Introduction to NLP | GloVe & Word2Vec Transfer Learning

[@WahranRai]

It is not G love model : it is Glove (glo ve) model ( related to GLObal VEctor)

[@shrutiiyyer2783]

Please keep going in the NLP section! your videos are great!

[@NormalizedNerd]

Thanks! Will do!

[@ecofriend360]

Very nicely explained👌

[@RS-vu5um]

Can you please do a video on BERT using Keras for Sentiment Analysis?

[@codepathsala]

Explained very nicely. Please tell how can we make the pipeline for prediction. Do we need yo apply pretrained word2vec in this case as well ?

[@dianafebrihatin7518]

Hi, I'm so new to this topic and programming itself. On 14:43 by the compilation you used 'adam' as optimiser. What does this adma optimiser do? I have zero clues on it and apparently you didn't explain :(

[@NormalizedNerd]

Basically, an optimizer is an algorithm that solves the optimization problem (minimizing the loss/cost function). I guess you are familiar with Gradient Descent, the most common optimization algorithm. Adam is just an advanced algorithm to solve the optimization problem.

[@dianafebrihatin7518]

@@NormalizedNerd thank you!!! I didn't expect you'd answer to my question so quickly. Another question would be, would you consider to also explain about any of LSA, FastText, ELMo, and Bert in the near future?

[@NormalizedNerd]

Yeah some of them are in my bucket list!

[@Gogeta0991]

Thanks for the video! I am trying to take word vectors from Spacy and update them by training on new data from a specific domain. My goal is to change the vectors affiliated with the new domain and to not change the other vectors much. Is it possible with this approach? Thanks!

[@mdsharifuddinkhan7064]

Is it under the transfer learning, machine learning or deep learning?

[@NormalizedNerd]

Both GloVe and Word2Vec are trained using neural networks hence they come under deep learning. Here, we are using the pre-trained versions of the models hence transfer learning.

[@abxy2197]

Can I use CNN and logistic regression? If yes, please let me know how!

[@NormalizedNerd]

Yes. We can apply CNN on text data. Maybe I'll make a video about it in future.

[@rohitjain9331]

The video was awesome!!! But I have one doubt, why we initialized embedding matrix dimension with 'len(word_index)+1' instead of just 'len(word_index)' ??

[@NormalizedNerd]

Great point. The extra one row is for the words that are not present in the vocabulary. This row will contain only 0's.

[@adityadarak1935]

i am getting error 'word_index' not defined. can you explain

[@mahery_ranaivoson]

Awesome content bro. Keep it up 👆

[@NormalizedNerd]

Thank you bro

[@md.tarekhasan2206]

Can't download pre-trained word vectors. :(

[@NormalizedNerd]

Try these links: nlp.stanford.edu/projects/glove/ github.com/mmihaltz/word2vec-GoogleNews-vectors Also I believe using gensim api you can download these very easily.

[@abxy2197]

My model is not giving accuracy above 40%. What should I do? Can you tell me how to predict from the model?

[@NormalizedNerd]

You can watch this video(towards the end) to know how to predict a single(new) example. kzbin.info/www/bejne/pn7NhWt9fr1jmdE Btw what model and dataset are you using? Accuracy shouldn't be that low.

[@abxy2197]

@@NormalizedNerd I tried LSTM, CNN, and also LSTM+CNN on dataset "text_emotion.csv" from data world data.world/crowdflower/sentiment-analysis-in-text But my model won't give accuracy above 40 :( Pls help

[@preetamsoni5141]

g-love

[@NormalizedNerd]

haha...I should have pronounced it like Glo-ve but I like G-love more XD