Research on Multirelational Entity Modeling based on Knowledge Graph Representation Learning

Background: A research concern revolves around as to what can make the representation of entities and relationships fully integrate the structural information of the knowledge atlas to solve the entity modeling capability in complex relationships. World knowledge can be organized into a structured knowledge network by mining entity and relationship information in real texts. In order to apply the rich structured information in the knowledge map to downstream applications, it is particularly important to express and learn the knowledge map. In the knowledge atlas with expanding scale and more diversified knowledge sources, there are many types of relationships with complex types. The frequency of a single relationship in all triples is further reduced, which increases the difficulty of relational reasoning. Thus, this study aimed to improve the accuracy of relational reasoning and entity reasoning in complex relational models. Methods: For the multi-relational knowledge map, CTransR based on the TransE model and TransR model adopts the idea of piecewise linear regression to cluster the potential relationships between head and tail entities, and establishes a vector representation for each cluster separately, so that the same relationship represented by different clusters still has a certain degree of similarity. Results: The CTransR model carried out knowledge reasoning experiments in the open dataset, and achieved good performance. Conclusion: The CTransR model is highly effective and progressive for complex relationships. In this experiment, we have evaluated the model, including link prediction, triad classification, and text relationship extraction. The results show that the CTransR model has achieved significant improvement.