Enabling a positive in-game chat experience for HiberWorld.
HiberWorld is looking for its creative and social gaming platform to facilitate even more positive engagement whilst ensuring a safe environment for the players.
“A community wants to be social: People want to share content, chat with their friends, send messages back and forth, comment on games, ask questions and have fun.”
Sean Kauppinen, CSMO at Hiber.
Over time, users will also want to talk to new people they don’t know, which means they’ll be interacting with strangers. To ensure a pleasant experience for the users, a platform holder needs to, beyond the typical privacy and legal compliance issues, make rules on what kind of communication and behavior is acceptable and what is not.
“You also need to decide if people can share personal information, e.g. passwords, phone numbers, address information or recently as an emerging behavior, sharing their Zoom room numbers” Kauppinen adds.
The challenge is how to enforce those rules and guidelines with real-time chats in high volumes in multiple languages. It’s also business-critical, as the unsavory behavior doesn’t only reflect badly on the other users, but it also reflects negatively on the platform.
Hiber’s approach to enforce their guidelines was to work in stages. In the initial stage they put in place a general filtering for e.g. bad words, passwords, addresses, phone numbers (and most common abbreviations related to those words). This worked to an extent, but they knew people are clever in finding ways around the rules. So Hiber had to be more efficient. They needed to analyse the actual meaning of the messages instead of just using lists of not suitable words to ensure a positive chat experience.
Piloting with Utopia
Hiber started a pilot with Utopia during which Utopia built a tailor-made advanced AI moderation model for them. The challenge Hiber had was that it didn’t have training data, i.e. examples of previous human moderator decisions on good and bad chat messages according to their own community policy, to build the machine learning model on. Utopia’s Snowball Process was used to train the model.
“It has been really great working with Utopia Analytics. We had no data, no previous moderation history, we didn’t even have a tool to moderate the chats. In just 5 weeks, starting literally from scratch, Utopia created us a very accurate AI moderation model using what they call a “Snowball Process”, our effort in the process was less than 6 work days.”
“We knew that in-game chat would require AI-based pre-moderation to keep players safe and happy,” says Michael Yngfors, CEO at HiberWorld. “How easily and fast a high-quality in-game chat moderation model can be developed with Utopia surprised us in a very positive way”, he continues.
Integration done on a Monday afternoon
“Integration of the Utopia API was a breeze, it was actually done in one afternoon. Utopia even advised us on building our own moderation tool for human moderators to monitor traffic and warn or ban users, if needed,” says Kauppinen.
HiberWorld is a Swedish innovative game start-up. Its social game platform lets anyone create, play and share their games. There are thousands of new games built by the community every week. HiberWorld is free to play and can be played on pretty much any of the 1.4 billion devices with a web browser, including mobile phones, tablets and PC / MACs. The beta was quietly released in 2019, and the society is growing quickly.