Selene Knowledge Discovery: The Interface Effect

Debbie Denise Reese · Ben A. Hitt

Wed., June 10, 5:00–7:00, Great Hall (4th floor, Central)

The application of knowledge discovery techniques to Selene gameplay data and subsequent follow-up with descriptive and inferential statistics revealed evidence of an interface effect independent of player demographics. Regardless of player age, grade in school, gender, human capital, race, or self-reported academic achievement, the interface element designed to lead players toward the game goal worked as intended. Results support retention of this element in subsequent game versions.

Selene: A Lunar Creation GaME is an online, single-player game. Players form the Earth’s moon. The environment enables player discovery of science concepts (accretion, differentiation, impact cratering, and volcanism) through procedural gameplay. Selene collects all player gameplay data, as well as self-reported self-perceptions of skill and challenge known as flow (Csikszentmihalyi & Schneider, 2000).

During the accretion module, particles from a giant impact orbit in a ring around the early Earth. Players must select just the correct proportion of high-, mid-, and low-density particles to form the early moon. Players must also slingshot those particles into the existing proto-moon with just the right velocity to make them accrete yet impact with enough energy to cause heating and melting. The interface directs the player through a “best choice” gauge that displays a high-, mid-, or low-density particle based upon current proto-moon composition.

We suspected player demographics might influence patterns of Selene gameplay. The Self Organizing Map (SOM) technique provides a means of discovering patterns of gameplay and subsequently allows for association of demographic data with those patterns. The process makes no a priori assumptions. Normally distributed data will spread out across the entire map provided that the variable set are more or less equivalent in discriminatory power. This was not the case with Selene data. The data gravitated toward two edges of the map leaving a band of centroids to which no exemplars matched down the center. This indicated suggested the presence a single strong classifier among the exemplar variables.

We investigated this through hierarchical clustering. Hierarchical clustering looks at the ability of each variable to segment the data into two groups. After the initial segmentation, each reaming variable is similarly assessed. This analysis showed that choice variable strongly segmented the data. When the choice variable was removed from the variable set present to the SOM, the data spread across the map as would be expected. Analysis of demographics with respect to SOM cluster membership revealed no clear relationship of demographics to gameplay. In other words, players’ performance and flow reports are independent of player demographics.

Players (n=96, primarily aged 13–18, with 2 adults) followed the interface prompt by selecting the best choice 70.5% of the time.

A repeated measures analysis of variance using players’ choice mean score aggregated at four sub-segments indicated choice performance improved significantly over time F(3,81)=10.7, p<.001, partial eta squared=.28. Performance gain was greatest during subsegment 2. Performance fell slightly during sub-segment 3 and rose to its highest level during sub-segment 4, but neither change was significant.

References

Csikszentmihalyi, M., & Schneider, B. (2000). Becoming adult: How teenagers prepare for the world of work (1st Ed.). New York: Basic Books.