Understanding Players as if They Are Talking to the Game in a Customized Language: A Pilot Study

TL;DR

This pilot study investigates the application of language models to model game event sequences, treating them as a customized natural language, and demonstrates the potential of self-supervised LMs in enhancing game design and personalization without relying on ground-truth labels.

Abstract

This pilot study explores the application of language models (LMs) to model game event sequences, treating them as a customized natural language. We investigate a popular mobile game, transforming raw event data into textual sequences and pretraining a Longformer model on this data. Our approach captures the rich and nuanced interactions within game sessions, effectively identifying meaningful player segments. The results demonstrate the potential of self-supervised LMs in enhancing game design and personalization without relying on ground-truth labels.

Figures (4)

• Figure 1: Example events segmented into semantic sessions. The final game-end event in "Session 1" is expanded to show details about its associated fields and values.
• Figure 2: (a) Histogram of session lengths and (b) the distribution of session quantities over a 15-day period shown up to the 99th percentile.
• Figure 3: The pipeline to convert event streams to word streams.
• Figure 4: (a) t-SNE of the latent embedding space from the pretrained large Longformer with Gaussian Mixture Model clustering. (b) Histogram of quantized player events in clusters (excluding cluster 8 due to small size and lack of gameplay).