MinIE is an Open Information Extraction system which provides useful extractions (in unsupervised manner) from natural language sentences. The extractions are in the form of triples. It represents information about polarity, modality, attribution, and quantities with semantic annotations instead of in the actual extraction. MinIE also identifies and removes words in the extractions that are considered overly specific. As a result, it gives shorter, semantically enriched extractions with high precision and recall.


Kiril Gashteovski, Rainer Gemulla, Luciano del Corro
MinIE: Minimizing Facts in Open Information Extraction [pdf]
Conference on Empirical Methods in Natural Language Processing (EMNLP), 2017

Source code

The source code is available on GitHub. For more detailed documentation, please go to the documentation page.


Datasets from the paper's experiments:

Dictionary created from frequent relations and arguments:

  • Wikipedia: frequent relations and arguments [zip]

Labeling guide

Labelling guidelines for the human judges [pdf]