Research On Deep Learning Technology For Information Extraction Applications

Information Extraction(IE)is a technology that gathers relevant information and produce structured representation from unstructured text.It is widely used in knowledge base construction and information retrieval applications.In recent years,the focus of Information Extraction research has shifted from traditional statistical learning approaches to deep neural models.In this paper,we investigate the deep learning technologies relevant to Information Extraction and summarize the approaches of deep learning based Information Extraction algorithms.In addition,we study the use of deep learning technology in two typical Information Extraction applications: relation extraction with distant supervision and machine reading comprehension.For relation extraction,we propose the Multi-Path Convolutional Neural Network algorithm to improve performance of distant relation extraction.We improve the neural network structure by focusing on the difficulty of extracting relational information from sentence level data.By experiments we show that our method achieves higher performance than the current state of the art.For Machine reading comprehension,we recognize that current machine learning algorithms focus on the selection of a single answer for a question and propose the multi-answer machine reading comprehension algorithm.By experiments we show that our method achieves comparable results to the current algorithms on single answer questions and achieves drastically better performance on multiple answer questions.Finally,we develop a prototype system for the online deployment of deep learning-based information extraction systems.By using a fully contained package and APIs our system features rapid deployment and great compatibility.