Predicting Preference with Sparse Data in Personalized Gamification via Deep Learning

Personalized gamification is a practice that is relatively well defined and improves the effectiveness of a gamified system. However, in practical application the improvement is not as significant as expected. The process of personalizing a gamified system is taxing and relatively unfeasible, with far too many aspects to consider to produce an effective result. Artificial intelligence, and neural networks, can step in to alleviate much of the work, but even still results are inconsistent at best. This project seeks to remove this inconsistency by attempting to personalize only one aspect of a gamified system, rather than the entire system as a whole. By attempting the personalization problem in this manner the amount of individual characteristics to consider is reduced dramatically, thus allowing for a neural network to more quickly and accurately determine personalization characteristics and apply them for any given user. Results show that an RNN can detect preference patterns and apply user preferences to a scheduling system. These results were produced with little run time and a more sparse dataset than normally expected for a neural network, which showcases the novel fact that detecting user preference does not require large datasets.