Extracting Causal Knowledge from a Medical Database Using Graphical Patterns

Extracting Causal Knowledge from a Medical Database Using Graphical Patterns Christopher S.G. Khoo, Syin Chan and Yun Niu Centre for Advanced Information Systems, School of Computer Engineering Blk N4, Rm2A-32, Nanyang Avenue Nanyang Technological University Singapore 639798 assgkhoo@ntu.edu.sg; asschan@ntu.edu.sg; niuyun@hotmail.com Abstract This paper reports the first part of a project that aims to develop a knowledge extrac- tion and knowledge discovery system that extracts causal knowledge from textual da- tabases. In this initial study, we develop a method to identify and extract cause-effect information that is explicitly expressed in medical abstracts in the Medline database. A set of graphical patterns were constructed that indicate the presence of a causal rela- tion in sentences, and which part of the sentence represents the cause and which part represents the effect. The patterns are matched with the syntactic parse trees of sentences, and the parts of the parse tree that match with the slots in the patterns are extracted as the cause or the effect. 1 Introduction Vast amounts of textual documents and data- bases are now accessible on the Internet and the World Wide Web. However, it is very difficult to retrieve useful information from this huge disorganized storehouse. Programs that can identify and extract useful information, and re- late and integrate information from multiple sources are increasingly needed. The World Wide Web presents tremendous opportunities for developing knowledge extraction and knowl- edge discovery programs that automatically ex- tract and acquire knowledge about a domain by integrating information from multiple sources. New knowledge can be discovered by relating disparate pieces of information and by infer- encing from the extracted knowledge. This paper reports the first phase of a project to develop a knowledge extraction and knowl- edge discovery system that focuses on causal knowledge. A system is being developed to identify and extract cause-effect information from the Medline database – a database of ab- stracts of medical journal articles and conference papers. In this initial study, we focus on cause- effect information that is explicitly expressed (i.e. indicated using some linguistic marker) in sentences. We have selected four medical areas for this study – heart disease, AIDS, depression and schizophrenia. The medical domain was selected for two reasons: 1. The causal relation is particular important in medicine, which is concerned with devel- oping treatments and drugs that can effect a cure for some disease 2. Because of the importance of the causal re- lation in medicine, the relation is more likely to be explicitly indicated using linguistic means (i.e. using words such as result, ef- fect, cause, etc.). 2 Previous Studies The goal of information extraction research is to develop systems that can identify the passage(s) in a document that contains information that is relevant to a prescribed task, extract the infor- mation and relate the pieces of information by filling a structured template or a database record (Cardie, 1997; Cowie & Lehnert, 1996; Gai- zauskas & Wilks, 1998). Information extraction research has been influenced tremendously by the series of Mes- sage …

Content truncated to 500 words