Automatic term and relation extraction for medical question answering system

PhD ceremony: I. Fahmi, 14.45 uur, Academiegebouw, Broerstraat 5, Groningen

Thesis: Automatic term and relation extraction for medical question answering system

Promotor(s): prof. J. Nerbonne

Faculty: Arts

 

The main research question of Fahmi’s thesis is how term and relation extraction techniques can contribute to medical question answering'.This question is motivated by the fact that many of the recent open-domain Question Answering (QA) systems benefit from the vast amount of information on the Web, while for a domain-restricted QA system, such as a medical QA system, existing methods to extract information could be less promising. With relatively simple techniques (e.g., information retrieval and lexico-syntactic filtering), answers for open-domain QA systems can be extracted directly from the Web based on information redundancy, while for a medical QA system very little redundant information is available to support the answers since the size of its data sources is usually limited.

In his thesis Fahmi aims to increase the coverage of relation tables used by our medical QA system by generating relation patterns semi-automatically from corpora. He also aims to increase their precision by using semantic information from the Unified Medical Language System (UMLS) and by selecting relation instances related to medical terms.

The extraction of the relation tables involves several issues such as the extraction and labeling of medical terms and the extraction of medical term relations. In the course of his experiments, Fahmi raised six research questions to cover the issues. As a general conclusion, he found that relation tables generated using his method have a larger coverage and return higher precision answers compared to those generated using a manual method.