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

Show HN: Natural Language Processing Demystified

news.ycombinator.com

Post reply on HN
Print view View original

31–39 of 39 posts

Re: Show HN: Natural Language Processing Demystified

#31

Post by axiom92 »

brooksbp wrote: post #10

Thank you for sharing this! I am currently studying NLP.. Along the way, I've been struggling with a question and I hope someone can help me understand how to go about this: how would you build a model that does more than one NLP task? For a simple classifier like input: text (a tweet) and output: text (an emotion), you can fine-tune an existing classifier on such a data set. But, how would you build a model that doe…

If you have the dataset, you can try to train a model like T5 [1], notebook [2].

You just need to create [(input, output)] examples in the format you want.

For example

[(a Yelp review of a restaurant, [("tacos", "good"), ("margaritas", "good"), ("salsa", "bad")]].

With enough data, the model should be able to learn to generate the output in the right format.

> Python list of tuples

Things get interesting if you want to generate actual Python code. You can use a large language model with just a few examples of the task to generate such code. For example, see https://reasonwithpal.com/.

Happy to answer more questions!

[1] https://huggingface.co/docs/transformers/model_doc/t5

[2] https://colab.research.google.com/github/huggingface/noteboo...

Re: Show HN: Natural Language Processing Demystified

#32

Post by gattilorenz »

brooksbp wrote: post #10

Thank you for sharing this! I am currently studying NLP.. Along the way, I've been struggling with a question and I hope someone can help me understand how to go about this: how would you build a model that does more than one NLP task? For a simple classifier like input: text (a tweet) and output: text (an emotion), you can fine-tune an existing classifier on such a data set. But, how would you build a model that doe…

Yours is an example of aspect-based sentiment analysis. Typically it has been tackled in two steps: first extract the aspects, then classify them as positive/negative. GPT or T5 are possible options for doing both in one go, but splitting the task seems to be still a good option [1].

[1] http://essay.utwente.nl/91778/1/Middelraad_BA_EEMCS.pdf

Re: Show HN: Natural Language Processing Demystified

#33

Post by posharma »

mothcamp wrote: post #29
posharma wrote: post #26

Who is the intended audience for this course? Is it application developers looking to use NLP in their apps? Or machine/deep learning devs?

It's for anyone who wants to learn NLP such that they get (a) an understanding of what's going on under the hood and (b) knowledge of how to get stuff done. So the ideal outcome is someone who gets an end-to-end view from theory/concept to implementation. If someone just wants to learn how to use tools/frameworks, I'd stick to the Colab notebooks. If someone's already experienced in ML and wants to learn something NL…

Thanks. Excellent course.

Re: Show HN: Natural Language Processing Demystified

#35

Post by benjismith »

This is awesome! I just finished watching the Unit 10 video ("Neural Networks I") and it filled in quite a few gaps in my understanding.

I really love that you build a complete working example, all the way down to the matrix multiplications, so that we can see how everything works, at every layer of abstraction.

I'm looking forward to the next unit, and I can already tell this is going to be an indispensable reference I'll come back to review again and again.

Thank you!!

Re: Show HN: Natural Language Processing Demystified

#37

Post by escanor »

great work! just a note regarding tf-idf, when you mention log10: i think you're missing the point on the reason of log and most importantly base 10. namely, using log10 gives us a perspective on the number of digits of the term/document frequency. if a term "A" occurs 23 times and a term "B" occurs 50, they will have a very close representation (because both numbers are 2 digits ones).

anyway, thanks for the submission

Re: Show HN: Natural Language Processing Demystified

#39

Post by reichardt »

mothcamp wrote: post #15
reichardt wrote: post #14

Earlier quoted context omitted.

Yes, it's easy to see you put a lot of thought into that. I hope your course receives much more exposure. When I first found your videos a few weeks ago, I was surprised how few views they have given to the quality of the course. Do you record the voice track of your videos yourself? Glad to see you published the final lesson about transformers. Was looking forward to that!

I did record all voice tracks, yeah. If I do this again, I'll probably use a lot of generative tools now. :-D Hope you find the transformers module useful!

That's impressive. The audio track of your videos is so clean and well understandable that I was wondering if you used a studio setup or voice synthesis software. Well done!
Post reply on HN