Research On Weakly Supervised Relation Extraction

Relation extraction refers to extracting structured data in natural language texts and saving the results in the form of triplets.Since the presentation of text information on the Internet is disorganized and cannot be used directly.It is of great significance to convert unstructured text information into structured text information.This paper focuses on weakly supervised relation extraction.We analyzed noise data in distantly supervised relation extraction and semantic drift in semi-supervised relation extraction.The detailed work is as follows:1.For data set with noise,Markov Random Field model is designed to reduce noise in the data set.The model completes the correction of text labels through Markov Random Field theory.It reduces the labels in the text set,which is the basis for distantly supervised relation extraction and semi-supervised relation extraction.2.A neural network model for Markov Random Fields is designed for distantly supervised relation extraction.Based on the noise introduced during the process of distantly supervised relation extraction,Markov Random Field model is constructed.Then CNN classifier and BGRU classifier combined with the two-level attention model are used to complete the entire relation extraction process.3.For a semi-supervised relation extraction,a bootstrapping model combined with Markov Random Fields is designed.By using the Markov Random Field model,the problem of semantic drift in the semi-supervised process is alleviated,and the effect of bootstrapping relationship extraction is improved.