Avachat – A virtual autonomous agent for supported self management in COPD

Avachat was a co-designed artificially intelligent virtual agent to support self-management in individuals with complex physical comorbidities.  The app was co-designed via two workshops comprised of patients, health professionals and computer scientists. The first workshop explored challenges of daily living and ways in which these challenges could be overcome with everyone critiquing existing avatar examples and together designing an ideal avatar.  In the second workshop scenarios were role-played in order to develop content for the system. Finally, a proof-of-concept implementation was conducted, which was used in a video-based scenario testing of acceptability with older adults with a diagnosis of COPD and health professionals connected with this patient group’s care in order further development of the prototype.

Development of Avachat

Development of the prototype virtual agent was carried out by me.  The agent combined emerging technologies to create a seemingly autonomous agent.  The agent had the ability to convert spoken interactions to text through a speech-to-text (STT) service.  A sentiment service based on SentiWordNet was also present capable of carrying out analysis on the input text and generating positivity / negativity sentiment scores.  These scores could be collected, monitored and used to help determine the agents best possible responses over time.  Responses from the agent were then returned via a text-to-speech (TTS) service.

Avachat had two modes free text (conversation mode) and wizard (branching path mode).  In branching path mode a custom json based parsing language was interpreted by a state machine loop.  The parsing language was comprised of written modules that could be triggered in specific scenarios.  The agent was capable of diagnosis via this process by collating responses and carrying out actions as a result.  For example the COPD crisis module would attempt to help the user get their breathing under control before calling for medical assistance if the module failed to resolve the situation.

Publications
Easton K, Potter S, Bec R, et al. A Virtual Agent to Support Individuals Living With Physical and Mental Comorbidities: Co-Design and Acceptability Testing. J Med Internet Res 2019;21(5):e12996. doi: 10.2196/12996. Link to Paper

Project team:
Principal Investigator: Mark Hawley
Centre for Assistive Technology and Connected Healthcare (CATCH), University of Sheffield, UK
School of Health and Related Research (ScHARR) University of Sheffield, UK,

Co-Investigators:
Katherine Easton
Centre for Assistive Technology and Connected Healthcare (CATCH), University of Sheffield, UK
School of Health and Related Research (ScHARR) University of Sheffield, UK,

Heidi Christensen
Centre for Assistive Technology and Connected Healthcare (CATCH), University of Sheffield, UK
Department of Computer Science, University of Sheffield, UK

Stephen Potter
Centre for Assistive Technology and Connected Healthcare (CATCH), University of Sheffield, UK
School of Health and Related Research (ScHARR) University of Sheffield, UK,

Scott Weich
School of Health and Related Research (ScHARR) University of Sheffield, UK,

Luc de Witte
Centre for Assistive Technology and Connected Healthcare (CATCH) and University of Sheffield, UK

Dan Wolstenholme, Cheryl Grindell and Remi Bec
NIHR CLAHRC Yorkshire and Humber

Matthew Bennion
Centre for Assistive Technology and Connected Healthcare (CATCH), University of Sheffield, UK
Department of Psychology, University of Sheffield, UK

Bahman Mirheidari
Department of Computer Science, University of Sheffield, UK