Search

Scientific articles often contain relevant geographic information such as where field work was performed or where patients were treated. Most often, this information appears in the full-text article contents as a description in natural language including place names, with no accompanying machine-readable geographic metadata. Automatically extracting this geographic information could help conduct meta-analyses, find geographical research gaps, and retrieve articles using spatial search criteria. Research on this problem is still in its infancy, with many works manually processing corpora for locations and few cross-domain studies. In this paper, we develop a fully automatic pipeline to extract and represent relevant locations from scientific articles, applying it to two varied corpora. We obtain good performance, with full pipeline precision of 0.84 for an environmental corpus, and 0.78 for a biomedical corpus. Our results can be visualized as simple global maps, allowing human annotators to both explore corpus patterns in space and triage results for downstream analysis. Future work should not only focus on improving individual pipeline components, but also be informed by user needs derived from the potential spatial analysis and exploration of such corpora.

Cancers arise from the accumulation of somatic genome mutations, which can be influenced by inherited genomic variants and external factors such as environmental or lifestyle-related exposure. Due to the heterogeneity of cancers, precise information about the genomic composition of germline and malignant tissues has to be correlated with morphological, clinical and extrinsic features to advance medical knowledge and treatment options. With global differences in cancer frequencies and disease types, geographic data is of importance to understand the interplay between genetic ancestry and environmental influence in cancer incidence, progression and treatment outcome. In this study, we analyzed the current landscape of oncogenomic screening publications for geographic information content and quality, to address underrepresented study populations and thereby to fill prominent gaps in our understanding of interactions between somatic variations, population genetics and environmental factors in oncogenesis. We conclude that while the use of proxy-derived geographic annotations can be useful for coarse-grained associations, the study of geo-correlated factors in cancer causation and progression will benefit from standardized geographic provenance annotations. Additionally, publication-derived geographic provenance data allowed us to highlight stark inequality in the geographies of cancer genome profiling, with a near lack of sizable studies from Africa and other large regions.

Defining and identifying duplicate records in a dataset is a challenging task which grows more complex when the modeled entities themselves are hard to delineate. In the geospatial domain, it may not be clear where a mountain, stream, or valley ends and begins, a problem carried over when such entities are catalogued in gazetteers. In this paper, we take two gazetteers, GeoNames and SwissNames3D, and perform matching – identifying records in each that are about the same entity – across a sample of natural feature records. We first perform rule-based matching, establishing competitive results, then apply machine learning using Random Forests, a method well-suited to the matching task. We report on the performance of a wider array of matching features than has been previously studied, including domain-specific ones such as feature type, land cover class, and elevation. Our results show an increase in performance using machine learning over rules, with a notable performance gain from considering feature types, but negligible gains from other specialized matching features. We argue that future work in this area should strive to be more reproducible and report results on a realistic testing pipeline including candidate selection, feature extraction, and classification.

Spatial language is an important part of everyday communication, and placenames,like Zürich and Switzerland, are a major way to refer to geographical locations. Inwritten text documents, these placenames and their associated spatial languageconvey spatial information to the reader, setting the geographical context of news,events, blogs, and even scientific research. For the intended audience of a text,shared experience and spatio-temporal context makes understanding these referencesto places effortless. However, for computer systems, automatically extracting andrepresenting this geographical content is a challenging process which must deal withplacename ambiguity and vagueness in spatial language. Indeed, creative strategiesare required to turn text about locations into information that is explicitly spatial,such as points or regions that can be visualized on a map.It is thistext-to-spaceprocess which is the subject of this thesis, focusingboth onresourcesused to link placenames and geographical representations, andonapplicationsof text-to-space pipelines for particular use cases. Placenames formthe ‘glue’ to create geographical representations for text documents, because theycan be identified in text using tools like Named Entity Recognition (NER), andlinked to spatial representations (geometries) which are catalogued in placenameresources known as gazetteers. In their simplest form, gazetteers provide a name,feature type, and geometry for a set of named places. However, each gazetteerdiffers in terms of which places are catalogued and how, with no global authoritativegazetteer providing a definitive list.In the first part of this thesis, we look at this gazetteer heterogeneity, firstby analyzing and comparing the spatialcoverageof two global gazetteers usingaggregated record counts, and then by aligning ormatchingindividual records fromtwo gazetteers when they represent the same real-world entity. In our analysis ofgazetteer coverage, we find wide discrepancies in

How do people perceive landscapes? What elements of the landscape do they identify as characteristic of a landscape? And how can we arrive at descriptions, and ultimately representations that better reflect people’s notions of landscapes? In this study, we collected landscape descriptions from five landscape types at 10 study sites in Switzerland. For each site, we collected data from three sources: free lists with participants, hiking blogs, and Flickr tags. Free lists were obtained through on-site interviews with visitors, hiking blogs were gathered by focused crawling of web content, and Flickr tags were selected based on spatial footprints obtained from the hiking blogs. We quantitatively compared landscape descriptions between data sources and landscape types using cosine similarity. We found that descriptions from the same data source were significantly more similar, irrespective of landscape type. Descriptions from the same landscape type were more similar, but only within the same data source. Through a qualitative analysis of different aspects of landscape in our content, we found that each data source offered a different distribution of landscape aspects. For example, while Flickr tags contained high proportions of toponyms, they contained little content relating to sense of place. In contrast, hiking blogs contained more information about sense of place. Our approach combining these varied textual sources thus offers a more holistic view on landscapes. This study constitutes a step toward extracting semantically rich descriptions of landscapes from a variety of sources and using this information to distinguish different landscapes, with potential applications for landscape monitoring and management.

Explicitly linking text documents to geographical space is an important processing step for applications such as map visualization, spatial querying, and placename disambiguation. In this work, we present a proof-of-concept processing pipeline to generate spatial footprints for a spatially-rich, manually-annotated corpus of hiking blogs. We present preliminary results obtained by exploiting the spatially-focused nature of our input data and the rich placename resources at our disposal. Future work will fully automate the pipeline and systematically examine the influence of processing decisions on the footprints and downstream tasks.

Gazetteers are playing a central role in the current data revolution as key tools to link content to geographical space. However, their geographies and idiosyncrasies are poorly understood despite their potential to worsen application outcomes and information inequalities. In this study, we analyze two open gazetteers, GeoNames and the Getty Thesaurus of Geographic Names, in terms of the quantity and spatial distribution of features in Great Britain, illustrating how they provide a different and inconsistent picture of the region and are still far less detailed than the institutional gazetteers curated by Ordnance Survey.