SentTrack uses sentiment to detect bottlenecks in GitHub issues
New framework analyzes 9,000 issue threads, finding 49% stagnate
SentTrack is a novel framework from researchers Xinyu Hu, Ali Behbahani, Daniel Moon, Yaren Dogan, and Nasir U. Eisty that applies sentiment-driven analysis to GitHub issue threads to identify where development gets stuck. Unlike traditional code-metric tools, SentTrack looks at conversational dynamics. It has two pipelines: a horizontal pipeline uses an LLM to summarize raw issues, extract concern phrases, and cluster them via UMAP and HDBSCAN — producing 613 semantic clusters from the first 3,608 issues. The vertical pipeline applies the ABCDE collaborative interaction framework to classify each comment and infer thread outcomes.
Across roughly 9,000 threads from the AvaloniaUI open-source repository, SentTrack found stark metrics: 49% of threads ended in stagnation, only 13% reached resolution, and the resolution gap emerged as the dominant bottleneck signal. A weighted scoring engine combining negativity, stagnation, resolution gap, and thread length gives maintainers an interpretable way to prioritize high-friction discussions before they derail development. The paper is published on arXiv (2606.11476) and represents a shift toward understanding the human factors behind stalled software projects.
- Analyzed ~9,000 issue threads from AvaloniaUI using sentiment + LLM summarization
- Found 49% stalled and only 13% resolved — resolution gap is the key bottleneck
- Horizontal pipeline used UMAP and HDBSCAN to produce 613 semantic clusters from 3,608 issues
Why It Matters
Gives maintainers an early-warning system for toxic or stalled discussions before they block development.