Analyzing gameplay videos to support video game developers: from issue detection to game sessions reproduction

The gaming industry has seen significant growth in recent years. Video games present a unique challenge for testing due to their large number of states, which are reached through the different choices of players. Millions of players worldwide daily produce vast amounts of gameplay videos. Although such videos have significant potential as a testing resource since they might document video game issues, their use remains not widely explored in current research. This PhD thesis addresses this gap by proposing (i) strategies to automatically extract meaningful information from gameplay videos, and (ii) a technique to replicate the sequence of in-game actions. We introduce GELID, an approach that analyzes gameplay videos to identify game-related issues by extracting significant segments based on textual features. We observed that detecting performance and balance issues using textual features alone is challenging, which led us to develop targeted solutions. The first one is HASTE, which identifies performance issues through visual analysis, finding areas where stuttering events occur. Additionally, we conducted a study to capture players’ real engagement levels by gathering self-reported data from participants during gameplay sessions, providing a more accurate measure than previous reliance on facial expression interpretations. Finally, we present RePlay, a method that replicates player actions from gameplay videos to reproduce sequences leading to specific issues. Our findings demonstrate the value of automated gameplay video analysis in enhancing game quality assurance processes. Feedback from practitioners involved in our studies supports the relevance of these methodologies in improving the efficiency of game testing.