Abstract

A large body of work in natural language processing (NLP) for clinical applications is based on processing electronic health records (EHRs). While the latter are rich in information there are typically only few records per patient. More recently there has been interest in processing user generated content (UGC) such as social media posts collected over time to make predictions about individuals’ mental health. Such UGC data is available at much more frequent temporal intervals than EHRs but may be noisier. So far the majority of work in NLP on mental health prediction, even when using longitudinal social media data, involves distinguishing individuals with a condition from controls rather than assessing individuals’ mental health at different points in time. This talk will present the objectives addressed by my five-year Turing AI Fellowship through which I aim to establish a new area in natural language processing on personalised longitudinal language processing. I will give an overview of the state-of-the-art in this area, the challenges involved and work in progress on developing sensors for capturing digital biomarkers from language and heterogeneous UGC to understand the evolution of an individual over time.

Bio: Maria Liakata is a Turing AI fellow and Professor in Natural Language Processing (NLP) at the School of Electronic Engineering and Computer Science, Queen Mary University of London and the Department of Computer Science, University of Warwick. At the Turing she founded and co-leads the NLP and data science for mental health interest groups and supervises PhD students. She is in receipt of one of the five Turing AI fellowships, on Creating time sensitive sensors from user-generated language and heterogeneous content. She is the PI of projects on “Emotion sensing using heterogeneous mobile phone data”, “Language sensing for dementia monitoring & diagnosis” and “Opinion summarisation from social media”. Her research interests include opinion mining and summarisation, NLP for social and biomedical applications, longitudinal models of multi-modal and heterogeneous data, rumour verification.