A PhD position:

Adaptive Machine Learning for the Analysis of Dynamic Data Stream in Social Networks

The analysis of scalable data using classification methods faces the challenge of the evolution of the knowledge over time. In the study of extremely variable phenomena such as social networks, the learning data (for the supervised case) and the classification space may become obsolete as time progresses. In the case of local modifications, there are incremental learning rules (Ning, 2007) to handle the local transition from one mode to another. In the case of Social network data stream analysis one needs to design efficient algorithms  for  the  monitoring  of scalable classes generated  from  local  or  global  changes in  the dataset (nodes/links addition or removal). Very often in these cases, the profiles of a community can join or change community while retaining or substantially modifying their behavior (attributes). In addition, the class update stage is very dependent on the performance of the algorithm currently applied before the updating of classes.

Objective:

This thesis aims at proposing an algorithm for dynamic classification of social data (interactions and topics) which integrates an evidence accumulation approach to contain the inter-class instability. The results of this study will be applied to the detection/monitoring of persistent communities in social networks.

Keywords: Social networks, Community detection, Classification, data mining, SVM, k-means.

To Apply: Applicants should send:
- a CV,
- a one-page letter of interest stating candidate experience in data mining and at least one project in relation to the subject.
- the name and contact information of one or two references,

Contact:
Email: babiga.birregah@utt.fr , Phone (office): +33 3 25 71 58 69 Skype: babiga.birregah

References:
Ning, H., Xu, W., Chi, Y., Gong, Y., & Huang, T. S. (2007, January).
Incremental Spectral Clustering with Application to Monitoring of Evolving Blog Communities. In SDM (pp. 261-272).