Quality is crucial for any use of collected data. In addition to more traditional methods of data cleansing for data warehousing, new kinds of data provided by people have become important. To get truthful information from the answering individuals, proper motivation and compensation is vital. The book aims to discuss methods that allow for the efficient handling of numerous data providers, so it discusses mechanisms based on game theory and mathematical statistics.
Interesting examples of problems are introduced: treatment of service/product reviews about restaurants and hotels, voting opinion polls, crowdsensing of environmental data, and grading of work by peers. These examples are used to distinguish between queried information being objective (verifiable afterward) or subjective (a distribution of answers should be evaluated). Agents are motivated to choose a cooperative strategy through an individually rational benefit. The mechanisms should encourage agents to provide truthful information and demotivate/separate those giving valueless/false answers, thus enforcing positive self-selection. The authors clearly explain cases, introduce notions, and give their properties.
The chapters discuss problems and their properties in a series arranged from the simplest to the increasingly complicated. In cases involving verifiable information, incentives can be given to the agents according to the ground truth. When the answer to the question is subjective, agents can be treated according to the answers of the others. First, those mechanisms (based on game theory) depending on external parameters are discussed. After this chapter, it is shown how these parameters can be determined by additional information asked by the agents or by statistical evaluation of the answers. During these discussions, the authors refine methods by diluting/reducing their preconditions. “Truth serums” are described, coming from previous research results of the authors or from the work of others. To maintain brevity and highlight their essence, proofs of some statements are sketched only with reference to suitable publications. Examples demonstrate problems and solutions.
After the mechanisms are introduced, practical aspects of their application are discussed in three chapters: “Prediction Markets,” “Agents Motivated by Influence,” and “Decentralized Machine Learning.”
The well-structured book provides a clear and didactic overview of the theme. It summarizes not only the results the authors achieved during the last 1.5 decades; it gives the present status of the whole theory. The reference list is up to date, and the figures, diagrams, tables, and formulas are clear-cut.
The authors warn that this field is only in its infancy. However, because of its importance and the beauty of the theory, I recommend the book to those aiming to elicit truthful information for valuable analysis.
