Research On Knowledge Graph Completion Based On PTransD-HRS Representation Learning

As an important branch of artificial intelligence,knowledge graph has been widely applied in various fields,such as search engines,question answering systems,etc.In recent years,the representation learning method oriented to the completion of the knowledge map has received extensive attention from researchers.Its purpose is to map entities and relationships in the knowledge map to the continuous low dimensional vector space,and retain the original structural Semantic information,which promotes the reasoning analysis of large-scale knowledge maps.However,there are still some problems with current representation learning models,such as most models only consider structured information within triples,and the ability to represent entities and relationships is insufficient;Ignoring the potential knowledge hidden by the rich additional information.This leads to insufficient training of entity and relationship vectors in the knowledge graph,making it difficult to train the model more effectively.In response to the above issues,the main research work of this article is as follows:(1)A new knowledge graph representation learning algorithm PTrans D-HRS,which integrates relationship paths and hierarchical relationship structure information,has been proposed.By acquiring the potential Semantic information of the multi-step relationship path and adding the representation of entity information to the path,the algorithm makes the entity and relationship vector have richer knowledge representation;Secondly,a hierarchical relationship model is introduced to transform the vector representation of relationships into the sum of three levels of relationship vectors,so that the relationship vectors contain rich hierarchical structures to alleviate the problem of sparse knowledge graphs.In the experiments on the FB15 K and WN11 datasets,this algorithm compared with the Trans E model showed a comprehensive improvement of 13.1% in accuracy in triplet classification tasks,enhancing the ability to learn knowledge representation.(2)Propose a negative sampling method based on K-means.This algorithm overcomes the problem of low model efficiency caused by poor quality of negative samples.First,K-means clustering algorithm is used to cluster all entity vectors.When it is necessary to generate negative samples,first find the cluster where the replacement entity is located,and then select the entity closest to the replaced entity in the vector space to replace,so as to obtain high-quality negative samples.The experiment shows that the performance of the model is improved after using the K-means algorithm to generate negative samples HITS@10 The indicator has increased by 2.5%,proving that it provides high-quality negative samples for model training and improves training efficiency.(3)Designed an intelligent manufacturing knowledge graph system.The system includes a user module,a data management module,and a knowledge graph completion module.The knowledge graph completion module utilizes the PTrans D-HRS model proposed in this paper to perform entity prediction and link prediction tasks on all data to complete missing fact triplets.The visualization effect shows that it can effectively improve the quality of the knowledge graph.