Temporal Information Extraction in Clinical Domain

Abstract

In the present context, there are many records available in textual format that o ers a means to share knowledge and valuable informa- tion which are in most of the scenarios temporally bound. It is not straightforward to access the information which are relevant to a spe- ci c time frame without identifying the time, event information along with their relation in any form of documentation.These texts could contain various rules and attributes forming complex words describing time and events. The importance of temporality is especially crucial in the medical domain where the records of a patient are varying with age, medication and surgeries. Most of the existing approaches for information extraction are of lim- ited use due to its limited applicability in the medical domain in a comprehensive manner. A comprehensive mechanism was developed to extract the temporal information and their relationship that ex- ists among them pertaining to medical records namely, pathological, clinical and radiological reports. TIECA attempts aims to provide an e cient approach comparing the performance of rule based and machine learning approaches, along with a clear analysis elaborating the rationale behind the variation of the evaluation criteria values. Successful results were achieved by evaluating the proposed method using data obtained from Mayo Clinic.These experiments show state of the art promising results, prov- ing that the proposed approach suits information extraction within the medical domain.