Abstract
Given the large amount of data mining algorithms, their combinations (e.g. ensembles) and possible parameter settings, finding the most adequate method to analyze a new dataset becomes an ever more challenging task. This is because in many cases testing all possibly useful alternatives quickly becomes prohibitively expensive. In this paper we propose a novel technique, called active testing, that intelligently selects the most useful cross-validation tests. It proceeds in a tournament-style fashion, in each round selecting and testing the algorithm that is most likely to outperform the best algorithm of the previous round on the new dataset. This 'most promising' competitor is chosen based on a history of prior duels between both algorithms on similar datasets. Each new cross-validation test will contribute information to a better estimate of dataset similarity, and thus better predict which algorithms are most promising on the new dataset. We also follow a different path to estimate dataset similarity based on data characteristics. We have evaluated this approach using a set of 292 algorithm-parameter combinations on 76 UCI datasets for classification. The results show that active testing will quickly yield an algorithm whose performance is very close to the optimum, after relatively few tests. It also provides a better solution than previously proposed methods. The variants of our method that rely on crossvalidation tests to estimate dataset similarity provides better solutions than those that rely on data characteristics.

Abstract
This paper presents an inductive method that learns a logic program represented as an ordered list of clauses. The input consists of a training set of positive examples and background knowledge represented intensionally as a logic program. Our method starts by constructing the explanations of all the positive examples in terms of background knowledge, linking the input to the output arguments. These are used as candidate hypotheses and organized, by relation of generality, into a set of hierarchies (forest). In the second step the candidate hypotheses are analysed with the aim of establishing their effective coverage. In the third step all the inconsistencies are evaluated. This analysis permits to add, at each step, the best hypothesis to the theory. The method was applied to learn the past tense of English verbs. The method presented achieves more accurate results than the previous work by Mooney and Califf [7]. © Springer-Verlag Berlin Heidelberg 1998.

Abstract

Abstract

Abstract
Paraphrase detection can be seen as the task of aligning sentences that convey the same information but yet are written in different forms. Such resources are important to automatically learn text-to-text rewriting rules. In this paper, we present a new metric for unsupervised detection of paraphrases and apply it in the context of clustering of paraphrases. An exhaustive evaluation is conducted over a set of standard paraphrase corpora and real-world web news stories (WNS) corpora. The results are promising as they outperform state-of-the-art measures developed for similar tasks. Copyright

Abstract