Enhancing Named Entity Recognition (NER) Accuracy in Large Descriptions

[santhoshlinga]

Hello Spacy team,

Firstly, I would like to express my gratitude for providing such an excellent product to the open-source community. Spacy has been instrumental in our project, particularly in extracting information from large descriptions for each record.

In our use case, we developed a Named Entity Recognition (NER) system to extract relevant information by annotating based on our specific requirements. The results were promising, but we encountered some instances where the predictions didn't meet our expectations.

We're reaching out to the community to discuss strategies and techniques for improving the accuracy of our NER system. We believe that by collaborating with fellow users and experts, we can identify areas for enhancement and fine-tune our approach to better suit our needs.

We also have a specific question: How can we update the model when we encounter predictions that are completely wrong? Additionally, how can we streamline this updating process to ensure efficiency and accuracy?

We welcome any insights, suggestions, or experiences related to improving NER accuracy, whether it's through fine-tuning models, incorporating additional training data, or adopting alternative methodologies.

Looking forward to helpful discussion!

Warm regards,
Santhosh L

[santhoshlinga]

@svlandeg is their any possibility of finding what I am looking for? Your assistance and support in this matter are greatly appreciated, and I am grateful for your dedication to helping the community.

Could you please let me know if there is a possibility of fulfilling my request? Your expertise and insights would be invaluable in guiding me in the right direction.

Thank you for your time and consideration. I look forward to hearing from you soon.

[santhoshlinga]

Could someone please confirm whether the described use case is feasible?

[svlandeg]

Hi @santhoshlinga, please refrain from pinging individual people on this discussion forum. We already spend a lot of time and energy answering questions on this forum and we don't give priority to people who ping maintainers, as this makes our GitHub mentions almost unusable and creates a bigger burden on us.

[santhoshlinga]

Agree.

By any chance can you point me in the right direction?