Detecting hierarchical relationships and roles from online interaction networks

摘要: In social networks, analysing the explicit interactions among users can help in inferring hierarchical relationships and roles that may be implicit. this thesis, we focus on two objectives: detecting between and inferring of interacting via same online communication medium. both cases, we show considering temporal dimension of interaction substantially improves detection roles. The first thesis is problem inferring implicit relationships from users. Based promising results obtained by standard link-analysis methods such as PageRank Rooted-PageRank (RPR), introduce three novel time-based approaches, \Time-F" based a defined time function, Filter Refine (FiRe) which hybrid approach RPR Time-F, and Time-sensitive (T-RPR) applies in way that takes into account time-dimension process detecting hierarchical ties. We experiment datasets, Enron email dataset to infer managersubordinate relationships from exchanges, scientific publication coauthorship dataset detect PhD advisor-advisee paper co-authorships. Our experiments demonstrate perform better terms of recall. particular T-RPR turns out superior over most recent competitor methods well all other approaches propose. The second examining communication behaviour of working activity order identify different hierarchical roles played We propose approaches. approach, supervised learning used train classification algorithms. second approach, address sequence problem. A novel sequence framework generates time-dependent features frequent patterns at multiple levels granularity. Our is a exible technique for applied domains. We an educational collected asynchronous communication tool students accomplish underlying group project. Our experimental findings supervised achieves best mapping of their when individual attributes students, information about reply them quantitative time-based features are considered. Similarly, our multi-granularity pattern-based framework shows competitive performance students' roles. Both approaches are significantly than baselines