Embedded Files
RQ1: What is the overall performance of our approach compared with the baselines?
MINES is more effective in detecting attacks than baselines. The precision of MINES is 100%, and the recall is more than 15% higher than the baselines.
RQ2: What is the training cost and runtime cost of our approach?
RQ3: How do different components (deducing from schemas, joining extra information, and binary log history tracking) in our approach affect its performance?
Deducing from schemas and incorporating API-DB, API-API, API-Env relationships, and binlog history tracking significantly enhance MINES’s performance.
RQ4: How much can deducing from schemas in MINES reduce the input length compared to inducting from raw logs?
Deducing from schemas reduces input length by up to 2 orders of magnitude compared to inducing from raw logs.
RQ5: Is our approach robust when equipped with different language models?
MINES achieves consistent performance across dif- ferent LLMs, indicating robustness.
RQ6: Can our approach generalize to popular web applications?
Page updated
Google Sites
Report abuse