5. Sentiment Analysis

[@ryankrueger6594]

Love the channel. One small recommendation -- I prefer having the coding and the conceptual summaries in separate videos. I typically want to firm up the conceptual understanding in an isolated manner before diving into the code.

[@韩松-q1s]

Really love your lesson. From an undergraduate who struggled to do some nlp work for graduation project.

[@WeightsBiases]

Thanks for watching, we're glad it was helpful!

[@diagorasofmel0s]

please keep this going, i know i am late but im sure you'll be popping soon

[@daniel_partida]

Thank you for the excellent tutorial. Do you have a model recommendation to perform sentiment analysis with unlabeled data?

[@romanlodeizen9095]

Found this channel today keep up the great work, i have some homework to do

[@samratkorupolu]

Thankyou so much Lukas, for explaining issues we might run into with sentiment analysis

[@allanboutilier2724]

This is a really great video, thanks for sharing this information. You explain it very well

[@thelastone1643]

I have the sentiment analysis review task of two classes and after the machine algorithm classification model applied , the model learnt which words (features) indicate to positive review and words (features) indicate to negative review. is there any way to know what are the words (features) that model used to indicate to positive review and the words (features) that model used to indicate to negative review. what are the weight the model gave it to each word ( feature )?

[@ayushthakur3880]

Naive Bayes is an interpretable model. Check out this post: lilianweng.github.io/lil-log/2017/08/01/how-to-explain-the-prediction-of-a-machine-learning-model.html#naive-bayes You will get the theoretical sense of doing this. Sklearn provide this API to get the parameters. scikit-learn.org/stable/modules/generated/sklearn.naive_bayes.MultinomialNB.html#sklearn.naive_bayes.MultinomialNB.get_params

[@DanielWeikert]

Thanks. Like your channel. Found it today. Could you elaborate on how to deal with Emojis? Are there other ways then n-grams e.g. we decide 2-grams and then have text like "not that good" Best regards

[@ayushthakur3880]

Maybe something like TF-IDF can be useful here. It's a simple technique to convert text to vector. And then this vector can be used as a feature for the classifier. To use emojis maybe one can use the unicode of each emoji and assign some value to it.

[@thelastone1643]

please explain all Machine and deep learning topics

[@PetroRyndych]

your videos are very useful, keep doing it

[@thanigai838]

Can you to video about. Text classification for unlabeled data

[@abdulazizsalman6197]

you are really awesome, i wish to be like you one day

[@jamesang7861]

lol! only positive feedbacks for you ! thank you!

[@alainmuller5359]

Thanks a lot for making these videos available. Something is bothering me though : the fact that you include your test data when you do feature extractions with CountVectorizer(). Shouldn't you work only with training data at the beginning ?

[@ayushthakur3880]

Hello Alain, in machine learning the model accepts the same data modality to give prediction as it is trained on. For this text classification the model is trained on feature vector computed by CountVectorizer(). Thus he converted the test data to the same input modality. I hope you get the idea here.

[@charles_irl]

@@ayushthakur3880 Actually, it's not great practice to set the parameters of your pre-processing using both the training data and the test data. it can increase over-fitting, because information from the test set is leaking into your model through the pre-processing step. You're right that you still need to _apply_ your pre-processing on test data (and in production) but that doesn't mean you should use the test data before that point. In the specific case of a CountVectorizer, though, this effect is probably minimal, since it just means that words that only occur in the test set are included in your features.

[@maxhansen5166]

in line 42 it says: there are no line numbers visible on the screen!

[@Sagaofsr]

Thank you

[@thelastone1643]

you are amazing