Design And Implementation Of Knowledge Fusion System Based On Dual-Attention Mechanism

In recent years,the emergence of related recommendations and intelligent question answering has greatly promoted the development of knowledge graph technology.The construction of large-scale universal knowledge graph has become the focus of scholars at home and abroad.However,due to the serious heterogeneity and redundancy between the existing types of knowledge graphs,it is necessary to efficiently and accurately determine whether the two entities point to the same objective object when the knowledge graphs are fused,that is,the knowledge graphs are fused by entity alignment technology jobs.At present,the research on entity alignment technology has made great progress,but there are still some limitations.For example,as the amount of data increases,traditional probability models based on statistics have the problem of high time complexity.The entity alignment algorithms based on expert experience cannot meet the adaptation problems caused by the increase in data volume,resulting in lower accuracy.The main problem of the method based on machine learning or deep learning is that it is difficult to obtain a priori alignment data.In addition,there is a problem of low efficiency of entity vector generation under Chinese text.In the above context,this paper proposes an entity alignment algorithm based on the dual-attention mechanism,at the same time designs and implements a knowledge fusion system based on the dual-attention mechanism.The system can efficiently and accurately perform entity alignment,knowledge graph fusion,knowledge graph query and other functions.The research work in this paper is as follows:(1)An entity alignment algorithm based on the dual-attention mechanism is proposed.First,in view of the low efficiency of entity vector generation under the Chinese knowledge graph,this paper proposes an E-CBOW entity vector training model based on Continuous Bags Of Words(CBOW),using a Convolutional Neural Network(Convolutional Neural Network,CNN)further trains the public word vectors to obtain entity vectors containing entity semantic information and knowledge graph structure information.This method avoids word vector training starting from one-hot encoding,and trains entity vectors more efficiently and accurately.Then,in order to solve the problem that the accuracy of entity alignment judgment is low due to the inaccurate representation vector between two entities in the entity alignment algorithm based on deep learning,this paper designs a dual-attention mechanism.According to the characteristics of word information and sentence information,this mechanism designs different similarity learning methods to learn the attention weights of the entity's word information and sentence information to obtain a more accurate vector between the two entities,which further improves the accuracy of entity alignment judgment.Finally,for the small amount of Chinese prior alignment data,this paper proposes an incremental data learning method based on iterative ideas.In this way,the entity alignment model training and data set expansion are alternately performed to solve the problem of prior alignment data to a certain extent.We have performed simulation experiments on the algorithm designed in this paper.(2)Aiming at the user's visual operation requirements for functions such as data preprocessing,entity alignment,knowledge graph fusion,and knowledge graph query,this paper designs and implements a knowledge fusion system based on the dual channel attention mechanism based on the Django framework.This article describes in detail the requirements analysis,system architecture and system module processing flow of the knowledge fusion system.It designs and implements module functions such as data preprocessing,entity alignment,graph fusion,and knowledge graph query.Finally,we tested the system in detail from both functional and performance perspectives.According to the simulation experiment results of comparison with the existing entity alignment algorithms such as BootEA and SEEA,it can be seen that the algorithm proposed in this paper has greatly improved various indicators such as Precision,Recall,and F1-measure.At the same time,after systematic testing,the knowledge fusion system designed in this paper can accurately and efficiently perform knowledge graph entity data preprocessing,entity alignment,graph fusion and knowledge graph query operations.The system meets the function and performance requirements of the knowledge graph fusion system.