The 12th edition of the Dutch-Belgian Information Retrieval Workshop (DIR) was hosted in the medieval town of Ghent in Belgium. The primary aim of DIR is to provide an international meeting place where researchers from the domain of information retrieval and related disciplines can exchange information and present innovative research developments. Thanks to the hospitality of Thomas Demeester and his tireless helpers from Ghent University, as well as a Belgian beer reception, the event created a great atmosphere for fruitful discussion.
In his keynote On regarding document test collections as samples, Steven Robertson gave a preview on his upcoming ECIR’12 work. He argues that document collections should be considered a sample from the population of universally possible documents. To account for this, he proposes a statistical method under which collection-based performance scores can be interpreted as estimates of the underlying universal performance.
My personal highlights among the many interesting research talks, demonstrations and poster presentations include:
- Wouter Weerkamp et al. – People Searching for People: Analysis of a People Search Engine Log
The authors inspect a sizable query log from a people search engine and investigate different notions of search intent and search behaviour.
- Maya Sappelli et al. – Collection and Analysis of Ground Truth Data for Query Intent
The authors describe an experiment in which they observe 11 participants’ Web search behaviour by means of a browser plugin and subsequent intent annotation by the searchers themselves. They argue that this is a more feasible and realistic way of determining intent than relying on external annotators.
- Maral Dadvar et al. – Improved Cyberbullying Detection Using Gender Information
The detection of on-line harassment and bullying typically relies on the presence and absence of “blacklisted” terms in natural language communication. The authors demonstrate the existence of gender-specific differences in bullying styles. Employing dedicated gender models to account for these differences, they show convincing performance gains.
- Marc Bron et al. – Linking Archives using Document Enrichment and Term Selection
The authors use a retrieval approach for linking newspaper articles from a richly annotated archive to a sparsely-annotated multimedia video archive.