-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Enrico Zini schrieb: > Hello, > > On private mail one visitor suggested me to add a [related] link to the > package list that would start a new search with the package name. > > What he was doing was searching the package name instead of keywords; > that would indeed match the package at least, bring the package's tags > in the 'Wanted' taglist and show other packages with same tags. > > I hadn't realized that this would work, but it seems that it would! > > It would be interesting, in that case, to score the tags by 'relevance'. > I think TFIDF[1] could be used to say how relevant is a tag for a package > compared to the rest of the collection. Anyone would like to > experiment?
Hi Enrico, my knowledge of tf.idf is limited, but what would you consider as term frequency here? The count of a tag is always for a given package is always 1, and you're left with idf only. Still, it would be useful, the relevance of a tag some::tag for a package (that has been tagged with it, that is) would then be 1 / docfrequency(some::tag) or, at your option, log(df). What would be possible to do is to calculate the relevance of a tag for a given search keyword, instead of finding just the tags with the highest frequency, which is computationally more expensive, since it involves computing P(tag|keyword), P(!tag|keyword), P(tag|!keyword) and P(!tag|!keyword). I'm neither a master at information retrieval nor with statistics, but I've done some study-related work in that direction lately. Is that somehow along your lines of thought? greetings Torsten - -- Torsten Marek <[EMAIL PROTECTED]> ID: A244C858 -- FP: 1902 0002 5DFC 856B F146 894C 7CC5 451E A244 C858 Keyserver: subkeys.pgp.net -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.2 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iD8DBQFDfjYrfMVFHqJEyFgRAsUUAJ47kWNGeCf69QjckxrUj/arGmslpACfdgsT xQKNtJGIiE3DgOYm1DdVrDE= =eaAb -----END PGP SIGNATURE----- _______________________________________________ Debtags-devel mailing list [email protected] http://lists.alioth.debian.org/mailman/listinfo/debtags-devel
