Introduction to NLP | Word Embeddings & Word2Vec Model

[@harshpal1457]

The most underrated channel for Machine learning

[@sreedharsree361]

The best.. you elucidated this topic with charm !! Thanks Sujan

[@NormalizedNerd]

Glad it helped you...Keep supporting ❤️

[@magelauditore333]

Commenting after 1/3rd part f the video. It is really very clear. Wait up and continue this. You will get huge lots of subs. Keep it up

[@NormalizedNerd]

Thank you so much...it means a lot :D

[@TechResearch05]

Very clear description. I was struggling to understand it but your video was very simple and provided required information

[@NormalizedNerd]

I'm glad to hear that. Keep supporting. ❤️

[@ZohairAhmed007]

Thanks, that is one of the best explanations, I Understand a lot.

[@shivanineeli5392]

please continue making NLP videos , please we want more and more if possible entire AI we would love u hear from you!

[@NormalizedNerd]

Keep supporting!

[@debjyotibanerjee7750]

Really good explanation, now understood the concept completely!!!!!!!

[@NormalizedNerd]

Glad this video was helpful. Keep supporting man!

[@manikant1990]

Beautiful Explanation, I love it!! 👍👍

[@NormalizedNerd]

Thank you! 😃

[@swagatmishra9350]

Really awsum vdo.... So easy and clear explanation... Loved it.. Please make more vdos.. Thanks a lot..

[@NormalizedNerd]

I'm glad that you loved it. More videos are coming :D

[@pratibhagoudar6817]

Thanks bruhhh🤍.... it's more clear dan compare regular classes # nlp

[@NormalizedNerd]

Great to hear that :D

[@mastercomputersciencewitha5985]

Very Nice explaination sir. Thank you so much sir.

[@dipannita7436]

one of the best explanation

[@NormalizedNerd]

Thanks a lot!

[@jjtalks6797]

Super explanation

[@vaidehideshpande1489]

great explanation!!

[@NormalizedNerd]

Thank you!

[@rajivs9287]

awesome video

[@NormalizedNerd]

:D

[@spartacuspolok6624]

It was really helpful.

[@NormalizedNerd]

Thanks!

[@meetsavsani9739]

Great work buddy!

[@NormalizedNerd]

Thanks a lot!

[@aniketchhabra8912]

This is amazing!!

[@edwardrouth]

Hi, This could sound bit naive but i just want to know how did you figure out the parameter that you are passing to "api.load()" which is "word2vec-google-news-300". I mean there must be a list of API from where you got this right ? I googled it but i found there are just links and its bit confusing too. Thanks.

[@NormalizedNerd]

We were all naive once so don't worry. I'm using gensim api. So you can find the correct parameters from their documentation/repo. Here you can find a file called list.json => github.com/RaRe-Technologies/gensim-data You can also find the list of the models from the GitHub readme.

[@tobiascornille]

You said Skipgram predicts the context words from the target word, but then later you just compute the sigmoid (so not a softmax) to know if one pair of a target word and a context word is correct. I don't really see how this is "predicting" the context words. Is there something else going on? I'm very confused since it seems like every explanation is saying something different...

[@NormalizedNerd]

Because, we are taking one context (or random) word at a time and pair it with the target word. If we get context-target pair then class is 1. If we get random-target pair then class is 0.

[@intelligentinvesto9060]

What is the loss function used?

[@NormalizedNerd]

- log[P(w_target | w_context)

[@coxixx]

please make a video about how back propagation works in skip gram.

[@NormalizedNerd]

Backpropagation in the word2vec model is really hard to explain in a single video however, I found a great resource to learn about it. www.claudiobellei.com/2018/01/06/backprop-word2vec/ I hope one can understand this article after watching the video.

[@rexwan561]

intro video remind me its wednesday my dudddde

[@hemangshrimali6308]

Nice video

[@ccuuttww]

I want to ask a question is it all vector of words in same length? Because I have an idea if we use DNA sequence(of course not in same length) instead of just words can we train a model to get a better classify result?

[@NormalizedNerd]

The length of each word vector is same. Because the idea behind word2vec was to represent every word using vectors of fixed length.

[@ccuuttww]

@@NormalizedNerd can we fit DNA sequence into it? I know we can fit image into it

[@md.shafaatjamilrokon8587]

Thanks man

[@ThoTran-oi3xi]

Thank you so much for your video, can you turn on subtitles for this video? Because I'm not from England, I can't hear you clearly but the video has no subtitles

[@NormalizedNerd]

Currently, I don't have the resources to put subtitles on every video. However, I'll try to do it for some videos.

[@r_pydatascience]

Nice video.Does word2vec represent medical vocabularies? I have a medically text corpus that has about 100000 tokens. What do you think I should do?

[@magelauditore333]

Just Awesome.

[@NormalizedNerd]

Thanks again!

[@krishcp7718]

Hi, Your videos on NLP are great. For , most_similar(positive = ['boy', 'queen'], negative='girl', topn=1) I am getting : [('teenage_girl', 0.35459333658218384)]. What could be happening here? Krish

[@NormalizedNerd]

Strange!...that shouldn't happen. Please check your code. You can download my notebook and run it. The link is in the description. Edit: Ohh, I get it. You used "girl" instead of ["girl"]. Interesting...I didn't know it behaves like this when just a string is passed :o

[@debjyotibanerjee7750]

Bro just tell me one thing, while creating vectors of the words, do we need to remove stopwords, and lemmatize our text data, cause I believe if we do the mentioned text pre-processing steps, then may be the word2vec model may be not able to understand the context, and the training will not happen properly. If you could say something, that would help me a lot in my project.

[@NormalizedNerd]

Great question. TBH it depends on the project you are working on. Google's Word2Vec doesn't implement lemmatization (also removes very few stop words) so if you are planning to use that then don't lemmatize. But if you are gonna train your word2vec then you can do all sorts of preprocessing. A rule of thumb is if your data size is very large then don't lemmatize. For stop words I'd say remove only the ones that doesn't change the context very much (like a, an, etc.)

[@debjyotibanerjee7750]

@@NormalizedNerd okay.. Thank you for the information bro. Are you there on LinkedIn?

[@NormalizedNerd]

yeah...here's my profile www.linkedin.com/in/sujandutta99/

[@gulsanbor]

excellent

[@NormalizedNerd]

Thank you...keep supporting

[@gulsanbor]

@@NormalizedNerd sure . Connect with me in Linkedin if possible www.linkedin.com/in/gulsan19/

[@coxixx]

awesome

[@NormalizedNerd]

Thank you :)

[@Hephasto]

Shouldn’t it be grammatically correct I love making videos instead of to make