Relation Extraction Method And Application Based On Graph Convolution

Relation extraction is a critical stage in the field of natural processing,as well as a critical technical support in its implementation.Relation extraction is vital in the building of semantic networks,information retrieval,question answering systems,and other sophisticated applications since it provides the theoretical foundation of natural language processing.At present,Deep learning is the core of most current relationship extraction algorithms.This method overcomes the drawbacks of existing methods,such as extraction difficulty and a high workload.It can extract data automatically rather than relying on pre-defined rules.It’s characteristics extractors are mostly based on recurrent neural networks and convolutional neural networks.This paper analyzes and studies the relationship extraction method of graph convolutional neural network.But this approach still has issues like gradient blurring,improper learning of feature nodes and only node information is taken into account in feature extraction at this time,learning about differences and correlations isn’t enough.Meanwhile,the presence of noise nodes has an impact on the accuracy with which feature vectors are transferred.As a result,relevant improvement plans are offered,with the following precise contents:(1)A graph convolution relation extraction approach is devised that incorporates GRU and an attention mechanism.This method uses GRU to Learn the context of the sentence,strengthen the word correlation and solve difficulties like gradient blurring.Meanwhile,The attention mechanism weights the relationship between entities,prunes invalid nodes,improves information extraction efficiency and filters out faulty information in sentences.The relation extraction model is verified by the SemEval-2010Task8 and SemEval-2007Task4 datasets.The results of the experiments reveal that this strategy increased the percentage of its F1 value and decreased the impact of noise words.Finally,an optimized experimental model is developed by evaluating and testing the experimental parameters,which enhances extraction accuracy even further.(2)A method for fusing edge characteristics to derive edge-graph convolution relations is proposed.Edge features are introduced to further learn the differences of node information,To introduce edge features to further learn nodes,the attention mechanism and the gating mechanism are applied.To acquire fused entities perceived edge features,first use attention mechanism to assign relevant feature information to distinct edges and then utilize gating method to fuse entity information.The edge feature matrix is then added to the graph convolution,and the edge feature is utilized to realize the three-stage edge graph convolution of node-to-node,node-to-edge,and edge-to-node,which subsequently aggregates node and edge information.In the data sets SemEval-2010Task8 and SemEval-2007Task4,Adding edge features to the graph convolution relational extraction model to execute edge graph convolution improves the relationship extraction impact greatly over other techniques and its rich features are successful in inference,according to ablation experiments and comparison experiments,improve the relationship extraction model’s performance.(3)It is built and implemented a graph convolution relation extraction system.The system is designed with front and back ends separated,as well as a relation extraction module and a relation prediction visualization module.The front-end is built using a WeChat applet that mostly displays the user’s information registration and relationship extraction findings visually.The user’s information management,model training,and relationship extraction are all handled by the back-end,which is written in Python.The system successfully extracts relationships between elements in the text and the interplay between data and relationship extraction is presented to the user in a more intuitive manner.According to the above research,the relationship between entity relationships can be positioned more precisely by introducing new attributes,which significantly improves the accuracy of relationship extraction results.