《An Automatic Knowledge Graph Construction System for K-12 Education》阅读笔记

论文原址。本篇文章主要提出了一个自动化构建数学领域知识图谱的系统,主要应用的事NER和数据挖掘技术,其中NER主要是抽取数学概念,概念间的关系是作者自己构建的(例如先修关系)。对于数据集,作者主要从the Chinese curriculum standards of mathematics上提取的概念实体,从自己的SLP平台上,通过对学生表现来提取关系(把这部分作为数据挖掘)。本篇文章实际上可以作为构建特定领域的知识图谱的一个参考。

challenges

  • the desired educational concept entities are more abstract than real world entities like PERSON, ORGANIZATION, LOCATION
  • the desired relations are more cognitive and implicit, so cannot be derived from the literal meanings of text like generic knowledge graphs

contributions

  • a novel but practical system
  • entity recognition (NER) & association rule mining algorithms
  • demonstrate an exemplary case with constructing a knowledge graph for the subject of mathematics

SYSTEM OVERVIEW

  • Educational Concept Extraction Module:
  • Implicit Relation Identification Module

CONCEPT EXTRACTION

  • 线性链式CRF模型

  • 标签预测

RELATION IDENTIFICATION

两种方法

  • support
  • confidence

From the perspective of prerequisite relation, if concept si is a prerequisite of concept sj, learners who do not master sivery likely do not master sj, and learners who master sjmost likely master si.

EXEMPLARY CASE AND SYSTEM EVALUATION

Concept Extraction

Dataset

the Chinese curriculum standards of mathematics published by the ministry of education as the main data source

Evaluation

  • adopt precision, recall and F1- score
  • The ground truth is manually labeled by two domain experts.

Relation Identification

Dataset

students’ performance data collected by our SLP platform.

Evaluation

The ground truth of the prerequisite relations between selected 9 concepts are annotated manually by two domain experts.


 上一篇
2019年计划 2019年计划
读个博吉他可以弹唱几首流行歌曲假期报个班,请老师指点一下 精读50篇论文平均下来,每周都需要读一篇,每一篇都要勾划,发博客。 发2篇paper上半年积累知识,做实验,暑假开始写 考托福过90上半年背单词等 买个微单学个拍照不急,现在没钱,下
2019-02-09
下一篇 
《RECURRENT NEURAL NETWORK REGULARIZATION》阅读笔记 《RECURRENT NEURAL NETWORK REGULARIZATION》阅读笔记
论文链接。这篇论文提出了LSTM的dropout策略来防止过拟合,即只在非循环链接处采取dropout。在BasicLSTMCell的接口就是依据这篇论文实现的。 文章整体架构和重点 contribution present a sim
2019-01-07
  目录