Recommender systems are living through a crisis today. After the initial success of simple relevance measures that were developed to construct search and recommendation systems, users are facing more and more problems whit satisfying their informational needs. The context of more ubiquitous Web use demands more sophisticated systems, capable of anticipating users needs and providing the right information when needed, in a way adapted to the users situation and context.
I my PhD thesis I have explored the ways how rich Semantic Web graphs can be used to deliver advanced information search and recommendation functionalities, satisfy complex user needs. The design of information systems more adapted to humans, with more enriching impact to augmenting human intellect continues to occupy a central place in my research.
Finding semantically related terms has become important for Web search, online advertising and other scenarios. However, the semantic relatedness of terms, being socially defined, is an evolving phenomenon and changes over time. This projects aims to establish a way to recognize the changes in semantic relatedness of terms over time. For instance if the Olympics are announced to be held in a particular city, the name of the city suddenly becomes much more related to the term "olympics". This project uses the real-time Web to sense such changes in semantic relatedness. The new related terms can be used to adapt advertising campaigns by including new, temporarily relevant terms, or to change the behavior of other online systems.
hyProximity is a system for keyword recommendation based on the semantic proximity of concepts. It uses DBPedia - a Semantic Web version of Wikipedia to find concepts that are semantically related. As opposed to the systems that rely on co-occurrence of terms in text, this projects aims to find the related terms based on their meaning. The system is capable of recognizing the context of relevancy (two concepts are not related in the same way in all situations; e.g. clay is a concept close to artistic modeling, but this proximity is irrelevant if we are dealing with clay in the context of clay mining and raw material processing). The concept suggestions provided by the system tend to be more novel and surprising to the user while remaining highly relevant. hyProximity is used by hypios in the search of potentially interesting topics for Open Innovation problem solving, but can also be used in online advertising, and similar use cases.
As a part of my PhD thesis I am working on good ways to search for experts on Linked Data. As opposed to legacy approaches that took only one expertise hypothesis (assumption about what does it take to be an expert), I try to leverage the flexibility and multiple purposes of Linked Data to allow a more sophisticated approach.
People use emoticons every day as the easiest and most interesting way to express their emotions, actions and affective states. But the exchange of emoticons across different systems is limited. Smiley Ontology is an ontology that aims to enable interchange of emoticons between different systems, without any loose of its semantics, and to enable the capture of semantics of emotions and emotional states that can be found in smileys.
The project is aimed at enabling the integration and exchange of data related to a user’s presence in the online world. As opposed to static user profiles and the interoperability in that domain, well supported by the FOAF Vocabulary, the Online Presence project targets the interoperability of dynamic properties that determine a user’s current state of presence (e.g., custom messages, IM statuses etc.).