logo ws 3




An Introduction to j−δ Diagrams 

By Prof.Giuliano Armano

Abstract: Performance measures are used in various stages of the process aimed at solving a classification problem. In this talk, a pair ofnovel measures (i.e., ϕand δ) is defined, able to capture the concepts of characteristic and discriminant capability. These measures come in two forms, i.e., unbiased and biased. The former has the same expressive power of ROC diagrams, although ϕ-δ diagrams allow to inspect at a glance several quantities deemed relevant in the machine learning community –such as accuracy, bias, and break-even points. The latter kind of diagrams highlight to which extent the class ratio (namely, the imbalance between negative and positive samples) affects the corresponding metric space. The use of ϕ-δ diagrams can give important information to researchers involved in machine learning tasks, such as classifier performance assessment and feature ranking / selection.  

Summary: The talk will be focused on the following aspects:

  • analysis of the most acknowledged metrics –in particular, accuracy, negative predicted value, precision, specificity, sensitivity, MCC and F1,
  • motivationfor the need of novel metrics and their definition,
  • main propertiesand graphical representationof the (ϕ,δ) space –for feature and classifier assessment,
  • simple case studiesaimed at highlighting the usefulness of ϕand δ,
  • main propertiesand graphical representationof the generalized (ϕ,δ) space. 



Mining Lurking Behaviors in Online Social Networks 

By Andrea Tagarelli

Abstract: Research on social networks (SNs) has traditionally focused on influential users, experts and trendsetters. By contrast, less attention has been paid to the fact that all large-scale on-line communities are characterized by a participation inequality principle, i.e., the crowd of a SN does not actively contribute, rather it lurks. Lurkers are those silent members of a SN who gain benefit from others' information without giving back to the SN. However, because they acquire knowledge from the SN, lurkers are social capital holders. Within this view, a major goal is to de-lurk such users, that is, to encourage them to more actively participate in the SN. Developing solutions to de-lurking problems can support enhanced personalization of user access and adaptation of the design of web-based systems and their interfaces, thus ultimately helping sustain a community over time with fresh ideas and perspectives. Lurking analysis in SNs lies at the confluence of many disciplines, such as social science, human-computer interaction, and computer science.

The goal of this tutorial is to provide an overview of research issues and solutions related to the characterization and analysis of lurkers in SNs. More specifically, it will explain the several meanings of “lurking” and related implications, introduce the principles and motivational factors underlying lurking behaviors, and discuss the main strategies of de-lurking. It will also guide through the computational approaches and methods to mine lurkers and unveil their behavioral patterns in social networks. It will further cover use-cases in different types of online social environments, and present perspectives and challenges in the field.

primi sui motori con e-max