CLIC Framework Characterizes LLM Coding Behaviors via Token Frequencies

The CLIC visual analytics approach uses token-frequency analysis and decision trees to differentiate the coding styles of various LLMs. The method introduces robustness and concentration metrics to quantify how distinguishable models remain when their most discriminative tokens are removed.