A semantic sequence similarity based approach for extracting medical entities from clinical conversations

Clinical conversations between physicians and patients can provide a rich source of data, information, and knowledge. A plethora of tools and technologies have been developed to identify attributes of interest in unstructured text. However, identifying the name and correct value of an attribute, from real world data, in a timely manner is a nontrivial task. In this manuscript we present a novel pipeline using transfer learning, clinical concept dictionaries, and pattern matching to provide an end-to-end solution for identifying attributes and extracting their values from natural clinical text. On real-world data, with 1176 instances, we achieve an accuracy of 56.21%, which is 3% higher than the baseline methodology.