Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Classifying 200k articles in 7 hours using NLP

salt.agency

Post reply on HN
Print view View original

21–30 of 32 posts

Re: Classifying 200k articles in 7 hours using NLP

#21

Post by Der_Einzige »

jointpdf wrote: post #20

For those interested in related/alternative approaches, one or more of the following established open-source libraries might appeal to you: - Snorkel (training data curation, weak supervision, heuristic labeling functions, uncertainty sampling, relation extraction): https://github.com/snorkel-team/snorkel - AllenNLP (many pretrained NLP research models for tasks beyond text classification, model training and serving,…

Kill the LSA/NMF middle-man and use UMAP directly. It supports sparse (tf-idf) vectors.

Re: Classifying 200k articles in 7 hours using NLP

#22

Post by jointpdf »

Der_Einzige wrote: post #21
jointpdf wrote: post #20

For those interested in related/alternative approaches, one or more of the following established open-source libraries might appeal to you: - Snorkel (training data curation, weak supervision, heuristic labeling functions, uncertainty sampling, relation extraction): https://github.com/snorkel-team/snorkel - AllenNLP (many pretrained NLP research models for tasks beyond text classification, model training and serving,…

Kill the LSA/NMF middle-man and use UMAP directly. It supports sparse (tf-idf) vectors.

True, good point. That may be better for classification performance. But at least for visualization and interpretability purposes using NMF is extremely simple and versatile (e.g. you can induce sparsity in the representation, setting the rank to be artificially low can cause high-level structure to “pop out”). That is, it gives you a few more knobs to turn than UMAP alone.

Re: Classifying 200k articles in 7 hours using NLP

#23

Post by aledalgrande »

smeeth wrote: post #15

This article is a bit of a Frankenstein monster. There are too many possible target audiences for ML blogs and it seems like this post made an attempt to please everybody. This isn't a condemnation of the author, its just an impossible task. 1. Experienced ML practitioners will be unimpressed with the ML task generally (simple problem, no comparison with common models, no use of common dataset, no lit review) and wis…

5. CIOs will be ecstatic and go tell their team that they have to do NLP

Re: Classifying 200k articles in 7 hours using NLP

#24

Post by nxpnsv »

jointpdf wrote: post #20

For those interested in related/alternative approaches, one or more of the following established open-source libraries might appeal to you: - Snorkel (training data curation, weak supervision, heuristic labeling functions, uncertainty sampling, relation extraction): https://github.com/snorkel-team/snorkel - AllenNLP (many pretrained NLP research models for tasks beyond text classification, model training and serving,…

Spacy link is for allennlp!

Re: Classifying 200k articles in 7 hours using NLP

#25

Post by jointpdf »

nxpnsv wrote: post #24
jointpdf wrote: post #20

For those interested in related/alternative approaches, one or more of the following established open-source libraries might appeal to you: - Snorkel (training data curation, weak supervision, heuristic labeling functions, uncertainty sampling, relation extraction): https://github.com/snorkel-team/snorkel - AllenNLP (many pretrained NLP research models for tasks beyond text classification, model training and serving,…

Spacy link is for allennlp!

Oops good catch. I should really stop writing posts on my phone. That’s especially sad since their docs are so good. I can’t edit it now, but: https://spacy.io/usage/spacy-101

Bonus—An excellent interactive Spacy course from Ines Montani (also includes a template to build similar courses!): https://github.com/ines/spacy-course

Re: Classifying 200k articles in 7 hours using NLP

#26

Post by mobilio »

rezamoaiandin wrote: post #3
sixhobbits wrote: post #2

The title makes it sound like they talk about how they did it so efficiently. But all the info we get about that is "The last step was to combine the four binary models into one multiclass model, as explained in the previous section, and use it to classify 1M new documents automatically. To do this, we simply went on the UI and uploaded a new list of documents." Great intro to NLP article, but very light on the actua…

Given that is of interest, I'll do a follow up on the implementation details and dataset!

I also would like to take a look on code and all input files.

Re: Classifying 200k articles in 7 hours using NLP

#27

Post by 55555 »

rezamoaiandin wrote: post #3
sixhobbits wrote: post #2

The title makes it sound like they talk about how they did it so efficiently. But all the info we get about that is "The last step was to combine the four binary models into one multiclass model, as explained in the previous section, and use it to classify 1M new documents automatically. To do this, we simply went on the UI and uploaded a new list of documents." Great intro to NLP article, but very light on the actua…

Given that is of interest, I'll do a follow up on the implementation details and dataset!

Can you explain what you actually... did? What were the four categories? How did your client make money from this?

Re: Classifying 200k articles in 7 hours using NLP

#28

Post by pmayrgundter »

Glad to see work like this being shared!

There are some well known text classification datasets, e.g. the Reuters news dataset from David Lewis of Bell Labs:

  http://www.daviddlewis.com/resources/testcollections/reuters21578/
More background here:

  https://link.springer.com/content/pdf/bbm%3A978-3-642-04533-2%2F1.pdf
Here's a result from ReelTwo's Classification System circa 2003 (Based on a bayesian learner; related to the U Waikato WEKA ML system) if you'd be up for comparison:

  https://web.archive.org/web/20040606002449/http://www.reeltwo.com/datasets.html
10 categories 2,535 documents 15 build time (~170 docs/sec; these were short news abstracts; see pdf for example) 0.9121 F-measure

Build Time is the time to load, model and evaluate (using Leave-One-Out evaluation) a dataset on a WinXP/1GHz Celeron/256MB computer. F-Measure is the micro-averaged F-Measure across all categories in the dataset.

Re: Classifying 200k articles in 7 hours using NLP

#30

Post by whymauri »

kgarten wrote: post #29

Maybe off topic ... Is Stanford ML expert some type of accreditation? How do you become a Stanford ML expert? :) Attending the (excellent) Stanford ML online course on Machine Learning or do I have to read an ML book on Stanford Campus?

They have masters degrees from Stanford. I agree it's a bizarre accredidation since they have a few years of industry experience otherwise, which is, IMO, more relevant.
Post reply on HN