山东大学 基于web的信息检索和知识发现chap11.pptVIP

Support Vector Machines;Text classification: Last time and today;Linear classifiers: Which Hyperplane?;Another intuition (non-unique);assuption;Support Vector Machine (SVM);w: decision hyperplane normal （法线） xi: data point i yi: class of data point i (+1 or -1) (并不是 1/0） Classifier is: f(xi) = sign(wTxi + b) Functional margin of xi is: yi (wTxi + b) But note that we can increase this margin simply by scaling w, b…. Functional margin of dataset is minimum functional margin for any point b is the distance from data set to the plan.;x;Geometric Margin;Linear SVM Mathematically;Bi1 : WTx1+b = 1; Bi2: WTx2 + b = -1; 2= WT (x1-x2) since ||x1 – x2 || 2 = ?) ? = 2/||W||;Linear Support Vector Machine (SVM);Linear SVMs Mathematically (cont.);Solving the Optimization Problem;The Optimization Problem Solution;Soft Margin Classification ;Soft Margin Classification Mathematically;Soft Margin Classification – Solution;Classification with SVMs;Linear SVMs: Summary;Non-linear SVMs;Non-linear SVMs: Feature spaces;The “Kernel Trick”;Kernels;Most (over)used data set 21578 documents 9603 training, 3299 test articles (ModApte split) 118 categories An article can be in more than one category Learn 118 binary category distinctions Average document: about 90 types Average number of classes assigned 1.24 for docs with at least one category Only about 10 out of 118 categories are large ;Reuters Text Categorization data set (Reuters-21578) document;New Reuters: RCV1: 810,000 docs;Per class evaluation measures;evaluation;ROC curve;Dumais et al. 1998: Reuters - Accuracy;Reuters ROC - Category Grain;ROC for Category - Crude;ROC for Category - Ship;Results for Kernels (Joachims 1998);Table 15.1 Training and testing complexity of various classifiers including SVMs. Training is the time the learning method takes to learn a classifier over D, while testing is the time it takes a classifier to classify one document. For SVMs, multiclass classification is assumed to be done by a set of