Detecting Mobility Functions in Free Text as an Indicator of Disability
The detection of disability is of great importance in multiple medical domains and in social programs. For instance, Social Security Administration (SSA) suffers from a significant delay in disability eligibility decisions due to the expensive human resources needed to review dozens if not even hundreds of pages per applicant. In this talk, we will present a disability detection approach using LSTM models. We identify mobility mentions in text given the International Code Functionality (ICF) scheme, since mobility is one of the strongest indicators in disability. Additionally, we will discuss the impact of using in- vs. out-of-domain word embeddings and try to leverage large amount of non-medical data to enhance the performance of the LSTM model.
Ayah Zirikly is a postdoctoral fellow at the NIH working with the epidemiology and biostatics section. She focuses on identifying disability mentions in free text in both: medical specialists’ notes and patients. Additionally, she is very enthusiastic about incorporating the use of Natural Language Processing (NLP) techniques in identifying mental health issues in free text. Over the last two years in collaboration with UMD, she has been focusing on suicidal and depression prediction in social media. Currently, she is working with NIH in collaboration with Stanford University to identify suicidal ideology among the veterans. Her PhD dissertation, under the supervision of Dr. Mona Diab, focused on information extraction (specifically Named Entity Recognition) in low-resource languages, where limited training data is available.