Chinese NER and relation extraction, split into focused labs

What it does

This repository houses two main Chinese NLP pipelines: named-entity recognition built on IDCNN or biLSTM with a CRF layer, and relation extraction driven by a biGRU with character- and sentence-level attention mechanisms. It also points to an external word-segmentation project using iterated dilated convolutions. Each task lives in its own folder with a separate README, so the root repo acts more like an index than a monolithic framework.

The interesting bit

Rather than forcing everything into one coupled codebase, the project keeps each pipeline isolated. That modularity makes it easier to borrow just the NER or just the relation-extraction logic without untangling a shared training loop. It is essentially a curated wiring of established techniques—acknowledged references include thunlp/Tensorflow-NRE, zjy-ucas/ChineseNER, and koth/kcws—repurposed specifically for Chinese text.

Key highlights

• NER via IDCNN/biLSTM+CRF, relation extraction via biGRU+2ATT
• Self-contained subdirectories (NER_IDCNN_CRF/, RE_BGRU_2ATT/) with independent documentation
• Explicitly references upstream inspirations and related implementations
• Targets Chinese-language corpora throughout

Caveats

• The root README is sparse; you will need to dig into the sub-folder READMEs to find actual implementation details.
• One of the three advertised tasks (word segmentation) is actually an external link to hankcs/ID-CNN-CWS, not code hosted here.

Verdict

Worth bookmarking if you need decoupled reference implementations for Chinese NER and relation extraction. Pass if you want a single, unified framework.