Mind Reading to Predict the Success of Online Games

I Think He Likes It: Researchers in Taiwan recorded the electrical signals of muscles involved in positive and negative emotions when subjects played a new online game. They used it to predict whether people would find the game addictive.Photo: Multimedia Networking and Systems Lab/Institute of Information Science, Academia Sinica

On a first date, couples scrutinize each other’s facial expressions for a clue as to whether the date will turn into a long-term relationship. Game publishers and designers might start doing the same thing. By analyzing the movements of gamers’ smile and frown muscles in the first 45 minutes of play, Taiwanese researchers have found a way to predict a game’s addictiveness.

“Such forecast results might give game designers the green light to complete a new potential game or advise they drop a hopelessly doomed one,” says Kuan-Ta Chen (now known as Sheng-Wei Chen), an associate research fellow at the Institute of Information Science, Academia Sinica, in Taipei.

The online gaming industry sees a game that is played by a large number of fanatics and survives more than two years as a success, says Chen. But that success comes at a cost. Blizzard Entertainment, for example, reportedly spent 4.5 years and US $63 million to develop its popular online video game World of Warcraft, which was released in 2004. For upkeep and expansion, it invested tens of millions more.

Of course, not every online game makes it. According to Chen, more than 200 online games are released each year, globally. The cost of developing a game, jointly brainstormed by dozens of designers, ranges from less than $1 million to as much as $200 million. However, the humbling fact is that most games survive only four to nine months, says Chen. 

It’s difficult to evaluate an online game’s addictiveness prior to the release, says Chen. The gaming industry’s approach is simply based on designers’ intuition and experience and the feedback from focus groups, the latter of which could be limited and biased.

Chen’s team, composed of researchers at the institute and at the electrical engineering department of National Taiwan University, aims to help game publishers avoid risky or blind investments. Using archival game data and dozens of electromyography (EMG) experiments, they constructed a forecasting model that predicts a game’s ability to retain active players for a long time.

To do this, Chen says, the researchers had to sort out the relationship between the data from laboratory emotion studies and a game’s market performance during the first six months after its release. By analyzing account activity records of 11 games—five role-playing games, four action games, and two first-person shooter games—the researchers tried to produce a general addictiveness index. They came up with an index that takes into account things like how quickly players’ frequency of participation decreases during the subscription period in which the gamer actually played the game and found that the index correlated well with key measures of a game’s success, such as user focus group responses and the length of time players spent playing a game.

Meanwhile, in the lab, the researchers connected electrodes to 84 gamers, ages 19 to 34. The electrodes were set up to measure the electrical potentials generated by two facial muscles—the corrugator supercilii, or “frowning” muscle, whose motion primarily produces the appearance of suffering and unhappiness, and the zygomaticus major muscle, which is used in smiling and laughing. These facial EMG measurements were conducted for 45 minutes as players explored new games for the first time.

Each of the subjects was asked to play as many as three new games. In total, researchers gathered 155 hours of facial-expression data. From the data, they were able to discern positive and negative emotions. Analyzing those two separately and in combination, they were able to predict the games’ addictiveness index to within an average of 11 percent, Chen says. They reported their findings [PDF] at IEEE/ACM NetGames in November 2012. Both Taiwan’s National Science Council and Taiwan-based game developer Gamania Digital Entertainment sponsored the research.

Chen says the model his team created can be used to ensure that a new game’s design is on the right track in its early development stages. It can also help game operators assess the potential market value of a new game. Chen’s team plans to build more-sophisticated models by collecting additional emotional response data, such as heartbeat and galvanic skin response (the basis of lie detection). In addition, Chen says, they’ll expand the sample size to account for individual tastes.

Nicky Yeh, marketing manager of the Taiwan branch of Hong Kong–based Gameone Group, says that game publishers rely heavily on in-depth interviews with gamers in focus groups to predict a new game’s potential; thus the forecasting model created by Chen’s team might be an additional objective tool. “The 11 percent error sounds acceptable. It will help counterbalance some errors caused by human judgmental decisions,” Yeh says.

Gilbert Hsieh, senior technical director of Taiwan-based Gamania, says that for certain types of games with complicated plots it might be difficult to forecast the game’s potential based on emotional reactions of gamers engaged in a mere 45 minutes of play. “It’s hard to know whether viewers would like a new film if only a small part was shown,” Hsieh says.

Someday the process could even break out of the online gaming world. “We hope our forecasting addictiveness could become more mature and will be able to be applied to other genres of entertainment products, such as commercial films, pop music, and others,” Chen says.