Beyond Sub-Gaussian Detector Scores: Robust Weighted Profile-Loss Change Point Detection for Human-LLM Text Segmentation

2026-09-29

AI Summary

This study addresses the challenge of author boundary localization in mixed human-machine documents caused by varying detection score reliability. We propose Robust Weighted Contour Loss Change Point Detection (RWCP), which introduces truncated weights and Huber gains to suppress extreme score interference. The overall discrepancy between true and misaligned boundaries is modeled as a merge cost to circumvent nonlinear closed-form solutions, while quadratic loss limits are derived to recover weighted CUSUM properties. Precise segmentation is achieved through narrow-interval threshold search and shared-source centroid decoding. Experiments demonstrate that RWCP reduces WindowDiff by 17.6%, and its enhanced variant, RWCP-R, significantly improves boundary recovery accuracy for isolated change points.

Abstract

Mixed human-LLM documents require locating authorship transitions from detector scores whose reliability varies across text units. Existing weighted mean contrasts are vulnerable to extreme scores, while directly replacing means with robust centers obscures how a misplaced boundary changes the population objective. We propose Robust Weighted Profile-Loss Change Point Detection (RWCP), which combines capped reliability weights, Huber profile gains, and narrowest-over-threshold search in reliability coordinates. Our key analysis expresses the population gap between a true and a displaced split as a merge cost, avoiding a closed-form solution for the nonlinear center of a mixed segment. Under explicit curvature, spacing, and dependence conditions, core RWCP recovers the number of changes and localizes their boundaries; its quadratic-loss limit recovers squared weighted CUSUM. We also study RWCP-R, a separately evaluated decoder that shares source centers across nonadjacent passages. Across five retrospective cached-score benchmark families, core RWCP reduces family-macro WindowDiff by 17.6\% relative to weighted change-point detection, and RWCP-R lowers it further. Boundary recovery improves most clearly for isolated changes, while both fixed configurations miss changes in collaborative and densely alternating text.

Problem

• change point detection
• human-LLM text segmentation
• authorship transition
• detector scores
• robustness

Innovation

• Change Point Detection
• Robust Weighted Profile-Loss
• Human-LLM Text Segmentation
• Huber Profile Gains
• Merge Cost