Daily life is full of location-related decisions: where to go on vacation, which house to buy, which job to apply for, etc. These decisions are not only influenced by the characteristics of this holiday home, house, or company, but also by the region it is located in. What is the distance to the beach, the nearest train station, or the schools and kindergartens? In this thesis, that is inspired by the case of an online holiday home broker, we introduce the geosocial recommender system GeoSoRS: a system that supports people in their decision-making process for location-bound objects, such as holiday homes or real estate, using public data only, to keep the threshold for using GeoSoRS to a minimum. Current location-based recommender systems are focused on the recommendation of a single point-of-interest (POI), based on the characteristics of that POI only. In this thesis, we combine this information about a region in its geoprofile: a description of those characteristics of a region that are relevant to the decision to be made. We try to find the match between users and location-bound objects through a user profile and a geoprofile respectively, and look at the shared interests that users have and regions can satisfy. We find this match by answering four main research questions. First of all, we present a software architecture suitable for the combination of web content, social media data, user-generated content (UGC) and several other sources that provide useful information for recommendation selection. In the second, and largest, part of the thesis, we detect which places are visited by which people through a three-step process: (1) POI collection from the web, (2) estimation of their shape and size, called their polygon-of-interest, using public data only, and (3) matching such two-dimensional polygons-of-interest with user trajectories to detect true visits. Thirdly, we find a way to assess the quality of UGC, based on trajectory characteristics of their creators. We introduce a method for human pattern recognition in trajectory data, and show how the outcome of this pattern detection can be used for the creation of UGC quality prediction models. With the fourth and final research question, we focus on knowledge-based recommendations, combining user interests with the interests that can be satisfied at certain locations. We combine social media data, public generic knowledge-bases and tagged item sets to recommend items from multiple domains, among which holiday homes.
|Award date||10 Dec 2015|
|Place of Publication||Enschede, The Netherlands|
|Publication status||Published - 10 Dec 2015|