Kostas Christidis defended successfully his thesis "Recommender Systems with Probabilistic Topic Models" on Monday 11 Feb, 2013

His thesis is positioned in the research area of decision support systems and specifically recommender systems. It focused on the design and development of recommender systems based on probabilistic topic models.

Kostas has explored the possibility to design improved recommender systems inside enterprises, communities and in electronic commerce based on latent topics.An approach is presented for integrating existing domain knowledge in a recommender system. Additionally, a methodology is proposed for utilizing probabilistic topic models for complete and effective modeling of employee expertise on addressing problems. A methodology is described for extracting consumer preferences from datasets using topic models. Finally, a methodology is presented for utilizing the unstructured text found in electronic auction marketplaces in order to provide recommendations to buyers and sellers.

The recommender systems proposed in this thesis have displayed a number of common characteristics:

• They reduce the dimensions of the recommendation problem and provide fast online recommendations, having trained the topic models.
• They satisfy the user needs for accuracy and recall of all interesting objects.
• The topics extracted can provide significant insight to the system manager or owner.

The presentation (in Greek) can be found here.