/trine

This page is under progress! Will be updated soon !

Primary LanguageJava

TRINE

This repository hosts the code for "Inferring Temporal Knowledge for Near-Periodic Recurrent Events, IJCAI 2018"

Usage

The paper has three components: schedule extractor, instance extractor and a temporal inference engine (PGM). The extractors are stand-alone, while the PGM module requires the output of the extractors.

Schedule Extractor

Refer to the comments in extractor.py.

Instance Extractor

The following programs have to be run in order:

  1. Run edu.iitd.nlp.ee.freebase.FreebaseEventExtractor to get a list of events from FreeBase
  2. Run the programs in the edu.iitd.nlp.ee.corpus package to create an index of paragraphs that contain recurrent events. Choose the program based on the corpus (ClueWeb or NYT Corpus).
  3. Run edu.iitd.nlp.ee.core.InstanceCandidateOccurrenceDatePairExtractor to extract instance, candidate-occurrence-date pairs
  4. To train a occurrence date classifier use edu.iitd.nlp.ee.classify.LingpipeClassifier

PGM

The instructions are compiled as a Jupyter Notebook (TRINE_tutorial.ipynb)

References

Inferring Temporal Knowledge for Near-Periodic Recurrent Events. Dinesh Raghu, Surag Nair, Mausam. 2018. IJCAI.