Scientific and Technological Objectives

Emergent metadata

The initial phase of the project will deal with collecting actual raw data from existing, live systems. By “raw data” we mean the emergent metadata that arise because of agent interactions in online social communities, as described in the introduction. Several online communities are readily accessible over the web: for a selected set of these systems, tools will be developed and deployed to harvest the relevant data, metadata and temporal dynamics, and to store the acquired information in a form amenable for data analysis.

Data analysis of emergent properties

Examining quantitative aspects of folksonomy is a highly important area of research. Our objective is the set up of several protocols of data analysis to be performed on the raw data sets. A data analysis protocol is defined by: (1) indicating a specific quantity / observable / estimator suitable of a quantitative measure on the raw data sets; (2) acquiring the existing software tools, or developing new specific tools, needed to perform the measure; (3) extracting the relevant statistical information characterizing the analyzed data sets.

The aim of the data analysis is to identify and quantify emergent properties of the system in study, i.e. properties that can not be simply inferred from the behavior of the single agent. Beyond suggesting the collection of new or more refined raw data, the results of the data analysis will be used to

• identify general features common to the different systems in study
• characterize/discriminate the specific features of different systems in study
• orient the modelling phase of the research project (see below)
• providing benchmarks to test/improve existing systems or to suggest the creation of new more performing systems

Modeling and simulations

The objectives of this research area are twofold:

• understanding complexity: develop models that captures the essence of the emergent dynamics and explain how it might arise from the interactions of single agents
• taming complexity: formulate design strategies that allow controlling the behavior of the system at the emergent level by suitably choosing the microscopic dynamics of the interacting agents

One of the most important goals is to construct, implement and study specific modeling schemes aiming at reproducing, predict and control the emergent properties seen in the semiotic dynamics orchestrated in on-line communities. We plan in particular a modeling activity at different scales. On the one hand it will be important to construct microscopic models of communicating agents performing language games without any central control. At a different scale we shall consider more coarse-grained probabilistic models. Several models will be proposed to address specific aspects/scales of folksonomy. The models will allow computer simulation aimed at measuring emergent features to be compared with the results of the data analysis activity. The simulations should give an insight in how users select tags, what kind of categories and category structures underlying the evolving system of tags, how categories and tags are related to the objects being tagged, etc. It will also give information on what kind of more global structures (such as the most frequent tags) can be provided to users to optimize their on-line community infrastructure. The models will require components for assigning or adopting tags, categorizing data, and collective dynamics. However the approach will be to keep the models as simple as possible, identifying the minimal ingredients responsible for the emergent properties. The minimal character of the models should make a more analytical mathematical study feasible.

A possible way to tackle the complexity of the systems is to individuate different time scales, which can be separated. For instance, we expect that the dynamics of the social network of the folksonomy could be different from the time scale of the dynamics of the resources and of the tags. In this case one can, as a first approximation, propose a model of tags and/or resource dynamics based on a given, slowly evolving social network topology. This kind of assumption should be tested and corroborated as much as possible with the observations coming from the real data analysis.

Finally, the output of this activity has the potential to feed back into the data collection activity, specifically to the live social tagging system developed as part of, in order to experimentally verify the devised control strategies and demonstrate the technological advantage achieved by the present project.