Project: Personalized Product Recommendations using Machine Learning (Flipkart Grid)
Team Name: Velle Team Member: Garvit Mukhija
Objective: The project aims to create a personalized product recommendation system for an e-commerce platform like Flipkart. This system will utilize machine learning techniques to enhance user experience and increase sales by suggesting products based on user behavior.
Dataset: The team has used an Amazon dataset that contains user ratings for electronic products. Each product and user is assigned a unique identifier to avoid biases.
Approach:
- Rank Based Product Recommendation:
Objective: Recommend popular products based on the highest number of ratings.
Outputs: Recommend top 5 products with a minimum of 50 or 100 ratings.
Approach:
Calculating average and total ratings per product, creating a sorted DataFrame, and designing a function for retrieving top 'n' products with a certain minimum interactions count.
- Similarity-based Collaborative Filtering:
Objective: Provide personalized recommendations based on similar users' interactions.
Outputs: Recommend top 5 products based on interactions of similar users.
Approach:
Converting user IDs to numbers, creating functions to find similar users using cosine similarity, and making product recommendations based on similar users' interactions to solve the "cold start" issue and provide personalized suggestions.
- Model-based Collaborative Filtering:
Objective: Provide personalized recommendations based on past user behavior and preferences.
Outputs: Recommend top 5 products for a specific user.
Approach: Converting ratings into a compressed sparse matrix, performing dimension reduction with SVD, predicting ratings using U-sigma-Vt matrices, storing predictions in a user-product DataFrame, recommending products based on predictions, evaluating model accuracy through RMSE calculations.
For detailed implementation and code, please refer to the provided links to the dataset and the corresponding resources.