There are many historical itineraries that describe routes as a list of settlements and the travel distances along the way. They are an important source of information for various kinds of research in the humanities, providing insights into for example the development of human mobility and historical road networks. In this paper, we develop an approach for aligning these itineraries with a modern gazetteer (database of places). We combine textual information (historical toponyms) and spatial information (travel distances) into a Hidden Markov model. Naively calculating a maximum likelihood explanation is slow, but careful algorithm engineering achieves high performance suitable for user interaction. We demonstrate the practical potential of our approach by geo-referencing 48 itineraries (containing 691 stops) from two important historical guidebooks published in 1563 and 1597: our approach is fast and accurate. Additionally, we show how to use sensitivity analysis to power an efficient user interface for quality assurance.
Geographical and spatial descriptions in the premodern world are structurally different from the modern era, where spatial understanding is based on cartographic navigation. This paper presents an experimental process to tag, retrieve, and identify geographical information as described in premodern primary sources, together with the issues and possible solutions. The proposed method defines specific categories of geographical information and a markdown system to mark these categories in the source. Having tagged the data, we extract it and geographical locations and their connections are identified through a heuristic approach: the extracted geographical entities are initially aligned with existing geographical references and secondary sources. String similarity approaches might provide fuzzy identifications which need to be verified and disambiguated. In this paper, we describe the process of annotation and extraction of geographical descriptions, experiment some toponyms matching metrics, report the results, and offer possible solutions to handle disambiguation through the existing contextual information in the source. The process is applied to two different datasets, proposed as test cases: a classical Arabic geographical text and a Roman itinerary.
For the retrieval of concise entity relation information from large collections or streams of documents, existing approaches can be grouped into the categories of (multi-document) summarization and knowledge extraction. The former tend to fall short for this task due to the involved amount of information that cannot be easily condensed, while knowledge extraction approaches are often pattern-based and too discriminative for exploratory purposes. For location relations in particular, this translates to a set of very short relationship descriptors that predominantly encode hierarchical or containment relations such as located in or capital of. As a result, available knowledge bases that are typically populated through knowledge extraction are limited to these discrete and typed relations. In contrast, the representation of document collections as implicit networks of entities, terms, and sentences has emerged as a way to encode a much wider range of entity relations and occurrences, which can be leveraged for filtering the relevant information and enabling subsequent interactive explorations.In this paper, we discuss the extraction of descriptive sentences for sets of entities from such implicit networks to support an interactive exploration, and apply them to the extraction of complex location relations that are not hierarchical or containment-based. We introduce and compare efficient ranking methods for sentence extraction that address this entity-centric search task by leveraging entity and term relations in implicit network representations of large document collections. Based on Wikipedia articles and Wikidata as a knowledge base, we demonstrate the extraction of novel location relations that are not contained in the knowledge base.
In this short paper we present ongoing work on an approach to extract spatial information from natural language place descriptions. Our focus is on handling context-dependent spatial information and strategies for sensibly resolving contextual dependencies.
We examine the GeoNames gazetteer as a hub of Linked Geospatial Data. We survey quality and linkage characteristics to understand how well it supports the traversal of the Semantic Web and which limitations exist. We examine different traversal scenarios originating with GeoNames and present findings related to link availability and distribution along language and geospatial dimensions and discuss the role of cross-lingual issues.
This article presents an effort to integrate spatial and textual data processing tools into a modular software package which features preprocessing, geocoding, disambiguation and visualization.
We investigate gazetteer matching between natural features in Switzerland. We produce a gold standard dataset for a subset of features from 8 natural feature types in GeoNames aligned with their corresponding match(es) in SwissNames3D, an authoritative gazetteer. Based on this dataset, we comment on feature type alignments between the two resources and on type-specific differences to take into consideration for the matching task. We present preliminary results of rule-based matching and plans for future work.
We consider events as complex named entities, recursively consisting of simple named entities (people, places, organizations, dates, etc.) and/or other complex named entities. We have developed a processing chain dedicated to extracting, indexing and searching for social events in Web pages. In this context, we are proposing a generic similarity computation function that targets any type of complex named entity. In this paper, we implement this function according to three approaches which we shall describe and then experiment on a set of social events.
New retrieval models promise deeper integration of multiple features and sources of information. The inclusion of thematic and location features in a joint factorization model allows location to be modeled as a first-class feature and can improve a range of tasks in geographic information retrieval and recommendation. In this position paper, we describe these factorization models and how they can be useful for corpus and user need understanding and further GIR use cases. We argue that using joint factorization models can be a powerful tool in the integration of complex features and relationships present in many GIR data sources and applications.
Historical itineraries, often accessible as lists or tables describing places visited in sequence, are abundant resources and also important objects of study for humanities scholars. This article advances a novel method for automatically geocoding tabular itineraries, combining approximate string matching with a cost optimization algorithm based on dynamic programming. Experiments with a dataset of historical itineraries, with ground-truth geocoding annotations provided by domain experts and leveraging also the GeoNames gazetteer, attest to the effectiveness of the proposed method. The obtained results show that while approximate string matching can already achieve very low median errors, with many toponyms matching exactly against GeoNames entries, the combination with cost optimization can significantly improve results in terms of the average distance towards the correct disambiguations.
This paper is concerned with automatic georeferencing of river networks from raster images such as aerial photos or maps. Determining reasonable assignments between a given network of rivers derived from a textual description and an image is subject to high combinatorial complexity and uncertainty. We investigate the application of spatial reasoning in automatic georeferencing.
Events as composites of temporal, spatial and actor information are a central object of interest in many information retrieval (IR) scenarios. There are several challenges to such event-centric IR, which range from the detection and extraction of geographic, temporal and actor mentions in documents to the construction of event descriptions as triples of locations, dates, and actors that can support event query scenarios. For the latter challenge, existing approaches fall short when dealing with imprecise event components. For example, if the exact location or date is unknown, existing IR methods are often unaware of different granularity levels and the conceptual proximity of dates or locations. To address these problems, we present a framework that efficiently answers imprecise event queries , whose geographic or temporal component is given only at a coarse granularity level. Our approach utilizes a network-based event model that includes location, date, and actor components that are extracted from large document collections. Instances of entity and event mentions in the network are weighted based on both their frequency of occurrence and textual distance to reflect semantic relatedness. We demonstrate the utility and flexibility of our approach for evaluating imprecise event queries based on a large collection of events extracted from the English Wikipedia for a ground truth of news events.
The work in this paper is motivated from two different perspectives: First, gazetteers as an important data source for Geographic Information Retrieval (GIR) applications often lack historic place name information. More focused historic gazetteers are a far cry from being complete and often specialize only on certain geographic regions or time periods. Second, research on historic route descriptions---so called itineraries---is an important task in many research disciplines such as geography, linguistics, history, religion, or even medicine. This research on historic itineraries is characterized by manual, time-consuming work with only minimalistic IT support through gazetteers and map services. We address both perspectives and present a depth-first branch-and-bound (DFBnB) algorithm for deducing historic place names and thus the stops of ancient travel routes from itinerary tables. Multiple phonetic and character-based string distances are evaluated when resolving parts of an itinerary first published in 1563.
We propose a categorization algorithm for text content description such as tags for images from social media or crowd sourcing services, to identify places characteristics. The algorithm is based on a spatial coverage and a multi-facets categorization. We describe how it can be applied to individually process images from Flickr in order to extract geo-spatial knowledge. It is particularly dedicated for places with a small number of photos. The extraction process is done using categorization rules based on geographic and terminological knowledge resources.
Advances in technology have continually progressed our understanding of where people are, how they use the environment around them, and why they are at their current location. Having a better knowledge of when various locations become popular through space and time could have large impacts on research fields like urban dynamics and energy consumption. In this paper, we discuss the ability to identify and locate various facility types (e.g. restaurant, airport, stadiums) using social media, and assess methods in determining when these facilities become popular over time. We use standard natural language processing tools and machine learning classifiers to interpret geotagged Twitter text and determine if a user is seemingly at a location of interest when the tweet was sent. On average our classifiers are approximately 85% accurate varying across multiple facility types, with a peak precision of 98%. By using these standard methods to classify unstructured text, geotagged social media data can be an extremely useful tool to better understanding the composition of places and how and when people use them.
Ongoing initiatives promoted by cultural institutions and public administrations engage in the development of textual corpora issued from the general public. In this work, we deal with a spoken corpus of life stories and a crowd-sourced Web corpus of people's contributions related to urban planning issues in their city. Located information constitutes an essential component in these corpora. Toponyms refer to official names (e.g. Congo) which are listed in gazetteers but often to generic locations such as un endroit très beau (a beautiful place). Because of the nature of the corpora, these generic locations are inherently subjective, vague and descriptive. For enabling automated exploitation of these texts, it is crucial to properly detect such kinds of place mentions. In this sense, the present work provides a comparative study of state-of-art NER1 systems, most importantly of supervised tools such as Stanford NER, for the identification of generic locations in thematic corpora.
Local news articles are an important source of knowledge about local events, place-specific culture, and peoples' thoughts about their environment. Reliable geocoding of such articles is the first step towards unlocking such local knowledge for community engagement and development. However, existing geo-referencing methods and tools do not work well for local news because they do not reflect the ways local people encode and communicate geographical knowledge. This paper argues that local news requires a different method and infrastructure support for effective geo-referencing. To gain insights on the unique aspects of local gazetteers and the nature of ambiguities, we present an analysis of a collection of local new articles. We found that place references in local news have their special vocabulary, and that their ambiguities are handled differently by local people. We translated such insights into a gazetteer-based geocoding solution that combines progressive geocoding with a smart footprint recommender. Progressive geocoding service uses Nominatim (OpenStreetMap) as the initial gazetteer to jump-start the construction of local gazetteer for a community and by the community. LocusRecommender automatically suggests the best matches from gazetteer ranked by a set of heuristic rules. Preliminary evaluation shows that our smart footprint recommender predicts 80% of the answers by its top-three recommendations.
Semantic aligning of heterogeneous geographical data from different sources behaves unsatisfactory on Geographical Semantic Web (GSW) due to the flat structure of GSW and the influence of spatial features. To solve this problem, this paper proposes a holistic framework for GSW aligning. This holistic framework firstly produces the initial matched results respectively for classes, properties and instances by the approval voting strategy, and then enhances these results by the mutual cooperating mechanism. Especially, spatial distance and spatial index are introduced to align instances and to improve the performance of aligning class and aligning property. To demonstrate its ability, this holistic framework is tested with two real GSWs. Compared with the state-of-the-art holistic alignment system, namely PARIS, this framework gains a large number of matched pairs. The Fl values of aligning class, aligning property and aligning instance respectively are 0.562, 0.545 and 0.646, all of which are higher than PARIS's.
Farmers face pressure to respond to unpredictable weather, the spread of pests, and other variable events on their farms. This paper proposes a framework for data aggregation from diverse sources that extracts named places impacted by events relevant to agricultural practices. Our vision is to couple natural language processing, geocoding, and existing geographic information retrieval techniques to increase the value of already-available data through aggregation, filtering, validation, and notifications, helping farmers make timely and informed decisions with greater ease.
In this paper, we point out to the shortcomings of precision and recall in evaluating the performance of geoparsing algorithms. We propose separate processes for evaluating toponym recognition and toponym resolution stages, and also propose new metrics that quantify the performance of toponym resolution.
