ECOMMERCE PRODUCT
RECOMMENDATION SYSTEM
Presented by Team 8: Ritik Tiwari, Kushal
Aggarwal, Naman Sharma, Pravesh Garg
INTRODUCTION
• An Ecommerce Product Recommendation System
improves the shopping experience by suggesting relevant
products to users.
• Goal: Enhance user engagement and drive sales through
tailored product suggestions.
• Dataset: Amazon Electronics Rating dataset, containing user
ratings for various products.
DATASET OVERVIEW
Amazon Electronics Rating Dataset Structure:
• user_id: Unique identifier for each user.
• prod_id: Unique identifier for each product.
• rating: Rating given by users (1-5 scale).
• timestamp: Time of the rating.
DATASET OVERVIEW
• Amazon Electronics Rating Dataset Structure:
• - user_id: Unique identifier for each user.
• - prod_id: Unique identifier for each product.
• - rating: Rating given by users (1-5 scale).
• - timestamp: Time of the rating.
DATASET CLEANING
• Remove Duplicates
• Handle Missing Data
• Fix Data Types:
• Handle Outliers
• Standardize and Clean Data
RECOMMENDATION APPROACHES
The system uses two approaches:
1. Rank-Based Product Recommendation: Recommends
popular products.
2. Similarity-Based Collaborative Filtering: Recommends
products based on similar users.
RANK BASED RECOMMENDATION
APPROACH
• To make the database after the EDA to make the rank
based recommendation We have prepared the new DB
where we have taken into the consideration of only those
users who have rated the product of more than 50
products to make the recommendation feasible, Here are
the steps:
• 1. Took the counts of the user_id column from the
processed data.
• 2. Created the rating matrix which contains the rows as the
user_id and prod_id.
RANK BASED RECOMMENDATION
APPROACH
Here are the few screenshots from the previous
discussion.
RANK-BASED PRODUCT
RECOMMENDATION APPROACH
• Objective: Recommend popular products to new users.
• Steps:
• 1. Calculate average ratings.
• 2. Count total ratings.
• 3. Filter based on popularity.
• 4. Recommend top products.
RANK-BASED PRODUCT
RECOMMENDATION APPROACH
Few screenshots of the previous slides where we have calculated the
average rating, total rating, recommend top 5 products based on the
popularity of products.
COLLABORATIVE FILTERING
• Objective: Provide personalized recommendations based on similar
user preferences.
• Steps:
• 1. Create user-item matrix: Build a matrix representing user interactions
with items.
• Calculate user similarity: Use cosine similarity to find users with similar
preferences.
• Identify similar users: For a target user, find the most similar users based
on similarity scores.
• Find new items: Identify items that similar users have interacted with, but
the target user hasn't.
• Generate recommendations: Recommend these new items to the target
user, typically sorted by popularity among similar users.
Cosine Similarity
User Interaction Matrix
EVALUATION
• Metrics used:
Mean Squared Error (MSE): Measures prediction accuracy
of model. Any model is good if we have as least as MSE.
User Feedback Analysis: Assess relevance based on user
feedback.
• Visualization:
For Visualization part we have look into the plots and
graphs get from the data analysis which gives the overall
perspective of the database and how the model will behave
easily can be predicted from the visualization.
CONCLUSION
• Combining rank-based and collaborative filtering ensures
effective recommendations.
• Future Enhancements:
Matrix factorization or deep learning methods.
Hybrid models that combine content-based filtering.
Incorporate temporal dynamics which analyzes time based
changes in user preferences. As in our case we have
consider the static case where we have the database which
consists of the users rating but this temporal dynamics
helps in making the prediction if there is a change in the
user dynamics w.r.t time.
