users. The algorithm has been evaluated by using a Twitter dataset and its performance has been examined in terms of the ability to extract local information. The third research aims to discover communities of interest (CoIs) in local social media. This research may benefit the local information broadcasting and local marketing. In order to discover CoIs from the noisy social media data collected, this research employs natural language processing to clean the short messages (e.g., tweets) and proposes different ways to build sociograms. This research also proposes a graph clustering algorithm that enhances the fast-greedy optimization of modularity (FGM) with text similarity measures to eliminate the noises created by meaningless connections. The algorithm has been evaluated by using a Twitter dataset, and its performance has been examined in terms of the numbers of CoIs discovered under different conditions. Source: http://theses.ucalgary.ca/handle/11023/571