Flair: Pretrained medtext

[prabhatM]

Hi,
Flair looks interesting. Any pretrained model for medical text NER?

[alanakbik]

We don't currently have one, but would be very interested to see how Flair works on a biomedical task. We believe a character-level model would work really well here.

I think @tmills is working on something like this and has done a comparative evaluation that he'll share soon (#173).

Anyways, if you are interested in training an LM for medtext, we'd be happy to hear about your progress!

[tmills]

Hi @prabhatM, as @alanakbik mentioned I have been testing flair with some biomedical tasks, but I'm specifically focusing on extracting information from clinical questions right now. I have not tested yet on straightforward NER task like detecting diseases, signs/symptoms, etc., but if you're interested I'd be happy to share what I know and/or work together on an evaluation.

[prabhatM]

Dear Alan,

Thank you for responding to my query. I am working on analysis of free
clinical text.
After a few testing, I am trying to adopt an NER discovery process and for
that I am looking for a pre-trained model. If I come across, I will update
you.

Thank you for @tmills reference.

Regards

Prabhat MOHANTY

On Thu, 27 Dec 2018, 15:34 Alan Akbik <[email protected] wrote:

We don't currently have one, but would be very interested to see how Flair
works on a biomedical task. We believe a character-level model would work
really well here.

I think @tmills https://github.com/tmills is working on something like
this and has done a comparative evaluation that he'll share soon (#173
https://github.com/zalandoresearch/flair/issues/173).

Anyways, if you are interested in training an LM for medtext, we'd be
happy to hear about your progress!

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
https://github.com/zalandoresearch/flair/issues/331#issuecomment-450120010,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AG6jXD8g4pIKwtvlpGv6MIEsCDCCH85Wks5u9JsQgaJpZM4ZhRDR
.

[prabhatM]

Dear Tim,

It's great to know that you are working in the bionlp domain. I recently
stumbled upon Flair and am hugely impressed by its premise.

I too have interest in diagnostic clinical questions. But I started with
identification the disease and findings from the free clinical text. I was
thinking of training multiword tokens with clinical terminologies. I am
sure you would be needing the same.
I am so eager to know about your work. Please point me to some links or
share a little. I would love to share whatever resources I come across
with you.

Regards

Prabhat MOHANTY

On Fri, 28 Dec 2018, 02:20 Tim Miller <[email protected] wrote:

Hi @prabhatM https://github.com/prabhatM, as @alanakbik
https://github.com/alanakbik mentioned I have been testing flair with
some biomedical tasks, but I'm specifically focusing on extracting
information from clinical questions right now. I have not tested yet on
straightforward NER task like detecting diseases, signs/symptoms, etc., but
if you're interested I'd be happy to share what I know and/or work together
on an evaluation.

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/zalandoresearch/flair/issues/331#issuecomment-450228752,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AG6jXJ-slcK4Qv1Yybz2BdRbm5Z6xb5Jks5u9TKWgaJpZM4ZhRDR
.

[tmills]

@prabhatM please take a look at my github profile for links to my research page. I work on apache ctakes and do lots of NLP/information extraction on medical texts. Feel free to email me at the address there to follow up so we don't totally derail the issue thread :).

[prabhatM]

Hi Tim,

Thank you for your response. I will follow your lead.

Happy New Year.

Regards

Prabhat MOHANTY

On Tue, 1 Jan 2019, 19:22 Tim Miller <[email protected] wrote:

@prabhatM https://github.com/prabhatM please take a look at my github
profile for links to my research page. I work on apache ctakes and do lots
of NLP/information extraction on medical texts. Feel free to email me at
the address there to follow up so we don't totally derail the issue thread
:).

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/zalandoresearch/flair/issues/331#issuecomment-450730134,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AG6jXFFI09OADm5lQRG9xq2LMXIGHUsGks5u-2gBgaJpZM4ZhRDR
.

[alanakbik]

Release 0.4.1 has several pre-trained embeddings (Flair and ELMo embeddings) for medical text. So closing this thread, but feel free to reopen if you have more questions!