From IMC wiki
Revision as of 10:27, 29 May 2015 by Mpurver (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

AOTD: Analysing Online Therapy Dialogue


The funding will support a pilot project into using computational language processing to support and improve mental health therapy. Recent QMUL research applying computational linguistics methods to psychiatric therapy data shows that important factors such as patient satisfaction, symptoms and future treatment adherence can be automatically predicted from transcripts with accuracy comparable to human experts (see PPAT). The recent adoption of online therapy within the NHS can provide large volumes of therapy data (once suitably anonymised), and thus an opportunity to develop methods for automatically measuring therapy progress, predicting treatment outcomes and improving effectiveness.

Staff: Matthew Purver (PI), Christine Howes (PDRA)

Timescale: 15th November 2013 - 31st January 2014

Funding body: EPSRC IAA (via QM Innovation Fund)

QMUL account number: ECSA1N6R


Christine Howes, Matthew Purver and Rose McCabe. Linguistic Indicators of Severity and Progress in Online Text-based Therapy for Depression. In Proceedings of the ACL 2014 workshop on Computational Linguistics and Clinical Psychology (CLCP), Baltimore, MD, June 2014.

Hamed Haddadi, Patrick Healey, Rose McCabe, Matthew Purver. Healthcare Informatics for Mental Health: Recent Advances and the Outlook for the Future. Background paper, Mental Health Foundation report Starting Today: the Future of Mental Health Services, 2013.

Relevant docs

IATP handbook:

One year analyses of IAPT recovery:

POL online therapy data

Codes used in POL online therapy data

Definitions of outcome measures

Restrict some analyses to people who have completed treatment (59)

Research Questions

  1. What topics (client and agent together but not distinguishing between speakers) are discussed in online CBT?
  2. What topics are associated with PHQ status (on a per session basis)?
  3. Do topics predict PHQ change (adjusting for number sessions and therapist ID)?
  4. Is level or variability of positive and negative sentiment associated with PHQ status?
  5. Is change in sentiment over the course of therapy associated with PHQ change?

Topic modelling

Diagnostics for MALLET

Chris Notes to self

Pipeline of work done

Relevant papers

SEMDIAL 2013: Zhou Yu, Stefan Scherer, David Devault, Jon Gratch, Giota Stratou, Louis-Philippe Morency and Justine Cassell. Multimodal Prediction of Psychological Disorder: Learning Nonverbal Commonality in Adjacency Pairs:

SIGDIAL 2013: David DeVault, Kallirroi Georgila, Ron Artstein, Fabrizio Morbini, David Traum, Stefan Scherer, Albert (Skip) Rizzo and Louis-Philippe Morency. Verbal indicators of psychological distress in interactive dialogue with a virtual human

PLOS ONE 2012. Visualising Conversation Structure across Time: Insights into Effective Doctor-Patient Consultations. Daniel Angus , Bernadette Watson, Andrew Smith, Cindy Gallois, Janet Wiles

PLOS ONE 2014. Predicting the Risk of Suicide by Analyzing the Text of Clinical Notes. Chris Poulin, Brian Shiner, Paul Thompson, Linas Vepstas, Yinong Young-Xu, Benjamin Goertzel, Bradley Watts, Laura Flashman, Thomas McAllister

PLOS Comp Biology 2013. Getting More Out of Biomedical Documents with GATE's Full Lifecycle Open Source Text Analytics. Hamish Cunningham, Valentin Tablan, Angus Roberts, Kalina Bontcheva.