You are here
This activity covers research in the areas of question answering and content acquisition and integration from textual documents. The developed techniques are applied in concrete use cases provided by the users involved in our various projects.
We have been working in Information Extraction for more than 10 years, moving from rule based systems to ones based on machine learning approaches. Concerning entity recognition, particular attention has been given to methods for speeding up SVMs, obtaining significant reduction on time execution. On the one hand, research on relation extraction has shown that methods based on shallow syntax analysis are competitive in many tasks when compared with their counterpart based on deep syntactic parsing; on the other hand, it assessed the inﬂuence of the accuracy of named entity recognition on the performance of the relation extraction. Experiments have been conducted on benchmarks wildly used in the literature. In addition, FBK participated to evaluation campaigns obtaining state-of-the-art results (e.g., extracting semantic relations between nominals at SemEval 2007). All tools developed are available with open source licenses. Particular attention has been given to the evaluation by contributing with concrete proposal for IE evaluation, including organization of international challenges.
Another research activity deals with the computational treatment of the emotional, pragmatic and stylistic features that contribute to linguitic communication, to complement pure semantic content.