Dear spotlight developers, I am working at usage of DBpedia Spotlight with other data than dbpedia/wikipedia information. Instead, I feed the lucene index with surface forms and context information I retrieve from some authority file. (I work with german language, fyi, if this makes any difference at any point) At the moment I use this attempt for historical persons and it shows some improvement already, compared with spotlight annotation with dbpedia links.
But the main problem at the moment seems to be the annotation of persons, when only the last name and some context information (profession and a place of activity e.g.) is provided by the text, but not a full name. My current surface forms are containing the full name, e.g. 'Alan Turing' because last name only surface forms would map some of the surface forms to several thousand entities (I have also tried this at some point, and, not surprising, the amount of false positives exploded). >From my point of view this seems to be an issue with the lucene back end in >general rather than a specific problem with my data. By looking at the >different web services I also get the feeling that the statistical back end >might handle this problem better? I haven't looked into the statistical version yet, so I don't know anything about how it works in general, about language support, adaptability (I know there is an i18n tutorial similar to the lucene one, which I used for adaption) and so on. Therefore, before I start this, I would love to know, if you would even recommend switching to the statistical version with this rather specific problem. thanks in advance, cheers, Germaine ------------------------------------------------------------------------------ Want excitement? Manually upgrade your production database. When you want reliability, choose Perforce Perforce version control. Predictably reliable. http://pubads.g.doubleclick.net/gampad/clk?id=157508191&iu=/4140/ostg.clktrk _______________________________________________ Dbp-spotlight-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/dbp-spotlight-users
