http://www.paperoftheweek.com/2007/01/19/discussion-of-joachims-svms/ Read. Learn. Discuss. Mon, 18 Jun 2007 15:05:50 +0000 hourly 1 http://wordpress.org/?v=3.0.1 http://www.paperoftheweek.com/2007/01/19/discussion-of-joachims-svms/comment-page-1/#comment-7 ozzy Thu, 01 Feb 2007 15:25:47 +0000 http://www.paperoftheweek.com/2007/01/19/discussion-of-joachims-svms/#comment-7 "4. Most Text Cat. problems are linearly separable. This is the key idea behind why they work." What does this mean really? If textcategorization problems were linearly separable you should be able to get 100% on the training set everytime, if your algorithm is working properly even with a linear kernel. (which is not the case in most of the Reuters Collections.). And also the word "Most" doesn't mean anything to me, what does it bring to the table that most of them are linearly separable? “4. Most Text Cat. problems are linearly separable. This is the key idea behind why they work.”
What does this mean really? If textcategorization problems were linearly separable you should be able to get 100% on the training set everytime, if your algorithm is working properly even with a linear kernel. (which is not the case in most of the Reuters Collections.).
And also the word “Most” doesn’t mean anything to me, what does it bring to the table that most of them are linearly separable?

]]>