Automatic information extraction and text mining in medical abstracts.

Abstract

This study is in the area of information extraction (IE), which seeks to extract pieces of related information from unstructured text to populate a database or an ontology. Most IE systems employ a pattern-matching technique to identify the information to be extracted. Patterns are learnt from a large annotated training set, which requires substantial human effort. This study investigates a semi-supervised learning approach to learn IE patterns. The approach uses a small number of seed patterns to automatically generate a training set and learns IE patterns from the training set by an Apriori algorithm. The study is carried out in the context of extracting information related to potential treatments of colon cancer from medical abstracts. It focuses on extracting 3 kinds of semantic relations: • Treatment relation: the disease and its potential medical treatment • Dosage relation: the treatment and its dose • Effect type relation: the treatment and its effect type.

The objectives of this study are to develop a method for automatic construction of IE patterns using semi-supervised learning, to develop an IE system for extracting disease-treatment information from medical abstracts, and to develop an ontology for representing disease-treatment information found in medical abstracts.