SWE-QA: A Dataset and Benchmark for Complex Code Understanding

This collection of recent CS papers focuses on code understanding, automated vulnerability detection, and LLM-driven software engineering workflows. Key items: SWE-QA — a 9K-question multi-hop code comprehension benchmark exposing cross-file reasoning challenges; a broad 'Programming with Data' methodology for traceable, test-driven data engineering for LLMs; a systematic literature review of transformer-based vulnerability detection that highlights data imbalance, interpretability, scalability and generalization problems; FGDM — a flow-graph, multi-agent LLM framework using Chain-of-Thought/​