| Natural Language Processing(NLP)is one of the most challenging research directions in the field of artificial intelligence(AI).Among the sub-problems of NLP,Relation Extraction(RE)is one of the basic ones.According to the scale of the analyzed text,RE can be divided into three categories,i.e.,inner-sentence-level,inter-sentencelevel and document-level.The inner-sentence-level RE is to analyze a given sentence and the entities in the sentence—that is,the key information that we pay attention to in the sentence—and analyze the semantic relationship among these entities,thereby the unstructured information could be transformed into structured and easy-to-understand.For example,there is a relation 'capital' between entities 'China' and 'Beijing' in the sentence 'China's capital is Beijing'.Although RE has been studied over decades,it still faces two kinds of research challenges that are not well addressed thus far:1)joint consideration of the global sentence structure and the local entity interaction,and 2)effective solution to handling the overlapping triplets within the same sentence.To tackle these issues,in this paper,we present Global-local grAph-based convolutional network towards Multi-relation Extraction,GAME for short.In particular,we devise two layers of graph neural network(GNN)with different structures to complete the extraction of features,which effectively improves the capability of relation extraction.Moreover,we implement the GCN layers with graph convolutional network and graph attention network respectively for further comparison.Besides,we adopt a classification strategy to extract relation among entity pairs,assisting in solving the more complicated problem of overlapping triplets in RE.Extensive experiments have been conducted on two widely-used benchmark datasets,demonstrating that our model significantly outperforms several state-of-the-art methods.In addition,we also conducted ablation study on our model to test each component. |
PDF Full Text Request