BERTopic-Based Public Opinion Topic Mining on Graduate Supervisors: A Case Study of Weibo

Abstract

Weibo public opinion on graduate advisor issues profoundly reflects the structural contradiction between institutional supply and student demands. This study conducted dynamic topic modeling on 187,238 relevant posts from 2017 to 2024 using the BERTopic model, revealing three core findings. First, the topic distribution exhibits significant dual‑track differentiation: academic routines and student mutual aid constitute a stable public opinion baseline, accounting for over 70% of the corpus, with high‑frequency terms such as "thesis" and "graduation" indicating resource competition pressures under the expansion of enrollment policies, while the "outbreak‑plunge" pulse pattern of extreme events confirms the short‑term nature of public opinion. Second, policy interventions follow a clear three‑stage response path; after the issuance of the Guidelines for Advisor Conduct in 2020, the symbolic phase of policy communication and the lagged accountability phase of institutional implementation highlight the risk of "symbolic governance." Third, cross‑topic analysis further exposes systemic contradictions: the infiltration of academic pressure into regional resource issues reveals inter‑institutional resource imbalances, and although the symbolic dissemination of extreme events prompts institutional reflection, it remains constrained by issue fragmentation. The theoretical contribution of this study lies in quantitatively verifying the correlations among higher education expansion, institutional change, and public opinion evolution. Its practical significance lies in providing data support for constructing an early‑warning mechanism for advisor‑student conflicts and optimizing the advisor authority checklist, while also innovatively confirming the effectiveness of BERTopic in short‑text public opinion modeling, thereby promoting a shift in educational governance paradigms from passive response to active intervention.