This study explores the use of natural language processing in the medical field. Using the i2b2 2012 data from Harvard we have explored various available models that aim to extract temporal relationships from unstructured clinical text. Through a thorough literature review we have explored the underlying mechanisms of industry standard models that perform this task and the temporal logic that drives them.
Our goal was to obtain a strong understanding of the task itself and a broader insight into the software and tools available to tackle the problem. We detail these tools, their strengths, and weaknesses and how they can be used when extracting temporal relationships from free text.
This approach allowed us to build up our knowledge, from simple named entity recognition models to far more advanced temporal relation extraction models. We then focused on two promising models to feed our testing data.
We compared the results of the models used by examining the standard F1-scores. The proprietary models outperformed open-source models by a significant margin. However, we propose that further testing and fine-tuning could close this gap.
Additionally, where possible, we’ve visualised the temporal relations within a given paragraph.
Hello world!
Welcome to WordPress. This is your first post. Edit or delete it, then start writing!
optimize2023
April 13, 2023
