A Fact Prediction Model based on Hyper Graph and Neural Network

We presents a knowledge graph reasoning model based on neural network, which aims to improve the capability of predicting missing entities in knowledge graphs. Traditional methods often struggle with the prediction of facts as they treat them like unseen entities, leading to poor performance. To address this challenge, we developed a robust representation for facts, leveraging a neural network model enhanced with data augmentation techniques. Our approach is evaluated using two datasets derived from real-world knowledge bases, including FBH and DBHE. These datasets feature a rich structure of associations and higher-level relationships, enhancing the complexity and expressiveness of the knowledge graph. Our experimental results show significant improvements over several state-of-the-art methods in terms of Mean Reciprocal Rank (MRR) and Hits@10, demonstrating the efficacy of our proposed method in predicting both entities and facts accurately. Furthermore, we observed that increasing the number of layers in our neural network model up to a certain point consistently improved performance, indicating the stability and effectiveness of our model.