Fields of interest Recommender systems and application in real-life context Use of thesauri and ontologies in information retrieval tasks Blog mining using time information and semantics Association rule mining with the use of semantics and ontologies Trust metrics for social networks

More specifically:

1. My current research on recommender systems focuses on scaling recommendation algorithms to large-scale social networks and the development of distributed solutions that improve the performance of the respective centralized, can handle much larger graphs and provide equally good recommendations. My research is also extended on more complex recommender system problems, such as package recommendations, person recommendations through common activity detection, as well as recommendations that lead to habit change.
2. Based on a new measure of semantic relatedness between terms of a thesaurus, we have defined, we are now able to define the semantic relatedness between texts. The two measures have been successfully applied on Word Sense Disambiguation tasks and have been extensively tested in word-to-word and text-to-text similarity benchmark data sets. A demo of these measures is available at http://omiotis.hua.gr.
   On top of these two measures, we are developing algorithms that classify, cluster and rank documents and terms and examine applications in various linguistic, text mining and information retrieval tasks. 
   
3. We focus mainly on web 2.0 applications, as well as in document collections that contain time information concering document editing. Current research on theblogosphere takes advantage of as much information from blogs as possible (publication date, author, categories, tags, in and out links, content etc.) in order to locate subsets of special interest. In a different application, we tested the hypothesis concerning the relation between blog contents and real world events. Since the hypothesis was validated experimentally, we are now working on reversing the result and use changes in blogosphere contents as alets for local-scale events that do not reach the media..
   In this same field of the blogosphere, we are recently applying several ranking and recommendation models that confront the problem of spam blogs.
4. An algorithm that processes transactions in web logs and locates frequent patters has been designed. The algorithm takes advantage of the taxonomic organization of content and locates frequent generalized patterns in content access.
5. Several metrics and algorithms that take into account the special characteristics of each social network, the semantics of hyperlinks and user provided ratings and generate local and global ranking of social network users.