Analyzing Web Profiles using Probabilistic Ontologies

In this paper, we discuss our probabilistic ontological solution for analysis of Web profiles. The analysis of Web profiles is a very demanding and multi-layered task; especially probabilistic information in terms of probability distributions and weights is often the key to an expressive analysis. In our research we designed the Probabilistic Profile Analysis Ontology (PPAO) using Markov Logic Networks (MLNs) afterwards we conducted experimentation to evaluate the scalability and expressiveness of this solution. MLNs were chosen as the underlying formalism because of their probabilistic nature, intuitive and expressive modeling ability due to first order logic as well as ability to use approximation algorithms to improve reasoning performance. A similarity benchmark between profiles was designed within the PPAO concept as a special task and real static and probabilistic data from the German IRCLOVE community www.irclove.de was used for the analysis. Although our solution turned out to achieve expressive results, the experimentation revealed a mixed picture on scalability of profile analyzing with MLNs. Therefore we will discuss these results and propose possible research directions for further improvements on using probabilistic ontologies for profile analysis.