Building a Content-based recommendation engine API to recommend similar movies using NLP, Flask and Heroku. We generate keywords using NLTK, use it to build the model. Generate an API request using Flask and deploy it to Heroku.

Query String: /api?title=Insert-Movie-Title-Here

API : https://retro-reel-recommendation-api.herokuapp.com/api?title= Insert-Movie-Title-Here

Sample API : https://retro-reel-recommendation-api.herokuapp.com/api?title=Charlie%20Chaplin's%20The%20Rink%20!

Query String: /api?movieID= Insert-Movie-ID-Here

API : https://retro-reel-movieid.herokuapp.com/api?movieID= Insert-Movie-ID-Here

Sample API : https://retro-reel-movieid.herokuapp.com/api?movieID=b70eec12b79148dfbc97ac3b69ca49bf

Complete list of Movies can be found here

I love movies and was interested in working on a recommendation engine. So, I reached out to a Popular Android developer to create a recommendation engine for their retro reel app which serves retro movies from the 1900's onwards. Their requirement was to recommend similar movies as an API by passing only the movieID.

• Python
• Numpy
• Pandas
• NLTK
• Rake
• Scikit learn
• Pickle
• Flask
• Heroku

After the initial EDA, I extracted keywords using Natural Language Processing from the Dataset. After feature engineering, I decided to build a recommender enginer using TF-ID vectorizer.

I had previously used the following approaches to build the recommendation system using a countvectorizer:

1. Keywords as the feature.
2. Movie Titles as the feature.
3. Generating my own keywords using NLP as the feature.
4. Combining 2 & 3 to get the best of both.

I found that TF-ID vectorizer gave out a better quality of recommendations and chose that model.

1. Data Acquisition Parse the deeply nested JSON file

2. Data Preprocessing Normalize the JSON file, make the data usable.

3. Feature Engineering Use the mentioned approaches for feature selection mentioned above.

4. Model building For each of the approaches, we determine the cosine similarity index and post training, the model is pickled.

5. Model Evaluation Compare the results with each of the models and determine the best model to productionize.

• Retro_Reel_Recommender : To demonstrate the entire data science lifecycle.
• recommender : Has functions to return the recommendations for the API calls.
• recommendation_api : Flask app to run the recommender engine as an API.

After opening Anaconda environment, to run Flask locally:

set FLASK_APP=hello.py
flask run

Please leave me a feedback of how my engine performed in the Issues tab or email me. Feel free to try out the API calls with different movieID's from the list below.

1. How to build a content-based movie recommender system with Natural Language Processing
2. Build a Movie Recommendation API using Scikit-Learn, Flask and Heroku
3. Simple Content-based Recommendation Engine API With Flask Heroku Deployed