Getting in Tech with 2019’s UNI.CORN+s
Imagine this — you have a question about a product, so you launch a customer service chatbot. You ask your questions, but the chatbot is just not getting it. You start to get frustrated, but it’s like the chatbot knows it’s not able to help and automatically directs you to a live chat with a customer service officer. The customer service officer promptly answers your question. Problem solved. This was the winning solution that the Joy Chatbot team from the UNI.CORN+ internship programme came up with.
The existing DBS corporate banking virtual assistant, Joy Chatbot, has wowed in its interactions with customers, having been recognised in the Global Finance’s Innovators Award 2019 and The Banker — Tech Project Awards 2019. The UNI.CORN+ Joy Chatbot team were tasked to innovate with sentiment analysis to make Joy more ‘human’, giving it the capabilities to ‘empathise’ with customers and deliver more personalised experiences. The team’s solution came out tops on Demo Day, where every UNI.CORN+ team had to pitch their ideas to the DBS business teams.
We caught up with Tedmund, Su Yee, Darren and Vincent from the team to find out more about their UNI.CORN+ experience and how it all came together.
Hey guys, congrats on your win! Tell us more about how you guys came up with your solution.
Tedmund: At the start we decided to just see what’s out there. There was a lot of trial and error, and we tried about 11 models before we landed on the one we eventually used. We experimented with whatever we could find, constantly looked for a better model, and examined how we could improve on it together.
Vincent: Yea, we focused on looking at state-of-the-art results and traced back to how they were achieved. We then followed up with further research to figure out how to adapt from the successful models.
Darren: And we had to learn a lot of things too. I personally did not know anything about machine learning before the internship, and I’ve had to learn it from scratch.
Your project is all about improving the chatbot experience. What were your experiences or impressions of chatbots before embarking on this project?
Tedmund: Previously, the only time I would use a chatbot was probably after customer service hours. I felt that chatbots don’t necessarily understand what you need all the time. I would rather speak to a human. If I shout through the phone, they can understand that I’m angry. Chatbots that I’ve encountered have not been able to understand my frustrations.
Vincent: I believe that artificial intelligence (AI) has to be as good as a human intelligence. I think that’s the main reason why the adoption of chatbots is generally low, because the interactions that take place are not as seamless and accurate.
That makes a lot of sense! So that’s what you guys set out to do — make the chatbot ‘human’ using sentiment analysis. Why is it important for us to humanise human-to-AI interactions?
Tedmund: The whole point is to automate what humans can provide our customers — so we needed to make our AI as human as possible. Sentiment analysis is a way for us to inject some form of empathy or feelings into our AI. The goal was to not simply deliver a result-based solution to our customers — we also want to deliver a chatbot of choice with soul and empathy.
So how will your sentiment analysis solution improve future customer interactions with chatbots?
Tedmund: One of the major ways that sentiment analysis could come into play is when the customer is unhappy. It can pick up on that, and direct customers to live agents who can step in and resolve problems more swiftly. Another one is to look at the positive sentiments and think on how we can replicate them to ‘wow’ our customers even more! So the bot becomes more intuitive and active, rather than in the past when it was more passive.
That’s really cool, but also sounds like a hefty task. Did you face any challenges?
Su Yee: Of course! For example, there were a lot of training data in the first stage that were messy and incorrectly labelled. This meant that our engine had a hard time recognising what was positive or negative. We just had to manually examine 40,000 sets of data.
Darren: For me, I had to learn Python, which was a whole new programming language for me. I used C++, JavaScript and HTML/CSS in school. Python was initially challenging at times because the naming conventions were very different.
Tedmund: Likewise, my school didn’t teach a lot of the skills that we picked up during this internship, such as machine learning. Even for Python, it was just the basics. So coming here and understanding all the machine learning codes was definitely a challenge.
Vincent: I think the hardest thing for me was version control. As with every project, there are many iterations and we were all training different models. We started naming our files “bestmodel”, “bestbestmodel”, and then “finalbestbestmodel”. Eventually we all agreed on a naming convention and things got better.
Now let us talk about your experience with UNI.CORN+. How has the past 10 weeks been like for you guys?
Darren: It was challenging right from the start. For example, just the second week into the project, we were thrown curveballs like “what about double negatives?” or “can you also do this in Chinese?”.
Tedmund: Yea, we weren’t hand-held and told what to do. We were given a dataset and had the space and opportunity to explore and discover on our own. But our mentors were very supportive along the way.
Vincent: The mentors’ perspectives were also valuable. As a student with a tech background, we may not know how things work on the business side. For example, whenever we had a presentation, they would sit us down and guide us on how to present our ideas in a way that would make most sense to non-tech folks in the business.
Su Yee: I also really liked the environment at DBS Asia X because it was very casual and conducive. We also weren’t restricted on how to work on our projects and were offered lots of support.
Tedmund: The team was very welcoming. Right from day one, they were already very warm and reached out to help us. The result is something I’m proud of, as we came up with the idea, improved on it and made it truly ours.
You guys are graduating from UNI.CORN+ soon. What is your fondest memory of UNI.CORN+?
Tedmund: For me, it would be presenting what we had during to the Institutional Banking Group. We didn’t expect our product to go far, and Bayu, our mentor, was also really proud of our work, even promoting it on his own LinkedIn. That was really something that moved me.
Darren: We were invited to a party in June, even before our internship started. As we were getting to know everyone, Bayu said to us, “my job as a boss is not to just push our projects, it’s to make everyone able to shine to on their own.” And that was the moment that stood out the most to me.
Vincent: I agree, working with Bayu was also the most memorable for me. He’s a really good mentor who pushes us and sets clear objectives, but at the same time gives us the autonomy to pursue our own solutions. Most importantly, he takes pride in us and in our work.
Su Yee: I also really liked the DBS Asia X space that we worked in, which gave us a good balance of work and play. I previously didn’t know how to play ping pong. After training with the three of them, I’m still no expert, but I think I can beat some people now.
Finally, would you recommend UNI.CORN+ to your peers?
Tedmund: I actually got to know about UNI.CORN+ through my senior who took part in the programme previously. I’ve had such a positive experience since my first day here, so yes, definitely!
Vincent: Yes! It is a really unique experience where you can get exposed to both the business and tech in a company. It’s not something you get to experience very often.
Darren: I think some of my classmates would not be inclined to try for the internship because it’s outside of the usual field and they would rather stay in their comfort zone. But it’s something I would recommend to them because they get to learn more.
Su Yee: I agree, I’ll also recommend to my friends and peers as a challenge to push themselves out of their comfort zones.
UNI.CORN+ is a 12-week summer internship programme that gives university students practical experience. This year, UNI.CORN+ took on a new focus on technology to explore how tech can be used to redefine the world of banking for the future. More than 60 participants were shortlisted from over 800 applicants to take part in the Idea Smash hackathon final selection, with 28 graduates eventually making the cut. Find out more here.
