SentTrack Finds 49% of GitHub Issues Stall Before Resolution - Using Sentiment

Forty-nine percent of issue threads in the AvaloniaUI repository ended in stagnation; only 13% reached resolution. That's the headline finding from a new study of roughly 9,000 GitHub issue discussions, and it points directly at a blind spot in how teams monitor repository health.

Most tooling tracks code velocity, build failures, or label counts. It ignores the conversational dynamics that actually drive—or kill—a thread. SentTrack is a dual-lens framework designed to close that gap, and the numbers show why it matters.

Two Pipelines for Conversational Bottlenecks

SentTrack splits its analysis into horizontal and vertical pipelines. The horizontal side takes raw issue text, runs it through an LLM to produce clean summaries, then extracts mid-level concern phrases. Those get clustered with UMAP and HDBSCAN, yielding 613 semantic clusters from the first 3,608 issues processed. That gives maintainers a map of recurring friction themes.

The vertical pipeline applies the ABCDE collaborative interaction framework to classify each comment's role—assertion, building, challenging, defending, evaluating—and infers thread-level outcomes. Across the full corpus, the ratio of stagnation to resolution is the paper's strongest signal: nearly half the conversations simply stopped without a decision.

Why Sentiment Beats Label-Based Methods

Traditional label tracking is reactive. A bug label doesn't tell you if the discussion is spiraling. SentTrack's weighted scoring engine combines four signals: negativity from sentiment analysis, stagnation duration, resolution gap, and thread length. Together they produce an interpretable priority score for each thread. Early detection of negative sentiment can surface risk before anyone tags an issue "blocked."

What This Means for Maintainers

Open-source maintainers already drown in issue volume. SentTrack doesn't add another dashboard metric—it offers a triage lens tuned to social friction. The 49% stagnation number is a wake-up call: half of your issue threads may be silently dying. A tool that catches the tone shift before the thread goes cold gives you a chance to intervene.

For maintainers, this shifts the question from 'which issue is oldest' to 'which conversation is about to die.'

Source: SentTrack: Sentiment-Driven Bottleneck Detection in GitHub Issue Repositories
Domain: arxiv.org