IndicNER is a state-of-the-art multilingual Named Entity Recognition (NER) model developed by Bhashini. It is designed to recognize and classify named entities such as names of persons, organizations, locations, dates, and more from text in 11 Indian languages:

Hindi, Bengali, Tamil, Telugu, Gujarati, Punjabi, Marathi, Assamese, Kannada, Malayalam and Oriya.

Training Dataset:

The model is fine-tuned using a large corpus derived from publicly available Indian NER datasets and human-annotated test sets, ensuring high accuracy across different languages. Additionally, it has been trained on data sourced from the Samanantar Corpus, India's largest parallel corpus, to enhance its contextual understanding. The base model used for fine-tuning is BERT-base-multilingual-uncased, which allows it to capture linguistic nuances effectively.

Use Cases:

IndicNER can be used for a wide range of Natural Language Processing (NLP) applications, including:

1. Automated document processing – Extracting key entities from government, legal, and business documents.

2. Chatbots and virtual assistants – Enhancing conversational AI by identifying user queries related to people, places, and organizations.

3. News and content analysis – Automatically tagging and categorizing entities in multilingual news articles.

4. Healthcare and medical records – Identifying patient details and medical terms for structured data extraction.

For more details and implementation, visit: https://huggingface.co/ai4bharat/IndicNER.