How to train NeuroNER on my own text data?

Format your data in CoNLL or BRAT format, organize it into train/valid/test folders, and run the training command with parameters. The README provides examples for adding new datasets, but be prepared for manual configuration and embedding downloads.

Does NeuroNER work for languages like Spanish or Chinese?

While theoretically possible, pretrained models and embeddings are mainly for English. You'd need to supply custom word embeddings and train from scratch, which requires significant effort and data.

NeuroNER vs SpaCy for named-entity recognition?

NeuroNER offers higher accuracy on benchmarks and better custom training flexibility, but SpaCy has easier installation, production-ready APIs, and broader language support. Choose NeuroNER for precision in domains like clinical text; SpaCy for rapid deployment.

Can I use NeuroNER with GPU acceleration?

Yes, install the GPU version via 'pip3 install pyneuroner[gpu]' and use CUDA_VISIBLE_DEVICES to control GPU usage, as noted in the README. However, ensure your system meets TensorFlow 1.x GPU requirements.

How accurate is NeuroNER compared to BERT-based models?

NeuroNER's LSTM-CRF models achieve state-of-the-art scores on older benchmarks like CoNLL-2003, but BERT-based models often outperform them on newer tasks. NeuroNER doesn't natively integrate BERT, so for cutting-edge performance, you might need alternative tools.

How to integrate NeuroNER into a web app?

You can call it from Python scripts using the neuromodel module and wrap it in a REST API, but the README lacks detailed deployment guides, requiring custom engineering for production environments.

Open-Awesome

NeuroNER

MITPython1.0-dev2

An easy-to-use, state-of-the-art named-entity recognition (NER) tool based on neural networks.

Visit Website GitHub

1.7k stars472 forks0 contributors

What is NeuroNER?

NeuroNER is an open-source named-entity recognition (NER) tool that uses neural networks to identify and classify entities like persons, organizations, and locations in text. It solves the problem of extracting structured information from unstructured text with state-of-the-art accuracy while being designed for ease of use.

Target Audience

Researchers, data scientists, and developers working on natural language processing tasks who need a reliable, high-performance NER system without extensive deep learning expertise.

Value Proposition

Developers choose NeuroNER because it combines academic-grade performance (e.g., matching top scores on benchmarks like CoNLL-2003) with a simple installation process and flexible usage via command line or Python, all in an open-source package.

Overview

Named-entity recognition using neural networks. Easy-to-use and state-of-the-art results.

Use Cases

Best For

Extracting medical entities from clinical notes for de-identification
Performing NER on news articles or general text using pretrained models
Research projects requiring reproducible, state-of-the-art NER baselines
Educational purposes to learn about neural network-based NLP
Building custom NER models for domain-specific text datasets
Integrating NER into pipelines that support CoNLL or BRAT formats

Not Ideal For

Projects needing real-time, low-latency NER inference due to computational overhead of neural networks
Teams requiring out-of-the-box support for non-English languages or modern transformer architectures like BERT
Environments restricted to TensorFlow 2.x or PyTorch, as NeuroNER depends on TensorFlow 1.x
Applications demanding extensive API or GUI interfaces beyond command-line and basic Python integration

Pros & Cons

Pros

Benchmark-Matching Accuracy

Achieves F1-scores around 0.90 on CoNLL-2003, comparable to state-of-the-art systems as highlighted in the README.

Straightforward Command Line

Runs via simple commands like 'neuroner' for training and prediction, making it accessible without deep learning expertise.

Multiple Format Support

Accepts both CoNLL-2003 and BRAT annotation formats, easing integration with existing NLP pipelines and tools.

Domain-Specific Pretrained Models

Includes ready models for clinical text (e.g., i2b2, MIMIC) and general news, saving time on training from scratch.

Cons

Outdated Framework Dependency

Relies on TensorFlow 1.0+, which is no longer actively maintained, posing compatibility issues with newer libraries and systems.

Cumbersome Initial Setup

Requires manual downloads of word embeddings, datasets, and models in multiple steps, increasing risk of errors and frustration.

Limited Architectural Updates

Uses LSTM-CRF models without native support for modern transformers, potentially lagging behind current NER advancements.

Frequently Asked Questions

Related Projects

HuggingFace Transformers

🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.

结巴中文分词

Last commit1 year ago

spacy

💫 Industrial-strength Natural Language Processing (NLP) in Python

Stars33,637

Forks4,686

Last commit20 days ago

Haystack

Open-source AI orchestration framework for building context-engineered, production-ready LLM applications. Design modular pipelines and agent workflows with explicit control over retrieval, routing, memory, and generation. Built for scalable agents, RAG, multimodal applications, semantic search, and conversational systems.

Stars25,487

Forks2,832

Last commit3 days ago

Community-curated · Updated weekly · 100% open source

Found a gem we're missing?

Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.

Submit a project Star on GitHub