Katarzyna Kopczewska (p)

Muhammad Usman

This paper develops the Quick Density Clustering (QDC) method which fills the gap in the toolbox of density clustering of spatially geo-located points. It uses a K-means algorithm which is run on two normalized spatial variables: fixed-radius nearest neighbours (NN) and a sum of distances to k nearest neighbours (NN) to find diverse densities of points in 2D. Clusters detected by QDC classify all (x,y) geo-points to high/mid/low-density clusters. QDC uses a standard clustering method on transformed data, unlike many other sophisticated methods that are run on 2D geo-coordinates. It is a quick, efficient, semi-autonomous and big-data tool applicable to static and streaming data. A major parameter in QDC, the number of K clusters to detect, is interpretation-driven, while the other two: the radius for counting NN and the number of NN to sum the distances are of secondary importance and in a minor way impact the outcome. Classification for new points (prediction) is quicker than a typical kNN algorithm by using thresholds of spatial variables. The approach is suitable for tracking human activity as traffic or crowd detection from spatially geo-located mobile data – it finds the high-density points independently of phenomenon intensity and works well with streaming data.

Moritz Schütz (p)

Katarzyna Kopczewska

Due to the increasing availability of web data, as well as the development of new methods for extracting information from this data, the last decade saw a rise of web mining approaches for empirical research in the social sciences. This has provided novel ways of analyzing regional economic activities with varying applications. The aim of our study is the conceptualization of a comprehensive approach for using web content and web structure data of local government institutions. By creating a novel database, we are the first study to propose an approach for the creation of indicators that depict local governance institutions based on web content data. With this, we contribute to progress in empirical research by providing a new way of quantifying and analyzing the role of local governance institutions as facilitators for regional (economic) development. Our approach involves creating an extensive html-database by scraping data from websites of German municipalities and counties. This process resulted in a comprehensive collection of text data, amounting to around 4 million paragraphs from municipal websites and half a million from county websites. The core aim of this research is to delineate regional policy priorities and narratives through this data. We employ ‘GoogleBERT’ for topic modeling and Few-Shot Learning for text classification. The application of GoogleBERT to municipal website data facilitated the identification of approximately 150 topics, distinguishing between general information and specific policy priorities. Additionally, we utilized Few-Shot Learning to classify sustainability-related content on both municipal and county websites. Focusing on the content of the text we find that it is possible to not only identify differing sustainability narratives, but also to identify policy strategies that are implemented to tackle the topic of sustainability. Future applications of this novel data source are expansive and varied, ranging for example from the assessment of the cohesiveness of regional policy narratives to the creation of novel databases for qualitative research. Even though this research is highly explorative, we believe that the combination of web mining and topic modeling with regional policy data from local governance institutions has the potential to be a valuable tool for understanding regional policy priorities and narratives and thereby providing a new way of analyzing the role these actors play in regional development. We hope that that the opportunities arising from this unique blend of data and methodology will encourage other researchers to use web mining to explore this area.

Freke Caset (p), Yitong Xia, Ben Derudder, Ate Poorthuis

Moritz Schütz

Polycentric urban regions (PURs) have become a key concept in regional studies both as an analytical framework to capture empirical realities as well as part of normative visions in regional development policies. Despite there being a rich scientific debate, ongoing PUR research paints a mixed picture. At the conceptual level, there is no consensus on what a polycentric urban system is and how its empirical diversity can translate into equally robust conceptual diversity. At the methodological level, a plethora of ad hoc analytical-operational frameworks are invoked, making it difficult to interpret and compare results and, ultimately, to draw robust conclusions on the impact of this regional form on people and places.
To push forward PUR research, recently calls were raised to establish a stronger tool-building community to work around collective, data-driven, computational geographic tools. These calls resonate more broadly with appeals to develop more accessible and modifiable software and tools with and within academic communities to advance the spatial sciences. In the context of PUR research, the proposition is that such pursuit could, at the very least, advance the comparability, reproducibility, replicability and reusability of findings.
While, today, there are several successful examples of open-source spatial software tools in GIScience and urban science communities, concrete explorations of what a PUR-oriented academic tool could look like remain thin on the ground. More importantly, there exist no prior efforts to elicit insights and epistemological positions from scholars who are actively shaping the field of PUR research with respect to the proposition to commit more strongly to tool-building. The latter venture is nonetheless crucial to anticipate to what extent, why, and to whom such interventions may be useful, and if these should be pursued at all.
Against this background, our contribution to this special session generates new and critical insights with respect to academic tool-building endeavors in the fields of urban and regional studies in general, and PUR-driven research in particular. We focus on the anticipated potential of a PUR-oriented academic tool to advance the field’s main research agendas, viewed from the perspective of the academic community itself. To this end, we have developed a web-based survey to elicit insights from scholars involved in PUR-driven research. Depending on the thrust of our findings, we will produce a set of specific recommendations (in terms of data requirements, tool capabilities, etc.) to develop a tool prototype, which will be evaluated in further rounds of validation.

Kateryna Zabarina (p), Muhammad Usman (p)

Freke Caset

This study examines the political violence at the regional level in Pakistan from the time period 2010- 2023 using the big data on conflicts provided by Armed Conflict Location & Event Data Project (ACLED). In particular, we use data on four different types of violence, which are battles, riots, explosions and violence against civilians. Due to some incorrect measurements (in a fatalities data) and duplicates (caused by providing information about several conflict sides), our study pattern consists of around 35000 of observations (varying from 1500 to 4000 point locations dependently on a year). We apply point pattern analysis techniques, specifically, spatial distribution of political violence to examine and visualize the phenomenon of political violence across various regions, intensity analysis to assess the severity of political violence and cross-dependence to investigate spatial coexistence between points of different types (battles, riots, explosions and violence against civilians). Choice of analysis instruments was easy because of availability of conflicts’ locations – we believe that results obtained with use of disaggregated data at the regional level will be more interesting and useful.