Mining Controller Inputs to Understand Gameplay
summary
Analyzes raw controller streams from Super Mario Bros. 3 (91 stages, ~8 players, ~15 hours) by turning 30-frame sliding windows into "input words" and applying LDA then a new PGA model that separates level gameplay from player style. Discovers about 12 gameplay types per level, detects water levels and play-choice signatures (frog suit), and trains a recognition system that achieves 90% accuracy in 20 seconds of play, showing the method works for categorization and player ID.
AI-written summary · about
chapters (8)
- 0:00Start
- 0:13Categorizing Video Games
- 4:06Super Mario Brothers 3
- 7:02Topic Modeling
- 10:47How Do We Categorize the Game Independently of Playing Styles
- 11:24The Player Gameplay Action Model
- 14:49Player Recognition System
- 19:28Emergent Complexity
related talks
53mBalancing gameplay of aRPG using player behavioral patterns
16mUser Guided Movement Analysis in Games using Semantic Trajectories
06mUnderstanding Player Patterns by Combining Knowledge-Based Data Abstraction with Interactive ...
59mPlaying Video Games One Frame at a Time
30mLeveraging Sentiment and Topic Detection Analysis for Decoding Player..
15mAssessing the Impact of Visual Design on the Interpretation of Aggregated Playtesting Data ...
06mGAME: Game As a Measurement Environment - Scheme to Evaluate Interfaces and Game Contents Based ...
03mDance and Mind Control in The Lands Between. Video-based Study-in-progress of Alternative Game In...