Understanding datasets
This section is divided into two parts. In the first part, we need to discuss the challenges we have faced in order to generate the dataset. In the later section, we will be discussing the attributes of the dataset.
Challenges in obtaining the dataset
As we all know, the health domain is a highly regulated domain when it comes to obtaining the dataset. These are some of the challenges I want to highlight:
For summarization, ideally, we need to have a corpus that contains original text as well as a summary of that text. This is called parallel corpus. Unfortunately, there is no good, free parallel corpus available for medical document summarization. We need to obtain this kind of parallel dataset for the English language.
There are some free datasets available, such as the MIMIC II and MIMIC III dataset, but they won't contain summaries of the medical transcription. We can access just the medical transcription from this dataset. Gaining access to this dataset is a lengthy...
