AntNRE

AntNRE is a neural entity relation extraction package built on PyTorch. It aims to provide efficient and flexible toolkits for building information extraction systems.

Modularity

AntNRE contains modules which can be used as building blocks for various entity relation extraction systems. For example,

• Encoders: CNN/RNN-based word representations, CNN/RNN-based sequence modelling.
• Entity Models: sequence labelling with RNN + CRF.
• Relation Models: Feature-enriched PCNN.

Flexibility

One could use the package to implement many state-of-the-art entity relation extraction systems. For example,

• Joint entity relation extraction model.
• Minimum risk training based model.
• Multi-task joint entity relation extraction model.

Future Plans

More advanced systems will be integrated in the future. For example,

• Multi-instance multi-label models.
• Deep latent models.