Content-Based Recommendation System

The business question

How do you suggest the right next product when a customer is new and has no purchase history yet? A content-based recommender answers it from what the products themselves are like. This project demonstrates the approach on a public movie catalogue (Kaggle The Movies Dataset). I am now applying the same thinking to real products at Brijwasi Sweets.

What I built

• An NLP pipeline (tokenisation and stemming with NLTK) over genres, cast, director, keywords and description
• CountVectorizer features and cosine similarity to rank the most similar items
• A Streamlit interface with live posters from the OMDb API
• A pre-computed similarity matrix, so recommendations load instantly

Where this applies in a business

Product catalogues, menus, course libraries and article archives: anywhere "customers who liked this may also like…" can lift engagement and order value, especially for new customers.

How success would be measured

Click-through on recommendations, add-to-cart from recommended items and average order value, compared with a group that does not see recommendations.