Research On Knowledge Reasoning Based On Knowledge Graph

With the continuous development of large-scale knowledge graphs,the number of entities and relationships continues to increase,but they are often incomplete.Knowledge reasoning technology can reason new knowledge based on the existing entity relationship information in the knowledge graph,so as to complement the knowledge graph.Most of the current knowledge reasoning technologies only consider the triple structure information in the knowledge graph,but ignore the rich latent semantic information contained in the knowledge graph,such as entity neighborhood information,semantic category information,etc.Based on the classic knowledge representation learning model Trans E,this thesis fuses the neighborhood semantic information of entities,and then uses the Bi-LSTM neural network and attention mechanism to model multi-step relational paths,so as to provide a new way of knowledge reasoning for the fusion of various information in the knowledge graph.The main research contents of the thesis are as follows:(1)Aiming at the problem that the current knowledge representation learning model does not fully utilize the information contained in the knowledge graph,this thesis proposes a knowledge representation learning model based on neighborhood semantics.First,take the entity as the center,construct a hybrid neighborhood representation that fuses entity neighborhood information.Secondly,fuse entity semantic category information,and extend the entity semantic category to the neighborhood semantic category with the mixed neighborhood as the unit.Finally,construct New score function and loss function for model training by minimizing the loss function.Comparative experiments show that the model in this thesis is superior to other comparative models in MR,HITS@K in the link prediction task,and MicroACC and Macro-ACC in the triplet classification task.(2)In the knowledge reasoning method based on multi-step relationship path,the contribution of entities and relationships contained in the relationship path is not differentiated,and multiple relationship paths are not comprehensively considered.This thesis proposes a knowledge reasoning model based on Bi-LSTM neural network and attention mechanisms.The model is divided into three layers: the first layer is the embedding layer,which uses the entity and relationship vectors trained by the knowledge representation learning model based on neighborhood semantics proposed in this thesis as the initialization vectors,and the corresponding initialization vectors of the entity-relationship pairs included in the relationship path are spliced as the input of the second layer.The second layer is the Bi-LSTM layer,which takes the embedding layer as the input and uses the Bi-LSTM network to capture the semantic associations between entities in the multi-step relational path and their long-term dependency information.The third layer is the attention layer.The output of the second layer is weighted through the attention mechanism to obtain the representation of the multistep relationship path.Finally,the mixed representation of the multiple multi-step relationship paths is obtained through the attention mechanism.Comparative experiments show that the MAP and MRR of the proposed model are better than other comparative models in relation prediction task and link prediction task.