I don't often left comments on youtube but, finally someone that explains everything from scratch...I am a JS developer. And it's really cool your that you explain every piece of code. That really helped, I was able to understand everything.
Your videos like gem to me learned a lot your use of modules packages are like cherry on cake. Currently I'm working as an Jr. Data scientist in KPMG but man oh man you taught me many things thank you 😊 🙏
Bro, I just need to talk to u. I wanted to ask few questions regarding the profile you are working on. I have secured a job with Deloitte but want to switch to KPMG (Gurgaon).
Hey brother , you just provided the best NLP sentiment project , your channel deserve million+ subscriber , nd now I am just one new subscriber now to reach you there
Really interesting video. I've been following a lot of your tutorials lately and I must say that I really like the way you explain things, it's so easy to understand and follow along. Thank you!
Thanks so much for the feedback Juan. It's always hard to tell when I'm recording these if they are any good, so it's great to hear that it is helpful to you.
This was a good tutorial. I'm trying to get my feet wet in data analytics and found myself overwhelmed while trying to read the NLTK documentation, so thanks for the structured guidance. I'm working on analyzing sentiment across a dataset I've gathered myself, so I wasn't following along in kaggle and hit a hiccup as AutoModelForSequenceClassification requires pytorch and I initialized a python 3.10 environment. Oopsy poopsy. All the same, you made my headache significantly less daunting. Thank you. :)
Thanks so much. I’m glad it helped you get started with NLTK it can be a lot easier when you see it in action once. Setting up an environment that works with all the packages can also sometimes be frustrating so I can relate!
Thanks for such a wonderful tutorial. I used your shared data on my own with Google Collab and worked so well. Just I had to download a few more libraries for tokenization. Wonderful content and I truly enjoyed it.
just did all of that as a thesis by myself without knowing you made a video about it lol, luckily I've used a different Bert model from hug face at least. Nice video btw!
Thank you so much for this step by step process it has opened up all sorts of new analysis opportunities for our customer insights. Really well explained and easy to follow
Huge thank you to you!!! I recently participated in a ML hackathon and they had sentiment analysis as one of their problem statements. I had watched your video prior to the competition and used hugging face whereas everyone else used the standard vader. I ended up getting the highest accuracy and placed first, all in my second year of engineering. Genuinely, can’t thank you enough for the information! Team random_state42
I've just recently found myself interested in Computer Vision and NLP and I've finally gotten to the right content creators, this video absolutely rocks! And I fouind it 2 years late, I wonder how far are you now in this topic, if ever you come back to this comment section could I ask how did you get so experienced in this topic and how did you learn how to tackle all this problems? Thank you!
Thanks for the video, we have a school project to do anything coding related and while my classmates are using scratch I wanted to do something flashier, and some kind of language analysis seemed the way to go. I'll use this video as inspiration.
Who are you? My saver! I was asked to conduct a sentiment analysis on reviews from my internship. I was doing computer vision at the graduate school. New to NLP. Thanks God.
Pls make more such videos, that was great. I am a data engineer and wants to move to Data Science, please make videos for guidance also. Love from India
Thank you very much for this video. I'm new to the field of Data Analysis and related disciplines so this sentimental analysis project is pretty insightful for me.
00:01 In today's video, we'll explore sentiment analysis on Amazon reviews using traditional and more complex models. 02:26 Importing and reading data for sentiment analysis 07:28 Tokenization and part of speech tagging in NLTK 09:55 Introduction to VADER for sentiment analysis 15:12 Looping through Amazon review data to calculate polarity scores. 17:33 Perform sentiment analysis with NLTK and 🤗 Transformers 22:04 Explains the positive, neutral, and negative sentiments in Amazon reviews 24:24 Transformer-based deep learning models from Hugging Face are easy to use and powerful 28:45 Introduction to sentiment analysis with NLTK and Transformers 31:02 Running sentiment analysis on text using Vader and Roberta 35:28 Comparing vader and roberta sentiment analysis scores using seaborn's pair plot. 37:45 Vader model less confident compared to Roberta model 41:59 Hugging Face Transformers makes sentiment analysis simple and efficient 44:09 Explored models and ran sentiment analysis on Amazon reviews. Crafted by Merlin AI.
dude in 26:03 while writing the pertained model from hugging face it throwing an error. "Connection error, and we cannot find the requested files in the cached path. Please try again or make sure your Internet connection is on. " and my connection is very good I had run this around 40 times with good connection and still throwing that error and also changed the model from hugging face please help me on this
You might want to check and make sure the source hasn't changed from the hugging face site. They might have changed this specific model and your refrence might need to be updated.
Had the same problem. Just solved it. Unlike aveage laptop, Kaggle notebook is not connected to internet. To get an internet access with your Kaggle notebook you need to go through a phone verification. Look for the notebook option menu on the right side.
One of the best tutorials on Vader and the Huggingface Transformers I have seen. One question I had: How is the confidence score calculated on the Pipeline model and is there a way to evaluate the model's performance on these calculations?
Thanks so much for the feedback. Glad you found it helpful. Evaluating the model performance is a bit tricky without ground truth labels. The output of the Pipeline model is essentially the probability the model predicts of each class given the dataset it was trained on. Check out the actual model description on the huggingface site here along with the noted limitations: huggingface.co/distilbert-base-uncased-finetuned-sst-2-english Specifically this part is interesting: ``` Based on a few experimentations, we observed that this model could produce biased predictions that target underrepresented populations. For instance, for sentences like This film was filmed in COUNTRY, this binary classification model will give radically different probabilities for the positive label depending on the country (0.89 if the country is France, but 0.08 if the country is Afghanistan) when nothing in the input indicates such a strong semantic shift. In this colab, Aurélien Géron made an interesting map plotting these probabilities for each country. ```
@@robmulla FWIW - I reached out to the creator of this and what I was told is that the score is calculated using the activation function after the final layer of the neural net. It is used to determine polarity (and is not a confidence score). The model returns an array with the score for each polarity, and the larger is the prediction. The values will always be positive, regardless of the actual sentiment class tagged to the text. This is unlike Vader's model which provides a composite polarity score that could be a positive or negative float based on the inferred sentiment (positive, negative, neutral).
Awesome video. Would be great to see you follow the sentiment analysis with a topic analysis. I’ve seen a few different options out there (LDA, Top2Vec and BERTopic), but would love to see your take on it.
Great content. Please do more content model which solves attrition prediction for org. Very complex subject because its hard to find already made models on such topics. It would be great help if you can make something attrition prediction model with variables more than 45-50.
Thanks for this video, it was descriptive, well structured and well explained. I have two questions and I would appreciate if you can give your opinion and guidence on that. 1. At the end of the day star reviews and sentiment are giving the same results so how can we justify going through all this process when we already have a very good indication of user sentiment based on the star reviews. 2. How can we get the strength and weakness of the product based on the reviews using the sentiment analysis.
Excellent explanation and material. Thank you for your efforts in making learning enjoyable. A brief query about reviews that are negative (5 stars) and positive (1 stars), where the algorithm is unable to forecast the relevancy score. Regarding these kinds of situations, how would you advise handling them??
Thank you so much for this video tutorial! I wanted to ask if you created the Amazon review dataset from scratch or was it already pre-made from somewhere else?
@robmulla, great presentation but I have looked through videos on your channel, it appears you have not done one on finetunning a BERT model with custom dataset. I am particularly wanting to learn how you would finetune a BERT model for multiclass text classification, maybe on Google collab. I think many of us subscribers would love it. Thanks.
Both the VADER and ROBERTA model struggled with sentences with more context. For instance, both rated the sentence "I have had better in the past. It works well enough, but temper your expectations." as overwhelmingly positive. Are there ways to capture that context?
23:42 Step 3. Roberta Pretrained Model. RoBERTa base sentiment I am getting a value error. That is we cannot find the requested files in the cached path. Please try again or make sure your Internet connection is on. But My internet connection is good. What can I do about it?
yeah im getting the same problem ValueError: Connection error, and we cannot find the requested files in the cached path. Please try again or make sure your Internet connection is on.
Thanks for the feedback Niklas. You are correct that VADER handles the parsing of text and assignment of sentiment per word so we don't have to tokenize like with the transformer model. Check out the source code for VADER and it might make a little more sense - it handles specific cases like if words should "boost" the intensity of the sentiment and/or specific idioms: www.nltk.org/_modules/nltk/sentiment/vader.html
Amazing video! One question though. Initially we tokenized the data, found their part of speech and then grouped them into entities. However the vader and roberta model were ran on the raw example. does it mean that data cleaning/manipulation like dropping stop words etc isnt required for the models or did i understand it incorrectly?
Hey Rob , I was trying to execute the code where you extract the model trained on twitter comments but I keep getting the error "Connection error, and we cannot find the requested files in the cached path. Please try again or make sure your Internet connection is on." even though I am connected to the Internet . Could you please help me out ?
Hi, thank you for the amazing video. Your presentation was informative and insightful. Looking forward to your future content! Btw, I want to ask how can I save my expected result, it seems like I had a good training and dont want to keep going. What should I do in this situation ? Thank you
I'm having an issue running the tokenizer. Some goggling suggests downloading 'punkt' which i have done but still not working...Anyone experience this? I restarting my notebook from scratch after installing punkt and even ran code to check its installed which it says it is.
Sir I know pandas and data cleaning.But there is a alot of models in data science. which model I learn. My nitche in data science is Sales and marketing.Give me Some tips thanks.
Thanks for the feedback. I'm still learning every day just like you. The great part about data science is that there is always something new to master.
Very well explained video and clear guidance! I have a question about the preprocessing part of the text before putting it into the tqdm sia loop, do we directly put the raw data into it, or do we do the tokenize, remove stop words and stuff first, and then go for the sentiment analysis? Looking forward to your reply!
Hey Huan! Glad you found the video helpful. I'm not sure about the loop you are referring to but typically the text needs to be tokenized, but depending on the model it may handle that within the predict function. Hope that helps.
Great content, Really loved the explanation. I'm new to sentiment analysis but was wondering this : My objective is to score a set of reviews online of products, so shouldn't i first do a set of text pre-processing like normalization, spell check , lemmatization, tokenization before feeding each sentence into the pre-trained transformers model ? . How much of a difference would this cause in accuracy of predictions ?
Stoaked you enjoyed it Navaneeth! This video only scratches the surface. The tokenization and preprocessing of the text is usually built into the model pipeline and would depend on the model you are using. I'm not sure abuot how it would impact the accuracy but for instance VADER I believe stop words are removed. Worth looking into for sure!
Great Video! Just wanted to know is there a way to find the performance of our model in terms of confusion matrix, accuracy and precision score? Thank you
Great question. You can read about the model evaluation on the huggingface model card. It’s hard to properly evaluate independently because it requires a labeled dataset.