My Topics

1 February 2019

Developing An Emotional Support Chatbot Using the Knowledge of the Crowd

Chatbots, computer-automated conversation partners, have the potential to provide emotional support to stressed users. At the 16th International Conference on Practical Applications of Agents and Multi-Agent Systems, Lenin Medeiros and Tibor Bosse presented a study in which they used around 10,000 tweets to find out how people help each other to cope with stressful situations via online social networks. This knowledge will help to improve the supportive messages of an online chatbot.

Take aways

  • For different stressful situations, people send each other different online support messages, such as ‘take care!’ and ‘think positively!”.  
  • With this knowledge, an emotionally supporting chatbot can be developed that serves two purposes: 
  1. Identifying social media users with mild signs of everyday stress and helping them in an initial phase of their problem 
  2. Bringing social media users in contact with other users with similar problems

Study information

  • Who?

    Around 10,000 tweets 

  • Where?


  • How?

    The researchers collected tweets that contained negative sentiment in combination with keywords that indicated stressful situations. Using crowdsourcing (turning to a large public for input and advice) the researchers asked people to classify the collected tweets into categories. They also asked the participants to come up with supportive messages in reply to the tweets, which they also classified into categories. Then, the researchers explored which types of support were most frequently used in which circumstances.

Facts and findings

  • For tweets about death and health issues, people most often provided ‘empathic’ emotional support, supporting someone by expressing empathy, love, trust or caring (for example, ‘take care!’). 
  • For tweets about school and work issues, people most often provided ‘cognitive change’ emotional support, trying to change the other’s state of mind (for example, ‘think positively!’). 
  • For tweets about other stressful situations, people used a variety of supportive messages. 
  • With this overview, the online support messages send by a chatbot can be adjusted to match the support a human would provide.