Zusammenfassung
This paper introduces Transductive Support Vector Machines (TSVMs) for text classifi cation. While regular Support Vector Ma chines (SVMs) try to induce a general deci sion function for a learning task, Transduc tive Support Vector Machines take into ac count a particular test set and try to mini mize misclassifications of just those particu lar examples. The paper presents an anal ysis of why TSVMs are well suited for text classification. These theoretical findings are supported by experiments on three test col lections. The experiments show substantial improvements over inductive methods, espe cially for small training sets, cutting the num ber of labeled training examples down to a twentieth on some tasks. This work also pro poses an algorithm for training TSVMs effi ciently, handling 10,000 examples and more.
Nutzer