Research Of Relation Extraction Based On Entity Type Embedding And Recurrent Piecewise Residual Networks

As a core mission in information extraction,effective relation extraction is a great challenge in the information explosion era.In this paper,we found that existing models usually fail in extracting the correct relation of entity pair effectively when a sentence is long and complex or there is not the firsthand trigger word of entity relation in a sentence.Whereas,such sentences are ubiquitous in texts formed of natural language,which makes the effect of relation extraction hit a bottleneck.In this paper,we penetrate into investigating and anlyzing the researches on relation extraction at home and abroad.Inspired by the idea of fusing more and deeper information,we propose Entity Type Embedding(ETE)to enrich the representation of the model of relation extraction,and the ETE,word embedding and position embedding are served as the joint embedding of our model;In addition,we devise an architecture of Recurrent Piecewise Residual Networks(RPRN)meticulously,which makes the latent representations underlying the context of a sentence into abstractions at different levels,and accordingly picking up the structural representation of a sentence with ease when confronted with a variety of complicated large-scale corpus.The model(ETE-RPRN)of relation extraction based on ETE and RPRN,proposed in this paper,is capable of obtaining deeper information in a sentence effectively.For the sake of evaluating the model preferably,the public dataset of New York Times,extensively employed in the field of relation extraction,is adopted when conducting experiments in this paper.Subsequently,our proposal of relation extraction dubbed ETE-RPRN is compared with the models of other scholars quantitatively from diverse evaluation metrics,and then different results extracted from different models are analyzed qualitatively.Experiment results show that our proposal outperforms the state of the art models in the metrics of F1,AUC and etc,and is capable of extracting more effective relations when dealing with large-scale complicated corpus,accordingly solving the problem aforementioned of existing models effectively.Simutaneously,the extracted relation facts are key foundation of building the upper applications such as semantic search,question answering system and etc.