Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/manoharvit/ecommerce-fashion-recommendation

Based on past interactions with customer and product metadata, including product description and appearance, recommend 12 products that are quite relevant to the customer.
https://github.com/manoharvit/ecommerce-fashion-recommendation

Last synced: about 2 months ago
JSON representation

Based on past interactions with customer and product metadata, including product description and appearance, recommend 12 products that are quite relevant to the customer.

Awesome Lists containing this project

README 

        
# Personalized-Fashion-Recommendation

Based on past interactions with customer and product metadata, including product description and appearance, recommend 12 products that are quite relevant to the customer.



## Introduction



Kaggle H&M Personalized Fashion Recommendations competition, where the challenge was to build a recommendation engine to predict which articles a customer would buy in a particular week starting on September 23, 2020.



### Key highlights of our work:

* Analyzed over 34GB of publicly available H&M datasets from Kaggle, which included images, articles, customer data, and time-series transaction data.

* Cleaned and merged unstructured and messy datasets to create a cohesive dataset.

* Integrated customer and transaction data to identify purchase patterns, and used classification techniques to categorize them into their respective article clusters.

* Developed a personalized fashion recommendation model that achieved an accuracy of 79%.



As Yuval Noah Harari points out in "Homo Deus," the same mathematical laws apply to both biochemical and electronic algorithms. This project aims to represent human decision-making processes as algorithms that process multidimensional input information and generate outputs in the form of decisions.



## Recommendations Everywhere



Recommendations have become integral to our daily lives, whether it's Spotify creating personalized playlists or Netflix suggesting TV series and movies. Similarly, H&M aims to save customers time by providing personalized clothing recommendations.



### Example Scenario: Emily's Decision-Making Process



1. **Need Identification**: Emily realizes she needs a new jacket for a cold evening out with friends.

2. **Search and Evaluation**: She browses H&M's online store and decides she prefers a coat over a jacket.

3. **Style and Preference**: Emily's style influences her decision, and she chooses a classic, cream coat within her budget.

4. **In-Store Experience**: She visits a local H&M store, tries on the coat, and decides to buy it along with matching pants and an umbrella.



## Data Sources



### Customers

The customer table contains information such as age, club membership status, and newsletter frequency. There are over 1.3 million customers to provide recommendations for.



### Transactions

Transactional data represents a customer's interests, needs, and style, with over 31 million transactions recorded.



### Article Hierarchy

Articles are described using multiple dimensions, such as product group, department, section name, garment group, perceived color name, and graphical appearance.



### Article Descriptions

Detailed descriptions of articles provide additional information about products, which can be transformed into embeddings to enhance ML model performance.



### Article Images

Images of articles were included in the dataset, offering another dimension for feature generation.



## Recommendation Engines



When thinking about recommendations, content-based and collaborative filtering models are the two most popular ML algorithms in this field.



### Content-Based Filtering

Content-based filtering uses item features to recommend other items similar to what the user likes, based on their previous actions or explicit feedback. For example, if a customer likes a leopard-print pattern, the model will recommend other products with the same pattern.



### Collaborative Filtering

Collaborative filtering uses similarities between users and items simultaneously to provide recommendations. This allows for serendipitous recommendations, where an item is recommended to user A based on the interests of a similar user B. The embeddings are learned automatically, without relying on hand-engineering of features.



## Tools and Frameworks



### Python Libraries



- **CUDF**: GPU DataFrame library for fast, parallel processing.

- **Pandas**: Data manipulation and analysis.

- **NumPy**: Numerical computing.

- **Scipy**: Scientific computing and technical computing.

- **Scikit-Learn**: Machine learning algorithms and tools.

- **TensorFlow / Keras**: Deep learning frameworks.

- **PyTorch**: Another popular deep learning framework.

- **Matplotlib / Seaborn**: Data visualization.

- **Plotly**: Interactive graphing library.

- **NLTK / SpaCy**: Natural language processing.

- **LightGBM / XGBoost**: Gradient boosting frameworks for efficient modeling.

- **TQDM**: Progress bar for loops.

- **Pillow (PIL)**: Image processing.

- **WordCloud**: Generate word clouds.

- **OpenCV**: Image processing.

- **WandB**: Experiment tracking and visualization.

- **Bokeh**: Interactive visualizations.



## Conceptualization



### ML Model: General Need for Clothing



1. **Expectations**: The model should estimate the probability of a customer needing a particular clothing type in a given time period.

2. **Solution**: Use propensity-to-buy models and time-series seasonality forecasting models.



### ML Model: Customer’s Style



1. **Expectations**: The model should sort articles within a product category based on how well they match a customer's style.

2. **Solution**: Generate features using image and text embeddings, then apply content-based and collaborative filtering models.



### ML Model: Available Stock / Lifetime of an Article



1. **Expectations**: Estimate the probability of an article being offered in a given week.

2. **Solution**: Use models to predict sales volume per article per week.



### ML Model: The Customer's Final Decision



1. **Concept**: Create a final-decision model that is an ensemble of all the individual models.

2. **Solution**: Use AutoML to generate final scores and validate the model.



## Final Concept



The solution involves generating features using multiple techniques, storing them in a feature store, and using an ensemble of ML models to generate recommendations. The expected input to the final ensemble model is a table with `customer_id`, `week_id`, and `article_id`, with columns representing results from various ML models.



## Summary



Our approach to the H&M Personalized Fashion Recommendations competition involved understanding the decision-making process, generating relevant features, and applying a combination of ML models to create a robust recommendation system. The project highlights the importance of data-driven insights in predicting customer behavior and enhancing the shopping experience.



## References



- [H&M Personalized Fashion Recommendations Kaggle Competition](https://www.kaggle.com/competitions/h-and-m-personalized-fashion-recommendations)