The app i’ve deployed can do Named Entity Recognition (NER) and Part-of-speech tagging (POS) at once, using the terrific Flair NLP Framework from Zalando Research department in collaboration with Berlin University.
Code is available on my Github account, with under 70 lines of Python hopefully worth checking out.
Feel free to give feedback on the App. It wouldn’t be possible without the community in the first place! so you’re thoughts are most welcome.
I will write a blog with more detailed info as soon as I can find the time
It is showing This app has gone over its resource limits. Please try again in a few minutes.
Kindly refer this for checking the previous post which had similar issue hopefully it will help you.
Long story short: the app uses 2 saved .pt (Torch) models. It runs locally just fine, however the recursive nature of Streamlit proved to be causing a memory leak, since every time a new input is given, a new model instance is loaded in the RAM.
I simply didn’t notice on my laptop, since i have quite a lot of RAM. However, I did end up crashing my laptop just as well, it only took quite some time.
Thanks to the community, SessionState.py has now been added and a cached version does prevent this unexpected behaviour.
It is still far from perfect, so please try again an let me now your honest feedback
The link is working now. I tried your project. It is amazing and works fine. However, I am not an expert to the related field but even a small initiative should be appreciated as you tried something out of your comfort zone.
Thank you for your feedback.
In order to answer your question:
spacy.explain("PRP$")
pronoun, possessive
Furthermore,
Functional documentation of the models used can be reviewed @ the links below (still thinking how to integrate this nicely in the App, suggestions are most welcome)
I’ve updated the App to accommodate Streamlit >= 1.8. Since I had the code under review in over a year, I also did some updating . Firstly as FlairNLP improved the Tagger functionality, it also supports Span tagging. Really useful if you have complex documents, that can easily have multiple spans of meaning across sentences.
Secondly I’ve received user feedback that the abbreviations where not clear, hence that’s fixed as well, input taken from the revised modelcards @HuggingFace.
If you care to take a look and share you’re thoughts, that would be awesome.
