In this project, we aim to develop a personalized fashion recommendation system for H&M, inspired by this Kaggle competition. The dataset comprises three key components:
- articles.csv: Contains information about the products in the catalog.
- customers.csv: Provides details about customers (note: customer information won't be used in the models).
- transactions.csv: Lists the purchases made by each customer for each date.
The project workflow includes the following steps:
-
Data Exploration: We will thoroughly explore the three datasets to understand the underlying patterns and characteristics.
-
Recommendation Systems:
- Collaborative Filtering: Utilizing the user-item interaction matrix to provide personalized recommendations.
- Hybrid Model: Combining the user-item interaction matrix with product information to enhance recommendation accuracy.
-
Focus on 'menswear' Category: Due to computational constraints, we will concentrate our efforts on the 'menswear' category, as it aligns with our available computing resources.
-
Library Usage: LightFM, a Python library for recommendation systems, will be employed to build and implement our models.
-
Conclusions and Suggestions: The project will conclude with a detailed comparison of the collaborative filtering and hybrid models. Additionally, we will provide suggestions for potential enhancements to the recommendation system.
The project structure includes Jupyter notebooks in the notebooks/ directory, housing code for data exploration, model implementation, and evaluation. The data/ directory contains the dataset files, and the main documentation is provided in the README.md file.
Ensure the following dependencies are installed:
- Python 3
- Jupyter Notebook
- NumPy
- Pandas
- Scikit-learn
- LightFM
- Matplotlib
- Seaborn