TSA: A Temporal Sequence Alignment Algorithm for No-Ground-Truth Traffic Data Labeling

High-quality network traffic data labeling is a critical task for network security, enabling precise, fine-grained representations of traffic flows. However, existing labeling approaches still face severe limitations in no-ground-truth network scenarios, including high manual effort, imprecise labeling, and an over-reliance on often-inaccessible auxiliary information. Consequently, high-quality labeled datasets remain scarce, significantly hindering the advancement of data-driven security techniques. Thus, we propose TSA, a Temporal Sequence Alignment algorithm designed for automated, fine-grained traffic labeling in no-ground-truth scenarios. TSA operates solely on packet sequences recorded in source-side logs and packet sequences captured from the network. By enforcing rigorous payload, delta-time, and chronological order constraints, the algorithm automatically aligns sequences at the packet level, achieving high labeling fidelity even in no-ground-truth scenarios. We evaluate TSA across 15 scenarios derived from CVE reproduction cases, public network datasets and real-world attacks. As a result, TSA achieves a labeling accuracy of up to 0.9982 and a throughput of 225 packets per second, representing an effective and practical solution for high-fidelity traffic labeling in no-ground-truth scenarios.