Sarcasm is Very Easy to Detect! GloVe + LSTM

[@garrettosborne4364]

Excellent. I am working on GloVe model and this adds some insights to that project.

[@NormalizedNerd]

That sounds great! Keep supporting :D

[@datahacker1405]

Your videos deserve more likes , you deepdive into the concepts which is really helpful for us

[@NormalizedNerd]

Thanks a lot :D

[@chetan923]

Nice advertising skills bro, I was up to a project on sentiment analysis and thought tf-idf would suffice, now you made me watch all your videos to understand your references and guess what, GLoVe is again interesting like anything new I come across in data. I guess I'll have to work on the same to understand more.

[@NormalizedNerd]

Do read more about it!

[@ChinmayiHR-y3s]

Bro which algorithm you used in this project

[@adamameen5]

Great video! Can I know how you are performing classification in this @Normalized Nerd

[@missmad5789]

Hello :) thank you so much for this video! I'm doing this same project at the moment and I have a question for you: After looking at both datasets and checking for duplicates I realized that both data sets seem to contain nearly the same exact headlines. This means adding both data sets into one dataframe, it will contain thousands of duplicates. Does that worsen the models performance or does it somehow help the model to learn better? I'm new to ML :D thank you

[@missmad5789]

by the way, I love your energy in this video! :)

[@NormalizedNerd]

Great point! You should remove the duplicates. I forgot to mention this step in the video.

[@adamameen5]

@@NormalizedNerd Can someone tell how to remove the duplicates.. this is a great video, thank you!

[@sindhu2410]

Is there any pretrained model or transformation model to detect sarcasm?

[@paulkerrigan9857]

Sarcasm is entirely dependent upon contextual clues, though. Two identical sentences could be genuine or sarcastic depending upon the situation. Is it really a problem that a neural network could solve by looking at strings?

[@NormalizedNerd]

If you look closely, some of the strings that I used in the examples do contain the context. If the corpus contains the context then yes it is possible for a neural network.

[@paulkerrigan9857]

@@NormalizedNerd I see. Thanks for your response.

[@-_BahauddinTaha]

Can we do it for other language, using the same procedure??

[@atreyamajumdar9836]

Very nice video! But, if you are removing words like 'to', 'the' etc, isn't it also affecting the meaning of the sentence? And that can cause some error?

[@NormalizedNerd]

@Atreya Majumdar good point. I actually tried including the stop words but the validation accuracy didn't improve mush. I guess the data is not so rich in that respect. However you can be creative and try something with parts of speech detection.

[@atreyamajumdar9836]

@@NormalizedNerd, I don't have any experience with NLP. Still- is it possible that the presence of the second sentence in an example is teaching the network to classify it as a sarcasm? Maybe you can check this easily by calculating the proportion of sarcasm sentences with two full stops in it, in the training set.

[@NormalizedNerd]

@@atreyamajumdar9836 As I am removing every punctuation (including full stop) so the model doesn't know how many sentences are there. It just receives a sequence of words.

[@abhijeetv4418]

Very nice bro...Can you please help me out reach the code? Can you please share it somehow?

[@midhileshmomidi2434]

Hi sir one doubt here finally some sentences gives incorrect output So how to use this to be learnt by the model in future to give correct output

[@NormalizedNerd]

For that you'll need online learning. A simple approach would be to train your model at a regular interval on the new sentences.

[@abrahammathew8698]

Good Video!! What if the words in the test sentence not present in corpus, will it work?

[@NormalizedNerd]

No, it will treat unknown words as zero vectors so no meaningful output.

[@greeshmaheshwari1620]

Amazing work bro. I want to know is there any method to not make incorrect guess again n again for same or similar sentences. Basically, any method to append this new input and target, without retraining the whole model every time.

[@NormalizedNerd]

Thanks man! Well appending input-output pairs is like using if-else instead of machine learning so it's a big no. However you can try to use online learning. In that case you can train the model periodically after you have gathered enough new data.

[@somunath1713]

Ml er sathe kichu emon playlist banao dada jate amra kivabe data science job a apply korte parbo kichu valo india based data science company te..

[@debanshumajumdar9883]

Your video should be more famous, it's simple and solves many issues in the heads of people struggling with Glove Model usage and LSTN. amazing! One question - For a multi class use case, do you recommend some specific changes?

[@NormalizedNerd]

Thanks man! For multi-class case, just change the activation function in the last layer (softmax) and the loss function (categorical cross entropy).

[@thetensordude]

Bhaiya please some seaborn tut...

[@NormalizedNerd]

Sure I'll try to make some.