Due to the proliferation of Internet devices and connections, data is generated at an extreme speed. Right now, zettabytes of data are generated daily. How to make meaning out of mostly unstructured data is very difficult. Among these vast amounts of data, one type of data deserves special attention: rank data. For example, you may be asked to make a preference about candidates in an election. Or when choosing a book to read, your preferences could be book A first, then book B, then book C, and so on.
When you have multiple sets of rank data, how can you make an informed decision from these sets of data? For example, Alice wants to elect candidate one in an election. Bob’s preferences are candidate one, then two, then three. Charlie has no comments for one, but he prefers three over two. Then, what election result is most logical and conforming to the ranked data?
Learning and decision-making from rank data surveys recent progress, including statistical models for rank data, parameter estimation algorithms, a rank-breaking framework, mixture models for rank data, Bayesian preference elicitation, and socially desirable group decision-making from rank data. Theoreticians can use it to better understand rank data, practitioners can apply the algorithms introduced, and professors can use it as a graduate-level textbook.
This book is rather mathematically oriented. The presentation uses a lot of mathematical symbols and Greek letters that are hard to pronounce and memorize. We know this kind of topic is not easy to narrate. Nevertheless, a more common set of symbols, instead of these strange Greek alphabets, would make reading easier.
In general, this book is well printed with good resolution for figures and everything. However, I found a spelling error on page 6, line 8; with the convenience of spell-check, this error is rather strange.
The author proposes three future directions: handling richer types of rank data, smarter interactions, and multi-criteria design and analysis. But the prerequisite is a set of clean and valid data. Therefore, in the future, how to validate data will be more important work. Without it, other endeavors are useless.