Creating Flexible Human-Agent Dialogues for Simulation-Based Training of Health Coaches

From Master Projects
Jump to: navigation, search

About Creating Flexible Human-Agent Dialogues for Simulation-Based Training of Health Coaches

  • Contact person: has supervisor::Tibor Bosse
  • Contact person: has supervisor::Aart van Halteren
  • This project has been fulfilled.
  • This project fits in the following Bachelor programs: {{#arraymap:|, |xXx|bachelorproject within::xXx|,}}
  • This project fits in the following masterareas: {{#arraymap:Knowledge Technology and Intelligent Internet Applications, Computational Intelligence and Selforganisation, Cognitive Science, Human Ambience, Technical Artificial Intelligence, AI and Communication, Internet and Web Technology, Multimedia, Information Sciences|, |xXx|project within::xXx|,}}


|free text=}} Background

One of the interests of Philips Research is to explore and develop innovative interventions to help people maintain a healthy lifestyle. A particular way to achieve this is by means of health coaching, a process that facilitates healthy, sustainable behaviour by challenging patients to identify their personal values and transform their goals into actions. Health coaching involves a conversation between a coach and a patient, in which the former listens actively to the latter, makes an analysis of his or her personal situation, and provides advice in the form of concrete steps to realise the healthy behaviour. As a starting point for such conversations, the coach usually has access to an individual 'profile' of the patient, which has been established in advance based on a variety of data items related to the patient's personality, cognitive/emotional states, and environmental factors.

Simulation-Based Training

The task of a health coach is not easy. Coaches need to master a number of skills, including the ability to establish a positive relationship with the patient, to act as a motivator, to create an adequate vision on how the patient should improve his or her wellbeing, and to make sure the patient actually implements this vision. An effective way for coaches to improve these skills is by training. However, training actual conversations with real humans is costly, time consuming, and difficult to set up. Instead, simulation-based training is an attractive alternative. The idea of simulation-based training is that the health coach is engaged in a conversation with a virtual patient, i.e. a human-like computer character in a virtual environment.


To offer the health coach an optimal learning experience, the virtual patients involved in the training environment should be believable: that is, they should behave similar to how real patients would behave. For instance, if the patient is stressed or depressed, this emotional state should be reflected in the (verbal and non-verbal) behaviour of the character. In a previous collaboration between Philips and VU, a prototype of such a virtual patient has been developed, enabling the human user to engage in a dialogue upon a particular aspect of healthy behaviour. This dialogue system is based on the notion of conversation tree: a tree structure that represents all possible developments of a conversation in an exhaustive manner. Although this method is easy to use, it is not considered as very flexible. In particular, the use of conversation trees makes it is difficult to endow virtual conversation partners with believable behaviour without explicitly specifying their response to each possible series of inputs they can receive.

Aim of this Project

A promising alternative to using conversation trees is to create dialogues based on cognitive models. By endowing virtual characters with cognitive models, their mental state (e.g., in terms of their beliefs, desires, or emotions) can be maintained over time during a conversation. Subsequently, this mental state might serve as a mediator to shape the agents' (verbal) responses to the new inputs they receive. Based on such an approach, it is possible to develop a larger variety of training scenarios with limited effort. Hence, the main goal of this project is to extend the existing virtual patient (based on conversation trees) with a cognitive model, thus creating more flexible human-agent dialogues for simulation-based training of health coaches. Ideally, the solution should be set up in a generic manner, enabling it to be re-used in other domains involving human-agent dialogues.

Organisational Issues

- The Master project is expected to last about 6 months.

- As part of the project, an intership at Philips Research may be included.

- For more information, please contact Tibor Bosse ( or Aart van Halteren (