Sprint2Vec in the ASE 2025 Journal-First track

Our IEEE TSE paper “Sprint2Vec: A Deep Characterization of Sprints in Iterative Software Development” was accepted in the Journal-First track of ASE 2025. Sprint2Vec builds vector representations of Agile sprints from structured data, such as sprint reports, issue attributes, and developer profiles, and from text, such as issue descriptions and sequences of developer activity. Unlike earlier methods that predict only how much a sprint delivers, it predicts both productivity and quality, where quality means the risk of reopened issues. The evaluation covered nearly 8,000 sprints, 118,000 issues, and 3.7 million developer activities across five large open-source projects, where Sprint2Vec beat baseline and state-of-the-art approaches.

• Sprint2Vec uses deep learning, including LSTM and BERT-based models, to learn features from issue descriptions and developer activity.
• It gives interpretable insights into the factors related to sprint outcomes.
• Morakot Choetkiertikul presented the paper at ASE 2025 in Seoul in November.