Project involves Web Scraping custom IMDB data between 2020 and 2021 of 10000 movies and shows sorted by number of votes ,fine tuning a pre trained DistilBERT Transformer using Transfer Learning and then saving and reusing the saved model for further use.

• DistilBERT Transformer
• Tensorflow
• Numpy and Pandas
• Selenium, BeautifulSoup4 and requests

• Accuracy achieved: 81.3492%
• ROC_AUC_Score achieved: 0.7217

1) Ensure Python and Jupyter Notebook are installed. Optionally Conda environment can also be used.

2. Install the required modules using

pip install -r requirements.txt 

or conda install -r requirements.txt

or !pip install -r requirements.txt for Google Colab.
     

3. Selenium requires browser specific drivers. Guides for Chrome and Firefox are mentioned below. Alternatively,this step is optional if the notebook is run on Google Colab.
   Chrome: https://chromedriver.chromium.org/getting-started
   Firefox: https://www.lambdatest.com/blog/selenium-firefox-driver-tutorial/

1)(Optional) Run the IMDB Web scraper . This generates the already provided csv file and imdb_movies pickle file.

2. Run the IMDB Web scraper on an environment which has GPU acceleration. Here it is used with Google Colab where Nvidia Tesla T4 or Nvidia Tesla K80 are allocated.
   

   Training Time: Roughly 20-25 mins
   Epochs: 10
   Training Batch Size: 8
   Max length of each Sentence: 512 
   

   A Movie_prediction_model directory is created with config.json file(provided) and a tf_model.h5 (not provided due to space constraints).

1) Ensure the model has been created inside Movie_prediction_model directory.

2. Run the python file using python DistilBERT_Movie_Classifier.py

3. Enter the description of the movie or TV show you want to predict for. An output will be generated with the binary prediction of success based of IMDB Ratings.