jimaras08 / mlops-lab-example-yelp Goto Github PK

• Distinct words and its frequency for bad and good reviews. Can also do word Cloud for each class
• Run a sentiment analysis model from BERT and check correlation with review score
• ...

Two classification tasks are constructed from this dataset – one predicting full number of stars the user has given, and the other predicting a polarity label by considering stars 1 and 2 negative, and 3 and 4 positive.

• target distribution
• Target distribution per business category
• Target distribution per city

For more ideas: https://github.com/Yelp/dataset-examples"

Prediction using pretrained transformers and fine tuning them

Create a dashboard for predicting the score of a review, given the text

Can also do a "game" where you try to guess the review score and it gives you the real score

Setup MLFlow Experiment Tracking Code.

How do we setup MLFlow?

How do we train a model and save its parameters? Can do this with a mock model from Pytorch if needed

Include a README explaining how to install everything

Create a .csv file containing the indices to use as splits for train, test and validation. This will be used to compare the several models.

Consider each row as a review with a score.

Prediction using character level CNN https://medium.com/datadriveninvestor/using-fastai-to-analyze-yelp-reviews-and-predict-user-ratings-polarity-4e4e89df358e

Using the folds defined in the other issue, train a model using:

• Bag of Words
• Average word Embedding of Review (use spacy english language model)

Can use one or both features.

Then just run a Logistic Regression over this to get a baseline performance for both tasks.

Prediction using pre-trained language models with sentiment classification (ULMFit)

Check FastAPI for the code