Enhancing Data Classification Quality of Volunteered Geographic Information

Abstract

Geographic data is one of the fundamental components of any Geographic Information System (GIS). Nowadays, the utility of GIS becomes part of everyday life activities, such as searching for a destination, planning a trip, looking for weather information, etc. Without a reliable data source, systems will not provide guaranteed services. In the past, geographic data was collected and processed exclusively by experts and professionals. However, the ubiquity of advanced technology results in the evolution of Volunteered Geographic Information (VGI), when the geographic data is collected and produced by the general public. These changes influence the availability of geographic data, when common people can work together to collect geographic data and produce maps. This particular trend is known as collaborative mapping. In collaborative mapping, the general public shares an online platform to collect, manipulate, and update information about geographic features. OpenStreetMap (OSM) is a prominent example of a collaborative mapping project, which aims to produce a free world map editable and accessible by anyone. During the last decade, VGI has expanded based on the power of crowdsourcing. The involvement of the public in data collection raises great concern about the resulting data quality. There exist various perspectives of geographic data quality this dissertation focuses particularly on the quality of data classification (i.e., thematic accuracy). In professional data collection, data is classified based on quantitative and/or qualitative ob- servations. According to a pre-defined classification model, which is usually constructed by experts, data is assigned to appropriate classes. In contrast, in most collaborative mapping projects data classification is mainly based on individualsa cognition. Through online platforms, contributors collect information about geographic features and trans- form their perceptions into classified entities. In VGI projects, the contributors mostly have limited experience in geography and cartography. Therefore, the acquired data may have a questionable classification quality. This dissertation investigates the challenges of data classification in VGI-based mapping projects (i.e., collaborative mapping projects). In particular, it lists the challenges relevant to the evolution of VGI as well as to the characteristics of geographic data. Furthermore, this work proposes a guiding approach to enhance the data classification quality in such projects. The proposed approach is based on the following premises (i) the availability of large amounts of data, which fosters applying machine learning techniques to extract useful knowledge, (ii) utilization of the extracted knowledge to guide contributors to appropriate data classification, (iii) the humanitarian spirit of contributors to provide precise data, when they are supported by a guidance system, and (iv) the power of crowdsourcing in data collection as well as in ensuring the data quality. This cumulative dissertation consists of five peer-reviewed publications in international conference proceedings and international journals. The publications divide the disser- tation into three parts the first part presents a comprehensive literature review about the relevant previous work of VGI quality assurance procedures (Chapter 2), the second part studies the foundations of the approach (Chapters 3-4), and the third part discusses the proposed approach and provides a validation example for implementing the approach (Chapters 5-6). Furthermore, Chapter 1 presents an overview about the research ques- tions and the adapted research methodology, while Chapter 7 concludes the findings and summarizes the contributions. The proposed approach is validated through empirical studies and an implemented web application. The findings reveal the feasibility of the proposed approach. The output shows that applying the proposed approach results in enhanced data classification quality. Furthermore, the research highlights the demands for intuitive data collection and data interpretation approaches adequate to VGI-based mapping projects. An interaction data collection approach is required to guide the contributors toward enhanced data quality, while an intuitive data interpretation approach is needed to derive more precise information from rich VGI resources.