Hybrid NLP Pipeline for Detecting Irregularities in Public Procurement
August 13, 2026
A framework combining Word2Vec, LLaMA, and RoBERTa embeddings with Gaussian Mixture Models to identify accusatory language in government procurement metadata. The pipeline uses unsupervised clustering and supervised classification to detect procedural irregularities in participant comments.
HOW THIS AFFECTS YOU
●
researcherYou can use this multi-stage embedding and clustering approach for domain-specific fraud detection tasks.