In the preface of the book, the authors boldly state: “SVMs [support vector machines] have revolutionized the research in the areas of machine learning and pattern recognition, specifically classification, so much that for a period of more than two decades they are used as state-of-the-art classifiers.” Although the book starts with a bit of a background on classification spread over three chapters, including a chapter on perceptrons, which by the way features in the title, it is clear that the main topic of the book is SVM. Now the question of course arises whether this book has added anything new given the availability of a large repertoire of books and (freely available) technical reports on SVM from diverse angles. The authors further have claimed in the preface that the underlying theory of SVM is difficult to comprehend. True, perhaps, but the skeptic reader would probably wonder whether the presentation in the book has made that any easier. In fact, every discussion seems to be very brief and too mathematical with somewhat awkward notations at times. Perhaps the reason for such brief depiction is because this belongs to the “SpringerBriefs in Computer Science” series that “presents concise summaries of cutting-edge research and practical applications across a wide spectrum of fields.”
Most of the background is presented in the style of papers rather than books. The first chapter presents some relevant notations and notions and then briefly reviews some basic classification schemes. Chapter 2 focuses on linear discriminant functions, and chapter 3 reviews perceptrons. One seemingly odd thing that starts at chapter 3 and continues onward is the presentation of experimental results without enough information. When someone says: “we have conducted experiments,” the reader expects that those experiments are reproducible, and hence complete information pertaining to the experiments should be available. In chapters 4 and 5, the authors discuss SVM. Then comes the penultimate chapter, “Application to Social Networks,” which perhaps could have been the most interesting chapter of the book. However, the chapter is mostly filled with background on networks and related terms, social networks, some relevant measures like clustering coefficient, and so on, followed by a review on link prediction and similarity functions. In the sequel, we find only one brief section, section 6.6.3, “Link Prediction Based on Supervised Learning,” to justify the title of the chapter, where the authors simply report that they have conducted an experiment to test how SVM can be used for link prediction. This is in fact the only application that is reported in this chapter.
The book has some useful features like end-of-chapter summaries and point-by-point outlines of issues and advantages of SVM and other classifiers. But overall, a reader will probably be unsatisfied with the content of the book. Maybe the authors are not satisfied either--perhaps it had to be this way because of the format of the SpringerBriefs series. That in fact brings us to the first question we posed: whether this was an appropriate venture at all.