Transformations  in
Business & Economics

RESEARCH ON THE GEOLOGICAL ENTITIES BUSINESS RELATION EXTRACTION BASED ON THE BOOTSTRAPPING METHOD
Lv Pengfei, Yao Zheng, Wang Chunning, Zhu Yueqin, Liu Wei

ABSTRACT. The purpose of entity-relation extraction research is to extract structured entity-relation data from unstructured documents. Through entity-relation extraction technology, scattered and independent literature resources can be constructed into a knowledge system with interrelated and multi dimensional content, which helps people quickly discover the structural context between entities, and obtain target information intuitively and systematically. Entity-relation extraction research methods have experienced evolution based on rules, knowledge bases, and machine learning. Addressing unlabelled data is the difficulty and keystone in entity-relation extraction research based on machine learning. The bootstrapping method is a solution that carries out multiple and repeated sampling for adaptive training of unlabeled data. This paper designs an entity-relation extraction method based on the bootstrapping method, which includes four steps: context construction, sentence template extraction, candidate seed extraction and candidate seed scoring. Compared with the results of manual annotation, it has been proved that the method has better adaptability without large-scale labelled data.

KEYWORDS:  geological literature, entity-relation, machine learning, unlabelled data, bootstrapping.

JEL classification:  C51, C81.