基于深度信念网络的地质实体识别方法

摘要

地质实体作为地质信息表达的核心要素,对其准确识别是地质文本数据挖掘和应用的重要基础.本文通过分析各种类型文本数据中地质实体信息的描述特点,构建了地质实体信息的标注规范和语料库,设计了基于深度信念网络(DeepBelief Networks)的地质实体识别模型,解决了文本数据中地质实体信息的结构化、规范化处理问题.以矿产资源地质调查报告为实验数据,对本文的地质实体识别方法性能进行了评估分析.结果表明,深度学习模型能够在较小规模语料库的基础上,达到较好的地质实体识别性能.%As the core element of a geological entity recognition,the correct recognition of geological entities is important in geological text data mining domain.This paper firstly analyzes the linguistic characteristics about geological entity information in various types of text,and then proposes an annotation scheme for the annotation of corpus for geological entity in natural language.Furthermore,a geological entity recognition model based on Deep Belief Networks is put forward,which is to solve the problem about the structuring and normalizing of geological entity information.In order to evaluate the recognition model's performance for geological entity recognition,an experimental study by using mineral resources geological investigation report as source data is carried out.The result shows that the recognition model based on Deep Belief Networks can obtain good recognition results on small-scale corpus.