Random Post: get_results("SELECT ID,post_title,guid FROM $wpdb->posts WHERE post_status= \"publish\" ORDER BY RAND() LIMIT 1"); $p=$post[0]; echo ('' . $p->post_title . ''); ?>
RSS .92| RSS 2.0| ATOM 0.3
  • Home
  • About
  •  

    Notes from Linking Geodata seminar at CeRch

    Note, this blog entry was originally published in May 2010.

    While on a bit of a road trip, had the chance to give a short seminar at the Centre for e-Research at Kings College London. This was informal, weren’t expecting much of a showing, so there are no slides, here is a quick summary.

    Introduced by Dr Stuart Dunn, and i talked about project ideas we had just been discussing – the attempt to mine the English Place Name Survey for its structure, now called CHALICE – mining archaelogical site records and artefact descriptions and attaching them to entities in OpenStreetmap using LinkedGeodata.org – mining key reference terms from documents in archives, attempting to link documents to reference data.

    Linked Geodata seems like a good place to start, pick out a sample entry and walk through the triples, at this point a bit of jumping about and graph-drawing on the whiteboard.

    There’s a list of mappings between items in Linked GeoData and in dbpedia.org, and likely thus through to geonames.org and other rich sources of Linked Data. Cf. Linked Geodata Datasets. Via sameas.org geographic links can be traversed to arrive at related media objects, resources, events.
    geonames.org has its 8m+ points and seems to be widely used in the academic geographic information retrieval community, due to its global coverage and open license.

    The text mining process used in the Edinburgh geoparser and elsewhere is two-phase, the first is the extraction purely looking at the text, of entities which seem likely to be placenames; the second phase is looking those names up in a gazetteer, and using relations between them to guess which of the suggested locations is the most likely.

    Point data, cartographic in origin. Polygon geoparsing.
    Machine learning approaches to both phases.

    We looked at UK-postcodes.com and the great work @pezholio has done on the RDF representations of postcodes there, with links across to some of the statistical area namespaces from data.gov.uk – along with the work that Ordnance Survey Research
    have in hand
    , there’s lots of new Linked Open Geodata in the UK.

    Historic names and shapes, temporal linking, these are areas where more practical, and open research has yet to be done.

    Comments are closed.