Pattern Matching Method Of Heterogeneous Data Based On Attention Mechanism

With the rapid development of information technology,there are still many heterogeneous data stored in different forms.The isolated information island caused by this has already become a serious problem,hampering the analysis,share and application of data.To deal with the problem of semantic heterogeneity,pattern matching method has become one of the hotspots in current research.However,it still has some shortcomings,such as too large matching space,low accuracy and efficiency.Based on the technology of deep learning,this paper focuses on how to reduce the matching space and improve the matching efficiency and accuracy.The research of this thesis mainly includes the following points:Based on the analysis of the existing pattern matching methods,the superiority of pattern matching method based on learning is clarified.Secondly,the pattern matching method represented by BP neural network is analyzed in detail.It is found that the original data index system used for feature extraction and the model used for similarity calculation have some disadvantages.The original data index system ignores the problem that different types of data have different sensitivities to the same characteristic index.Based on the attention mechanism,this paper calculates the attention weights of the original feature indicators in numerical data and character data.Through experiments,select the six indicators with higher weights as key indicators.The experimental results show that the precision,recall and F1 value is higher than that of the original one,whether applied to BP neural network or attention pattern matching model.Meanwhile,it can improve the efficiency of manual feature extraction and reduce the complexity and matching space of the model.Analyzed the reason of low matching accuracy of the BP neural network pattern matching model,a pattern matching model based on attention mechanism is established.Experiments show that in multiple scenes,the performance of attention pattern matching model in precision and recall is better than that of BP neural network.At the same time,it is theoretically verified that the model can further filter the interference data and improve the matching accuracy.The method proposed in this paper can improve the accuracy and efficiency of matching,and the precision can reach 90%.In addition,it reduces the complexity and matching space of the model,and brings a new idea for solving the semantic conflicts of heterogeneous data.