• Author:
  • Ing. Jaroslav Kuchař, Ph.D.
• Supervisor:
  • doc. Ing. Tomáš Vitvar, Ph.D. - FIT ČVUT v Praze
• Reviewers:
  • prof. RNDr. Peter Vojtáš, DrSc. - MFF UK v Praze
  • Dr. Roman Dumitru - University of Oslo, Norway
  • Priv. Doz. Mag. Dr. Gerhard Wohlgenannt - WU Wien, Austria

PhD Thesis, Reviews

With a growing number of users browsing various web sites, the need of proper analysis and understanding of their behaviour becomes one of the most studied areas last years. Users interacting with a specific content provide huge amount of data during the behaviour. Such interactions are not self-explanatory till they are not properly represented and connected to the well described content items. Technologies of the Semantic Web become a part of many areas of informatics and they play a significant role in a representation of knowledge. With help of the Semantic Web we can build rich representations connecting users and content they are interacting with. Those rich representations associate interactions performed by users and available knowledge about the content and they allow to infer and utilize multiple relations.

This doctoral thesis studies a particular aspect of the recent research in using semantics for building and utilizing rich representations connecting users and the content. Our contributions address specific issues of using semantics in following areas: 1) Data acquisition - we deal with situations of modern user interfaces when a user performs multiple interactions per one content item and the required output is one relation representing user interest in the content 2) Semantization - we focus on linking of content to a knowledge base allowing a consecutive extraction of additional features and build its semantic representation 3) Enhancement - since most of existing knowledge is created by humans and fragments of links within a knowledge base can be missing, there is a need to manage the knowledge base using link predictions in order to get rich semantic representations 4) Utilization in Preference learning and Recommendation - rich semantic representations allow to utilize their relations and perform an intelligent selection of resources based on existing relations or they allow to build readable and concise user preference models that are applicable for recommendations of content items.

In particular, the main contributions of the dissertation thesis are as follows:

1. Method for an aggregation of semantically enriched user interactions.
2. Algorithm for linking content to a public knowledge base and a method for semantic aggregation.
3. Link prediction method that allows enhancement of semantic representations with respect to temporal information.
4. Method for selection of the most relevant target among a predefined set of candidates.
5. Preference learning and recommendation technique profiting from semantic annota- tions.

Keywords: Semantic Web, Web Usage Mining, Link Prediction, Preference Learning, Association Rules, Recommendation.