Research on a Framework for Chinese Argot Recognition and Interpretation by Integrating Improved MECT Models

Area of Science:

• Computational Linguistics
• Artificial Intelligence
• Criminology

Background:

• Underground industries utilize specialized argots for covert communication and evading surveillance.
• Existing methods for deciphering these argots are insufficient for law enforcement.

Purpose of the Study:

• To develop and evaluate novel frameworks for recognizing and interpreting industry-specific argots.
• To provide enhanced technical support for law enforcement in combating illicit activities.

Main Methods:

• Utilized word vectors and large language models (LLMs) to analyze semantic differences in argots.
• Developed a labeled argot dataset (MNGG).
• Created an argot recognition framework (CSRMECT) and an interpretation framework (LLMResolve) using MECT, LLMs, prompt engineering, and DBSCAN clustering.

Main Results:

• The CSRMECT framework improved F1 score by 10% for argot recognition on the MNGG dataset compared to optimal models.
• The LLMResolve framework achieved 4% higher accuracy in argot interpretation.
• Identified a potential correlation between vector information entropy and model performance.

Conclusions:

• The proposed CSRMECT and LLMResolve frameworks offer significant improvements in argot recognition and interpretation.
• These AI-driven tools can enhance law enforcement's ability to detect and address illicit activities.
• Information entropy in word vectors may be a key factor in model performance.