Performance Comparison Experiment of LLM Blackened Intelligence Perception Method in Data Poisoning Context

The experimental dataset consists of five parts, targeting data poisoning identification, intelligence perception, blackening separation, blackening energy, blackening penetration, and blackening destruction tasks. Each part of the dataset contains multiple performance indicators. When using this dataset, preprocessing is necessary to ensure data validity, then the improvement rate and leading rate of the new method and the reference method are calculated and compared, and finally the results are displayed through visualization. The experimental results show that the new method exhibits significant advantages in terms of improvement rate and lead rate in all tasks. When using this dataset, attention should be paid to data integrity and consistency to ensure the reliability of the results.