Relational Knowledge Discovery

Transcription

1 Relational Knowledge Discovery What is knowledge and how is it represented? This book focuses on the idea of formalising knowledge as relations, interpreting knowledge represented in databases or logic programs as relational data, and discovering new knowledge by identifying hidden and defining new relations. After a brief introduction to representational issues, the author develops a relational language for abstract machine learning problems. He then uses this language to discuss traditional methods such as clustering and decision tree induction, before moving onto two previously underestimated topics that are again coming to the fore: rough set data analysis and inductive logic programming. Its clear and precise presentation is ideal for undergraduate computer science students. The book will also interest those who study artificial intelligence or machine learning at the graduate level. Exercises are provided and each concept is introduced using the same example domain, making it easier to compare the individual properties of different approaches. M. E. MÜLLER is a Professor of Computer Science at the University of Applied Sciences, Bonn-Rhein-Sieg.

2

3 Relational Knowledge Discovery University of Applied Sciences, Bonn-Rhein-Sieg

4 CAMBRIDGE UNIVERSITY PRESS Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo, Delhi, Mexico City Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York Information on this title: / c M. E. Müller 2012 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2012 Printed in the United Kingdom at the University Press, Cambridge A catalogue record for this publication is available from the British Library Library of Congress Cataloguing in Publication data Müller, M. E., (Martin E.), 1970 Relational knowledge discovery / M.E. Müller. p. cm. ISBN (hardback) 1. Computational learning theory. 2. Machine learning. 3. Relational databases. I. Title. Q325.7.M dc ISBN Hardback ISBN Paperback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

5 Contents About this book Page 1 1 Introduction Motivation Related disciplines 8 2 Relational knowledge Objects and their attributes Knowledge structures 32 3 From data to hypotheses Representation Changing the representation Samples Evaluation of hypotheses Learning Bias Overfitting Summary 74 4 Clustering Concepts as sets of objects k-nearest neighbours k-means clustering Incremental concept formation Relational clustering 90 5 Information gain Entropy Information and information gain Induction of decision trees Gain again Pruning Conclusion 119 v

6 vi Contents 6 Rough set theory Knowledge and discernability Rough knowledge Rough knowledge structures Relative knowledge Knowledge discovery Conclusion Inductive logic learning From information systems to logic programs Horn logic Heuristic rule induction Inducing Horn theories from data Summary Learning and ensemble learning Learnability Decomposing the learning problem Improving by focusing on errors A relational view on ensemble learning Summary The logic of knowledge Knowledge representation Learning Summary 256 Notation 258 References 261 Index 267