FIGMENT-MULTI

Multi-level Representations for Fine-Grained Typing of Knowledge Base Entities

This is an extension to the old FIGMENT.

There are implementations for learning character, word, and entity level representations for entity typing.

Installation

Python: 2.7.11 
Numpy: 1.10.4 
Theano: 0.8.2 
Blocks: 0.1.1 
Fuel: 0.1.1 

Prepare datasets

To download embeddings and preparing datasets, first run this command:

sh prepare.sh

Training

To train, test and evaluate one of the models, you can use this script ("swlr" is the example model in this case):

sh run_one_model.sh swlr

There should be a subdirectory with the name "swlr" in "configs" directory. All the models in the paper already have one config directory. Please see (https://github.com/yyaghoobzadeh/figment-multi/tree/master/configs) for the list.

References

More information about the models is in this paper:

Multi-level Representations for Fine-Grained Typing of Knowledge Base Entities, Yadollah Yaghoobzadeh, Hinrich schütze. (EACL2017).