We describe the participation of MIRACLE group in the QA track at CLEF. We participated in three subtasks and presented two systems that works in Spanish. The first system is a traditional QA system and was evaluated in the main task and the Real-Time QA pilot. The system features improved Named Entity recognition and shallow lin- guistic analysis and achieves moderate performance. In contrast, results obtained in RT-QA shows that this approach is promising to provide an- swers in constrained time. The second system focus in the WiQA pilot task, that aims at retrieving important snippets to complete a Wikipedia. The system uses collection link structure, cosine similarity and Named Entities to retrieve new and important snippets. Although the experi- ments have not been exhaustive it seems that the performance depends on the type of concept.
Add Comment