Evaluating Large Language Models for Optimized Intent Translation and Contradiction Detection Using KNN in IBN

Abstract

Intent-Based Networking (IBN) simplifies network management by enabling users to express high-level intents in natural language, but existing approaches often fail to ensure alignment with network policies, leading to misconfigurations. Moreover, many methods lack robust validation mechanisms, reducing their reliability in dynamic environments. This research addresses these gaps by evaluating advanced Large Language Models (LLMs) such as BERT-base uncased (BERT-bu), GPT2, LLaMA3, Claude2 and small deep learning model BiLSTM with attention for translating intents and detecting contradictions. Using a curated dataset of 10,000 intent pairs, the proposed hybrid framework integrates a K-Nearest Neighbors (KNN) classifier to validate translations and recalibrate erroneous outputs. Experimental results demonstrate up to 5% higher accuracy (88%) and F1 scores compared to existing methods, ensuring precise intent translation and reliable network orchestration. This approach significantly enhances scalability and policy compliance in automated network environments.