View source for Using Sentiment Analysis and Human-Agent Dialogues for Aggression De-escalation Training

{{Projectproposal
|Contact person=Tibor Bosse
|Contact person2=Ward van Breda
|Bachelorprograms=Lifestyle Informatics
|Master areas=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
|Fulfilled=Yes
}}

|free text=}}
Employees with a public task, like police officers, bus drivers, and ambulance personnel are regularly confronted with aggressive behaviour. To prepare such professionals for these confrontations, dedicated aggression de-escalation training is required. This is usually done using role-play, but a drawback of this approach is that it is costly, time-consuming and hard to organise. Therefore, automated training based on conversations with virtual agents is an interesting alternative, which is currently being explored. This type of training makes use of dialogues between a human user and an aggressive virtual character. However, these dialogues typically make use of multiple choice menus, which limit the freedom of the user. To improve this, this project investigates the added value of sentiment analysis to create more natural dialogues with aggressive characters. To limit the scope of the conversations, the idea is that the user has restricted possibilities regarding the content of what she says, but is completely free with respect to the style of her utterances. For example, the message that a passenger has to buy a ticket can be formulated in a more empathic way (e.g., "I am very sorry, but I'm afraid you do not have any other option than buying a ticket") or in a more directive manner (e.g., "That's not my business, you just have to buy a ticket"). By using a sentiment analysis module, it is possible to develop a system that understands the style of the user's utterance. As a result, a virtual conversation partner can be developed that responds differently to different styles of communication (e.g., it calms down if you address it empathically, but becomes more aggressive if you address it too bluntly). This main goal of this project is to develop such a conversation partner in the context of a specific domain where aggression de-escalation plays a role (e.g., selling tickets to tram passengers).