Searching and mining in enriched geo-spatial data
Klaus Arthur Schmid · Elektronische Hochschulschriften der LMU München (Ludwig-Maximilians-Universität München) · 2016
The emergence of new data collection mechanisms in geo-spatial applications paired with a heightened tendency of users to volunteer information provides an ever-increasing flow of data of high volume, complex nature, and often associated with inherent uncertainty. Such mechanisms include crowdsourcing, automated knowledge inference, tracking, and social media data repositories. Such data bearing additional information from multiple sources like probability distributions, text or numerical attributes, social context, or multimedia content can be called multi-enriched. Searching and mining this abundance of information holds many challenges, if all of the data's potential is to be released. This thesis addresses several major issues arising in that field, namely path queries using multi-enriched data, trend mining in social media data, and handling uncertainty in geo-spatial data. In all cases, the developed methods have made significant contributions and have appeared in or were accepted into various renowned international peer-reviewed venues. A common use of geo-spatial data is path queries in road networks where traditional methods optimise results based on absolute and ofttimes singular metrics, i.e., finding the shortest paths based on distance or the best trade-off between distance and travel time. Integrating additional aspects like qualitative or social data by enriching the data model with knowledge derived from sources as mentioned above allows for queries that can be issued to fit a broader scope of needs or preferences. This thesis presents two implementations of incorporating multi-enriched data into road networks. In one case, a range of qualitative data sources is evaluated to gain knowledge about user preferences which is subsequently matched with locations represented in a road network and integrated into its components. Several methods are presented for highly customisable path queries that incorporate a wide spectrum of data. In a second case, a framework is described for resource distribution with reappearance in road networks to serve one or more clients, resulting in paths that provide maximum gain based on a probabilistic evaluation of available resources. Applications for this include finding parking spots. Social media trends are an emerging research area giving insight in user sentiment and important topics. Such trends consist of bursts of messages concerning a certain topic within a time frame, significantly deviating from the average appearance frequency of the same topic. By investigating the dissemination of such trends in space and time, this thesis presents methods to classify trend archetypes to predict future dissemination of a trend. Processing and querying uncertain data is particularly demanding given the additional knowledge required to yield results with probabilistic guarantees. Since such knowledge is not always available and queries are not easily scaled to larger datasets due to the #P-complete nature of the problem, many existing approaches reduce the data to a deterministic representation of its underlying model to eliminate uncertainty. However, data uncertainty can also provide valuable insight into the nature of the data that cannot be represented in a deterministic manner. This thesis presents techniques for clustering uncertain data as well as query processing, that take the additional information from uncertainty models into account while preserving scalability using a sampling-based approach, while previous approaches could only provide one of the two. The given solutions enable the application of various existing clustering techniques or query types to a framework that manages the uncertainty.