Premise

Everyone needs a friend to fill the need of love and belonging, but not everyone has a friend to overcome loneliness. What if a virtual bot can become the friend that you need? – Inspired by Maslow (1943)

Synopsis

Loneliness and social isolation have caught my interest over the past year, where the pandemic flooded the world and people have become more distant to each other. Talking from my own experience, I went from a busy and fast-paced life where work, friends and personal growth were central, to a more isolated and silent environment. The news has been dominated by the developments of the virus, but the effects on society are hardly discussed. Though, social isolation has increased tremendously over the past year, especially among young adults. A chatbot can offer the solution, to offer a listening ear or a bit of distraction. By exploring and experimenting with different prototypes a chatbot has been created that serves as a listening and interacting friend, on a daily basis.

Topic

The topic of this project is about social isolation. In the Netherlands there are more than 200.000 young adults, between 18-30 years old, who feel lonely or isolated (Eenzamejongeren.com, 2021). And more than 46 per cent occasionally experience feelings of loneliness. During the Corona crisis this statistic increased to 69 per cent (Nederlands Jeugdinstituut, 2020). I learned that many people around me are experiencing social isolation and that online interaction has become the main source of communication. It has brought me to the idea of creating a chatbot. I remember talking to a chatbot ages ago on MSN, back then a popular communication platform. I started a conversation with the chatbot regularly. Almost a decade later, I am interested in exploring new technologies and how these can create social interaction with users.

Research Insights

To get a better understanding on chatbots and how it has been used by other researchers or institutions, I have done research on the topic of chatbots in relation to mental health. An article was found that showed the relationship between virtual agents, which include chatbots, and mental health. The findings are based on real experiences. For the experiment, the virtual agents act as role players in different applications, with the goal to guide the user to manage their emotional responses (R. Baumeister, 2019). The agents also try to prevent these mental health issues by identifying different responses when having a conversation with the user.

There has been evidence that implies that virtual agents are able to provoke more honesty and openness than when the user is interacting with a human. The outcome showed that topics with little sensitivity were easily talked about with humans, but young adults preferred to talk about more intimate topics with a virtual agent, because it lacks the ability to judge (M. Pickard, 2016).

Approach

The research has been done with the help of a chatbot made in Microsoft Power BI and Microsoft Azure QnA maker. In addition, visualisation have been made in AdobeXD. The experiments have been tested on users between the age of 18-25 years old. The aim was to create a chatbot that could make the user feel at ease by being interested in or by asking questions about the situation of the user.

The Process

The process started with research on other platforms and chatbots. I downloaded the mental health chatbot, Woebot from Facebook, which gave insights on the approach on the user. Woebot has been a helpful app, since it also targets young adults. It gave me an idea of how this group can be approached. Moreover, another study shows that user experience and usability are essential to consider to make the user feel comfortable (M. Pickard, 2016). By researching and exploring existing chatbots, I started my first prototype. 

IBM Watson

I started my first prototype in IBM Watson, because I had followed a workshop on this tool and with little experience in coding, it seemed like a good option. However, I quickly realised that the options were too limited for the goals I had in mind. So, I decided to explore other options.

Microsoft PowerBI

This tool was used to create an actual working prototype. With the approach of Woebot and tips from Healthline on how to get to know someone, I started the first conversation. Essential factors are that the user feels at ease and to make it personal, therefore  the chatbot has been given a name, Abby, and tries to get to know the user.

Greetings

After struggling with finding the right sequences, I found out that options with Power BI were too limited to ask open ended questions. So answer buttons have been added to create a flow and guide the conversation.

Feelings

Keeping in mind that the user is feeling socially isolated, I wanted to show that the chatbot is able to listen to the user. By listening the chatbot creates an understanding of the situation. The following video shows the tiers: ”Happy” and ”Sad”. Since there are different levels of mental health issues (M. Joshanloo & M. Nosratabadi, 2008), I found it important that there were different approaches to every level of emotion.

Happy

Sad

While experimenting with the chatbot, I realised the chatbot would serve a better purpose if it was able to learn from the user and become its friend. I first conducted a user test, to see if I was heading in the right direction, before experimenting with other tools.

User tests

For the user test, a tech-savvy person has been asked to test my chatbot. Before starting I asked the user to follow the ”sad tier”, to pay attention to user experience and usability, and expectations and evaluation.

Findings

• The user was pleasantly surprised by the guidance of the conversation.
• The user found that some answers of the chatbot were not entirely matching the conversation.
• The user mentioned that the design was quite boring, and thought that adding colour would positively influence the usability.
• The user wanted to see different recommendations.

Adjustments

To implement the feedback of the first user test, I decided to make a visual design in AdobeXD. Since I have little prototype experience, it seemed like a good challenge to create an idea of an interface. I searched which colours evoke which emotion with users. I decided to use light blue as the main colour, to allow the user to enter a calming environment. In addition, a friendly face was added to make the conversation more personal. An overview of the prototype in AdobeXD can be found here.

Because the user test showed that the conversation was not flowing well enough, I decided to create a more in-depth conversation with the user. More answer options are added and the user is able to answer more freely.

Microsoft Azure QnA

Using Power BI is very time consuming, because all entries have to be put in manually. So for the third experiment, I used Microsoft Azure QnA maker, which makes use of AI to learn and interact with the user. The tool is a little bit more complicated, but allows me to insert an API, which will generate a conversation for me. I used a website URL which contained questions and answers about a person’s mental health.

Result:

• When not using the right API, responds can be very wordy and not unfiltered, which makes responses inaccurate.
• The knowledge based can be customised, so by observing the target group for a longer time and finding out FAQ’s, an API can be made.
• Microsoft QnA allows to make use of (endless) chit-chat, which allows the users to chat for hours.
• The tool can be trained and will learn from users responses and will adjust to the user, this means higher usage is better understanding. Therefore, the chatbot will become better at being a friend.

Reflection and Conclusion

In conclusion, I learned that users are enthusiastic about the idea of having a Friendbot, someone who the user could talk to everyday, which was not to initial idea of this project. The different technologies that I experimented with have given me interesting insights. The most useful tool of this project appeared to be Microsoft QnA, with the help of the knowledge base, interesting conversation can take place. It is less time consuming, because the tool is able to learn and train itself, unlike Microsoft Power BI, where every conversation has to be put in manually. I also learned that secondary features, such as design and appearance are just as important for the user, because it has influence on their emotion. Furthermore, I looked into Microsoft Azure Bot Services and GPT3, which are tools that offer these features, but are also subscription tools. These tools seem like the perfect addition to create and finalise the chatbot tool.

For further research, I would recommend experimenting with the LUIS Portal of Microsoft, it needs beginner coding experience but allows a higher degree of Machine Learning. Less specific questions and answers can be used for the chatbot to still understand the user. Moreover, further research on the target group, for example, with the help of user observation the knowledge base can be customised to the needs of the user.

Bibliography

Maslow, A. H. (1943). A theory of human motivation. In A. Maslow, Psychological Review (pp. 370-96).

Stichting Join Us. (2021, 2). Jong en Eenzaam. Retrieved from Eenzamejongeren.com: https://www.eenzamejongeren.com/jongeren/eenzaamheid/

Nederlands Jeugdinstituut. (2020). Veel jongeren voelen zich eenzaam door coronacrisis. Utrecht : Nederlands Jeugdinstituut.

Baumeister, M. L. (2019). The need to belong: desire for interpersonal  attachments as a fundamental human motivation. Psychological Bulletin, 497-529.

Pickard, C. R. (2016). Revealing sensitive information in personal interviews: Is self-disclosure easier with humans or avatars and under what conditions? Computers in Human Behaviour, 23-30.

Cuijpers, I. M. (2009). Computer-aided psychotherapy for anxiety disorders: a meta-analytic review. Cognitive Behaviour Therapy, 66-82.

Joshanloo, & M. Nosratabadi. (2008). Levels of Mental Health Continuum and Personality Traits. Springer Science + Business Media BV.