Indonesian Open Domain Information Extractor

IWBIS 2017 presentation slides of this research is available here

This is a work in progress and still focusing on proven of concept instead of precision and recall

Indonesian open domain information extractor using Stanford NLP pipeline, heuristics triples candidate generator, heuristics token expander and Random Forest triple selector.

Input example:

"Sembungan adalah sebuah desa yang terletak di kecamatan Kejajar, kabupaten Wonosobo, Jawa Tengah, Indonesia."

Expected output:

"Sembungan" "adalah"  "desa"
"Sembungan" "terletak di" "kecamatan Kejajar"

Build

Docker build:

docker build -t id-openie:latest .

To build manually without docker (required to run experiments with Matplotlib charts):

• Install build dependencies:
  • Oracle JRE/JDK 8
  • Gradle
  • Python 2.7
  • Pip
  • Tkinter (or python-tk in Ubuntu) to display classifier comparison charts
• Download and clone this repo
• Install Python dependencies: pip install -r scripts/requirements.txt
• Run build command from repo directory: gradle clean build
• Get distribution .tar or .zip from build/distribution
• Extract distribution to your choice of installation directory

Usage

Command format:

usage: python extract_triples.py [-h] [-m MODEL_FILE] [-s SCALER_FILE] [-o OUTPUT_FILE] [-f {json,tsv}] input_file

Extract triples from Indonesian text

positional arguments:
  input_file            Input file containing 1 (one) Indonesian sentence per line

optional arguments:
  -h, --help            show this help message and exit
  -m MODEL_FILE, --model_file MODEL_FILE Triples classifier model file
  -s SCALER_FILE, --scaler_file SCALER_FILE Triples classifier scaler file
  -o OUTPUT_FILE, --output_file OUTPUT_FILE Output file containing triples
  -f {json,tsv}, --output_format {json,tsv} Output file format

Example running on Linux/MacOS:

on Windows use PowerShell or replace ${PWD} with %cd% in normal command prompt

docker run --rm -v ${PWD}/data:/data id-openie python extract_triples.py -o /data/plain.triples.json /data/plain.txt
docker run --rm -v ${PWD}/data:/data id-openie python extract_triples.py -o /data/plain.triples.json -f json /data/plain.txt
docker run --rm -v ${PWD}/data:/data id-openie python extract_triples.py -o /data/plain.triples.tsv -f tsv /data/plain.txt

Example without docker:

python extract_triples.py -o plain.triples.json plain.txt
python extract_triples.py -o plain.triples.json -f json plain.txt
python extract_triples.py -o plain.triples.tsv -f tsv plain.txt

Datasets

List of Indonesian NLP datasets collected from various sources:

• POS tagging (source: https://github.com/UniversalDependencies/UD_Indonesian)

  • data/tagger-id.universal.train
  • data/tagger-id.universal.test

• NER (source: Faculty of Computer Science, UI & https://github.com/yusufsyaifudin/indonesia-ner)

  • data/ner-id-1.train
  • data/ner-id-1.test
  • data/ner-id-2.train
  • data/ner-id-2.test

• Dependency parsing (source: https://github.com/UniversalDependencies/UD_Indonesian)

  • data/parser-id.conllu.train
  • data/parser-id.conllu.dev
  • data/parser-id.conllu.test

• Triples selection (manually annotated):

  • data/triple-selector.dataset.csv (human readable)
  • data/triple-selector.train.csv (vectorized)

Models

• Stanford NLP POS tagger: src/main/resources/tagger-id.universal.model

• Stanford NLP NER src/main/resources/ner-id.model.ser.gz

• Stanford NLP dependency parser: src/main/resources/parser-id.conllu.model.gz

• Triple selector model (Scikit-Learn Standard Scaler and Random Forest):

  • scripts/triples-classifier-scaler.pkl
  • scripts/triples-classifier-model.pkl

References

1. Banko, M., Cafarella, M.J., Soderland, S., Broadhead, M. and Etzioni, O., 2007, January. Open Information Extraction from the Web. In IJCAI (Vol. 7, pp. 2670-2676).
2. Angeli, G., Premkumar, M.J. and Manning, C.D., 2015, July. Leveraging linguistic structure for open domain information extraction. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics (ACL 2015).
3. Suhartono, D., 2014. Lemmatization Technique in Bahasa: Indonesian. Journal of Software, 9(5), p.1203.

Citation

BibTex format:

@inproceedings{gultom2017automatic,
  title={Automatic open domain information extraction from Indonesian text},
  author={Gultom, Yohanes and Wibowo, Wahyu Catur},
  booktitle={2017 International Workshop on Big Data and Information Security (IWBIS)},
  pages={23--30},
  year={2017},
  organization={IEEE}
}