KV-rule

About

KV-rule is a rule-based fact checker that makes weighted logical positive and negative rules, and uses them to calculate a truth score for a given triple.

For example, the triple (Wozniak, education, UC Berkely) is assigned the high truth score 0.96 close to true since it is logically consistent with the triple (Wozniak, almaMater, UC Berkely) in a knowledge graph according to the positive rule (x, education, y) ← (x, almaMater y).

In contrast, the triple (Wozniak, birthPlace, Florida) is assigned the low truth score 0.06 close to false since it is logically contradict to the path (Wozniak, birthPlace, California) ∧ (California, ≠, Florida) in a knowledge graph according to the negative rule ¬(x, birthPlace, y) ← (x, birthPlace, z) ∧ (z, ≠, y).

Prerequisite

• python 3
• bottle (optional)

How to use

Download the pretrained positive and negative rules

1. Download the compressed dataset inter.tar.bz2 (Link).

2. The compressed dataset inter.tar.bz2 contains the pre-processed knowledge graphs (English DBpedia and K-Box) and the pre-trained positive and negative rules.

3. Unzip the compressed dataset inter.tar.bz2 by the command tar -jxvf inter.tar.bz2.

4. Locate all the contents in the unzipped dataset into the directory inter in the main directory of KV-rule.

Calculate a truth score for a given triple

1. Go to the directory code in the main directory of KV-rule by the command cd code.

2. Calculate a truth score for a given triple by the following command:

python3 filter.py -i INPUT_FILE_PATH -o OUTPUT_DIR_PATH -cn CONFIG_FILE_PATH

An input file (INPUT_FILE_PATH) should contain a set of triples in the format of tab-separated-values (TSV), as follows:

Alanis_Morissette	nationality	Canada
Alanis_Morissette	nationality	United_Kingdom
Alanis_Morissette	nationality	Italy
Alanis_Morissette	nationality	Nigeria
Albert_Einstein	nationality	Germany
Albert_Einstein	nationality	Scotland
Albert_Einstein	nationality	Venezuela
Albert_Einstein	nationality	Iran

An output directory (OUTPUT_DIR_PATH) would contain the result file result-scored.tsv that contains a set of triples and calculated truth scores in the format of TSV, as follows:

Alanis_Morissette	nationality	Canada	1.0
Alanis_Morissette	nationality	United_Kingdom	0.9216958227441079
Alanis_Morissette	nationality	Italy	0.7240789450444748
Alanis_Morissette	nationality	Nigeria	0.720765944969006
Albert_Einstein	nationality	Germany	1.0
Albert_Einstein	nationality	Scotland	0.7249292427182386
Albert_Einstein	nationality	Venezuela	0.7225167109516941
Albert_Einstein	nationality	Iran	0.7180768931575049

If you want to calculate a truth score for an English DBpedia-style triple, set a configuration file (CONF_FILE_PATH) as conf/conf-dben.json.

Or, if you want to calculate a truth score for a K-Box-style triple, set a configuration file (CONF_FILE_PATH) as conf/conf-kbox.json.

Licenses

• CC BY-NC-SA Attribution-NonCommercial-ShareAlike
• If you want to commercialize this resource, please contact to us

Publisher

Machine Reading Lab @ KAIST

Contact

Jiseong Kim. jiseong@kaist.ac.kr, jiseongyee@gmail.com

Acknowledgement

This work was supported by Institute of Information & Communications Technology Planning & Evaluation(IITP) grant funded by the Korea government(MSIT) (2013-2-00109, WiseKB: Big data based self-evolving knowledge base and reasoning platform).