Relational knowledge discovery. (Q2891516)

This book presents a formal, relational view of knowledge discovery. In order to do so, the author first covers some basics of relational algebra that are used during the whole book as a unifying principle to describe several machine learning approaches to knowledge discovery. The book is intended to be used as a textbook, and it contains a very large and comprehensive collection of examples and exercises that should help the reader or the instructor in the sue of the book.NEWLINENEWLINEThe book is insightful and useful for people with already a good grasp of machine learning who are interested in a more formal perspective to knowledge discovery. It is refreshing to see a broad and encompassing collection of definitions of machine learning in a single piece of work. Typically, the literature of machine learning is very narrowly focused on their own niches. It is also nice to see the statistical, the search and the algorithmic learning theory definitions of machine learning in a single text.NEWLINENEWLINEIn Chapter 2, the author presents a formal explanation of set and relation algebra, which form the basis for knowledge representation. An interesting aspect is that it deals with the subject in a purely abstract and theoretical way without focusing on any particular knowledge representation, which makes all the concepts presented in the book very general and applicable to a wide range of relational machine learning formalisms. Maybe the only non-general aspect is to define objects by a set \(F\) of features, which does not encompass non-feature-based machine learning such as sequential data. After that point, the book considers four different approaches to machine learning (induction of decision trees, rough set data analysis, inductive logic programming and ensemble learning), and presents them from a relational point of view.NEWLINENEWLINEFinally, although it is intended as a text book and contains numerous exercises and examples, I would not recommend it as a first contact with machine learning, since the book focuses more on the relational perspective of the field than in a first insight into machine learning. Therefore, many of the fundamental principles and intuitions behind most machine learning techniques covered are out of the scope of the book. The author still includes a significant amount of introductory and inspirational material such as a discussion about the ``representation gap'' which is essential for students to learn, or the very well covered discussions about ``bias'' in machine learning which is masterfully covered in the book. However, many basic ideas around searching hypothesis spaces, such as version spaces, are not even mentioned (which is very surprising, given their key role in the topic being discussed). Another missing connection is that with formal concept analysis, which is unfortunate. Finally, the main ideas of why ensemble learning works are basically lacking in the book. Instead of covering the standard principles covered in similar texts (see, for example, the book by T. G. Dietterich from 2000), the author presents intuitive explanations that are sometimes very far from the actual reasons for which ensemble learning works.