ISBN: 978-3-319-22734-4 (print) – 978-3-319-22735-1 (digital)
Daniel Schall’s book presents a series of experiments in the scope of social network-based recommender systems. Our daily life increasingly involves interacting with digital social networks such as Facebook, Twitter, LinkedIn, and GitHub. Their complexity impedes deciding whom to add as a friend, expert, coworker, or collaborator. The book suggests novel methods which narrow down the choices in form of a ranked list. The author emphasises two main aspects. First, he focuses on social networks in form of directed graphs. This applies to Twitter, GitHub, and GooglePlus among others. Second, the author considers how authority affects communities. Thereby, he devises a novel method by combining the Hyperlink Induced Topic Search (HITS) and the PageRank algorithm. The method enables him to estimate how much of an authority entities represent.
The first experiments compares a collection of similarity metrics applicable to social networks on their ability to predict links. The author introduces a novel metric based on triads between people on GitHub, GooglePlus, and Twitter.
The second experiment investigates whom a given developer should follow on GitHub. The authors applies the previously mentioned combination of HITS and PageRank to obtain a context-sensitive, personalised ranked list.
The third and forth experiment shift the focus onto organisations rather than individuals. The third experiment evaluates strategies to decide which organisation to cooperate with in scientific projects. The authors uses a data set obtained from the European Union FP7 programme. The authors extends the previously used method to account for structural information and costs.
The forth experiment takes the same data but considers a different problem. The experiment asks which organisation should take the role as information broker. The author introduces a query language which enables requestees to discover collaborators.
The book does not represent a text book on social recommender systems. The author clearly states limitation. For instance, the book does not cover semi-local or global link prediction techniques. Instead, the book presents recent results regarding link prediction and community formation in social networks represented as directed graphs. The author directs readers to additional resources in each chapters’ related work section.
The author put noticable effort into illustrating results and presenting methods comprehensibly. The book contains plenty of figures and tables with detailed results. Unfortunately, the printed copy does not provide coloured figures. In addition, each chapter features pseudo code describing the methods applied.
Overall, the book represents a suited resource for researchers and practitioneers who deal with directed social networks.