MAPS-KB

Code and data for the paper "MAPS-KB: A Million-scale Probabilistic Simile Knowledge Base" (AAAI 2023)

Data

You can download the following files from this Google Drive URL:

MAPS-KB.csv: Our million-scale probabilistic simile knowledge base, including 4.3 million simile triplets in the form of (topic, property, vehicle) along with two probabilistic metrics, plausibility and typicality, to model them.
Simile Instances: The simile instances in the form of (simile sentence, topic, property, vehicle) that we used to construct MAPS-KB. We collect them from 70 GB corpora, including:
- simile_instance_bc.csv: BookCorpus
- simile_instance_gutenburg.csv: Gutenburg
- simile_instance_opw.csv: Openwebtext

Demo

You can run the MAPS-KB demo through the following Colab link, with supported operations: look_up, simile generation, and simile interpretation.