SNLP Fact Checker

Goal

Given a list of sentences test the correctness of each sentence

Method Overview

• Extract Subject, Relation, Object triples from the input sentence
• Find Wikipedia article of the extracted Subject
• Parse Wikipedia Infobox of the Subject to extract Subject, Releation, Object triples
• Compare Wikipedia triples against the input triples
• Generate ttl file of results

Example

• Input:

3347316	Nobel Peace Prize is Henry Dunant's honour.

• Subject: "Henry Dunant" Object: "Nobel Peace Prize" Relation: "AWARD"
• Related Wikipedia article: en.wikipedia.org/wiki/Henry_Dunant
• Infobox:

• Triples extracted from Wikipedia:

  1. "Henry Dunant", "BORN_IN", "Geneva, Switzerland"
  2. "Henry Dunant", "DIE_IN", "Heiden, Switzerland"
  3. "Henry Dunant", "AWARD", "Nobel Peace Prize"

• Comparing Input triple against Wikipedia triples yields the fact "Henry Dunant", "AWARD", "Nobel Peace Prize" is correct

• Output 1.0 for this sentence

• Output example

<http://swc2017.aksw.org/task2/dataset/3347316><http://swc2017.aksw.org/hasTruthValue>"1.0"^^<http://www.w3.org/2001/XMLSchema#double> .

Running the application

• create a folder called data and put test.tsv file in this folder
• set VM arguments as follows: -Xmx3060m -Dfile.encoding=UTF-8
• run Main file de.upb.snlp.scm.Main
• result.ttl file will be created in data folder
• use conf.properties file to manage configuration (NER classifiers should be downloaded from nlp.stanford.edu)

Presentation

Presentation of the process

License

MIT