Jolt Awards: The Best Books, October 01, 2013 Five notable books every serious programmer should read.
Productivity Award: Taming Text
By Grant S. Ingersoll, Thomas S. Morton, and Andrew L. Farris
Perhaps once a year, I have the experience of picking up a book on a topic I'm unfamiliar with and immediately becoming absorbed by it. I started reading this volume to get a sense of the major issues and solutions in text processing and found myself devouring the book sneaking it into meetings and any other place where I could spend a few more minutes working my way through it.
Handling unstructured text (that is, text in documents, rather than databases) presents all kinds of complex problems, most of them having to do with the central role of natural language. For example, the problem of stemming (which is the process of reducing variants such as plural of nouns, verb tenses, etc. to their canonical form), searching, fuzzy searches, indexing, parsing, and categorizing.
The authors' explanations of these problems are supremely readable and the solutions are presented with code implementations. Where possible, the authors rely on open-source tools, such as Lucene, Solr, Apache Mahout, OpenNLP, and others to do the heavy lifting. In addition to algorithms, this book shows how to build text-processing clusters, manage document libraries for processing, and handle other pragmatic concerns.
The final third of the book is dedicated to the unique challenges of how to tag and classify documents automatically. Despite the wealth of techniques provided earlier, the problem of entity recognition (for example, does "Polish doorknob" refer to an item or to a task?) is presented and explained in all its gory complexity. At times, the challenges seem intractable, but as the authors demonstrate time and again, there are good, available tools and specialized algorithms that can get you most of the way through the wilderness.
At no point does the presentation become so esoteric as to lose the reader rather, the authors of Taming Text keep finding stimulating ways to lead developers into new areas that need to be mined for optimal results to ensue.