SenFlow: Inter-Sentence Flow Modeling for AI-Generated Text Detection in Hybrid Documents

SenFlow is a new model for sentence-level AI-generated text detection (S-AGTD) in hybrid documents, addressing limitations of existing methods by incorporating inter-sentence dependencies through a structured prediction approach. It leverages a graph-based inter-sentence propagation mechanism and linear-chain CRF decoding, achieving state-of-the-art results on the MOSAIC benchmark, which consists of 16,000 hybrid documents generated by DeepSeek-V3.2 and Kimi K2. This advancement is crucial for practitioners as it enhances the detection of AI-generated content by considering contextual relationships between sentences, thereby improving the robustness of hybrid document analysis.