How and why I built Product Recommendation System
About me
I am a data scientist with 6 years of experience. working in MNC. I have experience in solving Computer vision, NLP, and machine problems.
The problem I wanted to solve
The banking client had 18-month transactional data for around 1M users. he wanted to build a recommendation system that can increase its conversion rate and business.
What is Product Recommendation System?
This product recommendation system is based on the user's demographical information and its transactional history with the bank. it can recommend products like credit cards, loans, taxes, etc.
Tech stack
I used pandas to clean the data. python to build the API. and sklearn to build the ranking model.
The process of building Product Recommendation System
at first, I cleaned the data and fill the null values. then I processed it using sklearn min max scaler. i did some feature engineering to include user's purchasing history to introduce new features then I trained model to predict recommendation score for each product using feature which includes user's info and its transaction history for products. I used MAP(mean average precision) metric to evaluate the model. at final, I got a MAP of 3.6% from the model.
Challenges I faced
it was very hard to train a model with a good MAP score. I had to do a lot of feature engineering to get a good score.
Key learnings
learning from this project was to enter in recommendation system and how we can imporve it using machine learning
Tips and advice
a good recommendation system can be build by using both user's information and its purchasing history.
Final thoughts and next steps
it's a good experience to enter in recommendation system. The next steps can be to improve the conversion rate by using production data and improve the MAP score.
