Dear community, I had to drop the idea of hosting my demo on github (the same way as the current Spotlight interface) since for pages created after 15.06.2016 they only serve them over HTTPS connection which causes troubles (mixed content) when asking Spotlight rest service at sztaku.hu over HTTP. The demo interface with enabled prediction of topic distribution can be found here [1] instead.
Please note that the predicted topics might seem not meaningful. First, Chang et al. [2] conclude that not always "good" topics are meaningful for humans. Second, this depends on the number of topics and feature types used to mine the model. The model currently loaded in the backend has 50 topics and feature vectors are built from entities, hypernyms and types. I will mine more models with different numbers of topics / feature types to see how the predictions change. Best, Wojciech [1]: http://de.dbpedia.org/spotlight-topics/ [2]: https://www.umiacs.umd.edu/~jbg/docs/nips2009-rtl.pdf On Tue, Aug 23, 2016 at 3:20 PM, Wojtek Lukasiewicz <woj.lu...@gmail.com> wrote: > Dear community, > > attached you'll find the link to my GSoC submission [1]. > I used the wiki page with my progress log and extended it with > submission-relevant information and links. > We (Alexandru and me) are currently having small difficulties with the > deployment of the demo to the server but it should be online tomorrow. > > Regards, > Wojciech Lukasiewicz > > > [1]: https://github.com/dbpedia/topicmodel-extractor/ > wiki/GSoC-2016-Progress-Log-Wojciech-Lukasiewicz >
------------------------------------------------------------------------------
_______________________________________________ DBpedia-discussion mailing list DBpedia-discussion@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dbpedia-discussion
