Mitigating Rater Bias in Social Network Analysis: A Multi-Threshold Robustness Testing Framework for Reliable Risk Identification

Social Network Analysis (SNA) has been widely applied to risk identification research. However, two key constraints, namely rating bias and the subjectivity of threshold selection, undermine the reliability and reproducibility of analytical results. To address this di-lemma, this study aims to construct a standardized and robust analytical framework for SNA-based risk identification. The core research objectives are as follows: elucidate the differential impact mechanism of threshold variation on the macro-topological structure and micro-level node ranking of risk networks, examine the cross-threshold robustness of core risk node rankings, and delimit the effective threshold range for stable risk identification. Accordingly, to fulfill the above objectives, this study proposes a multi-threshold robustness inspection method based on individual rating patterns, and conducts systematic empirical analysis with industrial projects in the post-support period of reservoir resettlement as research cases. The results indicate that threshold variation exerts marked systematic effects on the macro-topological structure of risk networks, whereas the relative rankings of core risk nodes remain robust. The effective threshold range for risk identification in such projects is α ∈ [0.1,0.3]. This study provides a repeatable quality control framework for SNA-based risk identification, with favorable cross-domain transferability.