2
$\begingroup$

What is the current state-of-the-art for pos tagging and named entity recognition for twitter data? Are industrial-strength programs like Spacy and SparkNLP accurate for such texts? How about FlairNLP and Stanford's CoreNLP accuracy measures?

$\endgroup$
2
  • 1
    $\begingroup$ this is good question however it borders on being opinion-based $\endgroup$ – Nikos M. Jul 28 '20 at 11:50
  • 1
    $\begingroup$ There are metrics to determine SOTA, about spacy and other modules I agree with you, it's a mostly a matter of personal preference and/or contingent project needs. $\endgroup$ – Leevo Jul 28 '20 at 12:02
6
$\begingroup$

SOTA is changing so rapidly in NLP that even Data Science professionists struggle to cope with it. I have two main sources that I constantly check to gain some insights on SOTA:

  • NLP Progress from Sebastian Ruder. It contains updates on NLP on a whole lot of subfields, NER and POST included.

  • Paper with code contains a section on NLP. That's a great website for ML in general.

I know these links do not tackle the problem of Twitter specifically, however I don't think that domain is qualitatively different from others. IMO, of course.


About your other question:

Are industrial-strength programs like Spacy and SparkNLP accurate for such texts? How about FlairNLP and Stanford's CoreNLP accuracy measures?

As I wrote above, it's mostly a matter of personal preference and/or contingent project needs. There's no right or wrong tool. Personally, I found Stanford tools to be the best, for either the quality of their predictions and the amount of models available from a single pipeline. But as I said it's very subjective.

$\endgroup$
4
  • $\begingroup$ Thanks @Leevo. I have been following NLP Progress site as well. Given the unique challenges that Twitter text presents, I was wondering about whether any of these models have been tested and how they fare. $\endgroup$ – SanMelkote Jul 28 '20 at 14:03
  • 1
    $\begingroup$ You're welcome, I'm sorry I don't know of sources specifically on Twitter models. What makes it different from more general NLP tasks? (I'm just curious, I'm goinig to work on Twitter data very soon.) $\endgroup$ – Leevo Jul 28 '20 at 14:50
  • 1
    $\begingroup$ This presentation details the difficulties with Twitter data: gate.ac.uk/sale/talks/gate-course-jun16/module-2-socmed/5.pdf $\endgroup$ – SanMelkote Jul 28 '20 at 15:48
  • 1
    $\begingroup$ Also check this: derczynski.com/sheffield/papers/ner_issues.pdf $\endgroup$ – SanMelkote Jul 28 '20 at 16:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.