{"id":53205542,"date":"2011-05-01T07:56:00","date_gmt":"2011-05-01T07:56:00","guid":{"rendered":"http:\/\/su.blog.bunty.tv\/2011\/05\/01\/http-www-cs-cmu-edu-bsettles-pub-law-ecml10-pdf\/"},"modified":"2011-05-01T02:43:35","modified_gmt":"2011-05-01T02:43:35","slug":"http-www-cs-cmu-edu-bsettles-pub-law-ecml10-pdf\/","status":"publish","type":"post","link":"http:\/\/su.blog.bunty.tv\/?p=53205542","title":{"rendered":"http:\/\/www.cs.cmu.edu\/~bsettles\/pub\/law.ecml10.pdf"},"content":{"rendered":"<div class=\"sustuff\">Stumbleupon Review of : <a href=\"http:\/\/www.cs.cmu.edu\/~bsettles\/pub\/law.ecml10.pdf\">http:\/\/www.cs.cmu.edu\/~bsettles\/pub\/law.ecml10.pdf<\/a><a href=\"http:\/\/www.stumbleupon.com\/to\/2PpZSK\/www.cs.cmu.edu\/~bsettles\/pub\/law.ecml10.pdf\/t:4dbcc8ce02641;src:reviews\"><img src=\"http:\/\/bunty.tv\/images\/smallstumble.png\"><\/a><\/div>\n<p><\/p>\n<div class=\"review\"> &quot;Abstract. Most approaches to classifying media content assume a xed,<br \/> closed vocabulary of labels. In contrast, we advocate machine learning<br \/> approaches which take advantage of the millions of free-form tags obtain-<br \/> able via online crowd-sourcing platforms and social tagging websites. The<br \/> use of such open vocabularies presents learning challenges due to typo-<br \/> graphical errors, synonymy, and a potentially unbounded set of tag la-<br \/> bels. In this work, we present a new approach that organizes these noisy<br \/> tags into well-behaved semantic classes using topic modeling, and learn to<br \/> predict tags accurately using a mixture of topic classes. This method can<br \/> utilize an arbitrary open vocabulary of tags, reduces training time by 94%<br \/> compared to learning from these tags directly, and achieves comparable<br \/> performance for classi cation and superior performance for retrieval. We<br \/> also demonstrate that on open vocabulary tasks, human evaluations are<br \/> essential for measuring the true performance of tag classi ers, which tra-<br \/> ditional evaluation methods will consistently underestimate. We focus<br \/> on the domain of tagging music clips, and demonstrate our results using<br \/> data collected with a human computation game called TagATune.&quot; <\/div>\n","protected":false},"excerpt":{"rendered":"<p>Stumbleupon Review of : http:\/\/www.cs.cmu.edu\/~bsettles\/pub\/law.ecml10.pdf &quot;Abstract. Most approaches to classifying media content assume a xed, closed vocabulary of labels. In contrast, we advocate machine learning approaches which take advantage of the millions of free-form tags obtain- able via online crowd-sourcing &hellip; <a href=\"http:\/\/su.blog.bunty.tv\/?p=53205542\">Continue reading <span class=\"meta-nav\">&rarr;<\/span><\/a><\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_et_pb_use_builder":"","_et_pb_old_content":""},"categories":[416],"tags":[1116678,96],"_links":{"self":[{"href":"http:\/\/su.blog.bunty.tv\/index.php?rest_route=\/wp\/v2\/posts\/53205542"}],"collection":[{"href":"http:\/\/su.blog.bunty.tv\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/su.blog.bunty.tv\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/su.blog.bunty.tv\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"http:\/\/su.blog.bunty.tv\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=53205542"}],"version-history":[{"count":0,"href":"http:\/\/su.blog.bunty.tv\/index.php?rest_route=\/wp\/v2\/posts\/53205542\/revisions"}],"wp:attachment":[{"href":"http:\/\/su.blog.bunty.tv\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=53205542"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/su.blog.bunty.tv\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=53205542"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/su.blog.bunty.tv\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=53205542"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}