I have focused my research on developing large and scalable language analyses with
applications to physical, mental, and population health. I'm interested in
computational methods for identifying linguistic signs of neurological disorders affecting
children and the elderly. In addition, I build
contextualized neural models for clinical concept extraction
and NER, and neural models to better
understand
social and behavioral determinants of health.
Clinical NLP
University of Florida | AI in the Health Sciences Initiative
Johns Hopkins Institute for Clinical and Translational Research
JHU Malone Center for Engineering in Healthcare
- Building enterprise-level NLP and deep learning solutions on cloud computing.
- Deep neural models to improve clinical text concept linking to medical ontologies.
- Developing large scale solutions for de-identification of clinical notes.
- Standardizing 350m clinical notes and converting into HL7 Clinical Document Architecture.
- Providing NLP research services to JHU clinical community (ranging from pilot work and seed grants to large and fundamental projects).
Identifying Social and Behavioral Determinants of Health from Electronic Medical Records
University of Florida | AI in the Health Sciences Initiative
Johns Hopkins Center for Population Health IT
JHU Malone Center for Engineering in Healthcare
- Contextualized neural representations for temporal and severity analysis of SBDH.
- Predicting mortality risk and clinical outcome in opioid overdose patients.
- Understanding suicide mortality outcomes in the Mid-Atlantic region
- Identifying social needs in pediatric primary care.
- Assessing disparities in outcomes of type 2 diabetes ADEs in minority populations.
- Contextualized neural language models for identifying smoking status.
- Determining risk factors related to early liver transplantation in severe alcoholic hepatitis.
Social Media Analysis for Health & Well-Being
University of Florida | AI in the Health Sciences Initiative
Penn World Well-Being Project
Penn Institute for Biomedical Informatics
- Collaborated in developing Penn WWBP Differential Language Analysis ToolKit (DLATK).
- Large-scale language modeling of Facebook users to explore health-related factors.
- Developed big data analysis pipeline for noisy social media language (text normalization, sentence segmentation, pos-tagging, parsing, language modeling, relation extraction, sentiment analysis).
- Computational models for exploring medication side-effects from Twitter.
- Determining pregnancy timeframe from contextual patient-generated content.
- Pharmacoepidemiologic evaluation of birth defects from social media postings.
Computational Analysis of Cognitive and Language Impairment
Johns Hopkins Institute for Clinical and Translational Research
OHSU/OGI Center for Spoken Language Understanding
- Developed a fundamental framework for utilizing computational models for the analysis of impaired language.
- Unsupervised and generalizable framework applicable to a wide range of language impairment conditions (e.g. dementia, MCI, Alzheimer's, and autism)
- Computational models for sentiment, affect, arousal, empathy, and locus of control (applicable to language analysis in mental conditions like anxiety and depression)
- Lexical ranking methods for identifying idiosyncratic topic digressions in narratives.
- Quantifying topic perseveration and responsiveness in spontaneous conversations.
Text-to-Scene Conversion, WordsEye
Columbia University Spoken Language Processing Group
OHSU/OGI Center for Spoken Language Understanding
- Creating a scenario-based lexical knowledge resource for text-to-scene conversion.
- Crowd-sourcing the annotation of frame-based semantic relations.
- Categorizing and labeling object models for use in 3D rendering.
Ontology Construction
Shahid Beheshti University NLP Lab
- Designing and creating VigNet, a semantic frame-based ontology for spatial relations.
- Developing FarsNet, the Persian WordNet.
- Information extraction from large and noisy text for building frame-based ontologies.