Daria Chubarova, Ivan Efimenkov
In law, finance, and journalism, identifying contradictions in texts is important to ensure data integrity. Traditional methods are often ineffective. The article presents a contradiction detection system using knowledge graphs (KG). The system includes data processing, KG construction, embedding training, classification models, and evaluation. Methodology: text transformation into knowledge triples, KG compliance (DBpedia, Wikidata, UML), application of reasoning algorithms. Experiments on FEVER show a high accuracy of detecting contradictions with the KG model. The system is an effective tool for analyzing text data. Prospects: improving models, adapting to other languages and domains
Results folder include metrics from the last execution (5 epoch)
code_gpu_10.py provides approach with 5 epochs
code_gpu_11.py - 2 epochs