Jing Cao (Xiangtan U) et al. have posted “Multi-Dimensional Risk Identification and Dynamic Evolution Analysis of Generative Artificial Intelligence: A Multi-Source Heterogeneous Data-Driven Approach” on SSRN. Here is the abstract:
Addressing the research gap in systematic Generative Artificial Intelligence (GenAI) risk identification, this study constructs a multi-source risk corpus (e.g., official government documents, micro-blog). AI-enhanced methods (TextRank, Word2Vec) extract and expand a domain-specific risk lexicon. Employing LDA topic modeling, a multi-dimensional risk indicator system is developed from public and government perspectives. Critical risk points are identified and evolution pathways traced using significance metrics. Key findings reveal a public focus on social-level impacts versus government emphasis on technical standards; legal and ethical issues as pivotal tensions; emerging interactive effects of composite risks. This provides methodological references for governmental AI risk governance.