| With the rapid development of network information technology,the scale of knowledge graph keeps increasing,resulting in low computational efficiency and the problem of long-tail sparsity of data.In order to solve the above problems,relevant researchers introduce distributed low-dimensional vectors to represent knowledge graph triples.Knowledge graph triplet entity and relation embedding contain rich information.However,the training process of traditional knowledge representation learning model takes triplet as the unit and conducts model training and learning relatively independently.Therefore,the potential knowledge graph feature information among triples cannot be effectively utilized,which limits the ability of model knowledge representation learning.Selecting an appropriate method for fusion the structure information of triplet entities and relationships can improve the representation learning effect of the model.Therefore,this thesis studies the knowledge graph representation learning method based on the triplet structure information,takes the entity embedding in the triplet as the information carrier,and delineates the range of information fusion with the triplet containing the entity.According to the triplet structure information,from different fusion perspectives such as entity and relationship,a series fusion method is proposed and the corresponding model is constructed.A link prediction experiment was carried out on the constructed model on public data sets,and a fusion method with better experimental performance was found out through comparative analysis of the experimental results,and further experimental verification was carried out on the new data set to obtain the final knowledge representation fusion model.The main research results are summarized as follows:1)In order to solve the problem that the traditional knowledge representation learning model is relatively independent in the training of knowledge graph triplet and does not integrate the graph information contained in the graph triplet entity and relationship,this thesis completed the construction of knowledge representation learning model that integrates the structural information such as triplet entity and relationship,and compared and analyzed the experimental results of series of fusion models.Finally,a presentation learning fusion model with excellent comprehensive performance is obtained by experimental verification on a new data set.2)In order to further explore the fusion method to improve the learning performance of model representation,a series of controlled experiments based on triplet structure information were designed and corresponding models were constructed,and the experimental results of fusion models were compared and analyzed.It was found that the experimental performance of graph linkage prediction was comprehensively improved after the model fused graph triplet relationship information.In particular,the hit@1 index of model for knowledge graph linkage prediction has been improved rapidly,which has certain reference significance for the research on knowledge representation learning. |
