Recommendation system for e-commerce using collaborative filtering

This project features a recommendation system to recommend products using item-based collaborative filtering. Required number of recommendations are obtained for a given product. Collaborative filtering is carried out on the basis of item details (item name, category, brand etc.)

An introduction to some terms

Recommendation System

A recommendation system is a system that uses available data to provide suggestions for a specific item or object that a user is interested in. In the case of recommendation systems for e-commerce, they are used to recommend a list of products with respect to a given product.

Collaborative filtering

This project focuses on using item-based collaborative filtering to obtain recommendations. Item-based collaborative filtering uses the similarity between items to filter data.

Cosine Similarity

In this project, collaborative filtering is done using cosine similarity. A cosine similarity matrix is obtained having similarity scores between each item. This in turn is used to get the top recommendations for a specific product.

Objectives

1. To create a recommendation system to provide recommendations for a given product
2. To create a command line interface to give recommendations using command-line arguments

Data

The data for this project was downloaded from kaggle. Click here to view the dataset source.

Repository structure

.
├── data
├── notebooks
└── src

Instructions to execute the recommendation system

1. Create and activate python virtual environment
2. Install packages
3. Clone repository
4. Run notebooks in order
5. Execute CLI by running recco.py

Note : This project was created using python version 3.9.5

Create and activate virtual environment

python -m venv venv
source venv/bin/activate

Install packages

pip install -r requirements.txt

Clone repository

git clone https://github.com/amruthaa08/Recommendation-system.git

Notebooks

1. flipkart_data_cleaning.ipynb

- removing unwanted columns from data
- removing duplicates
- adding fill values
- saving data to clean_flipkart_data

2. flipkart_data_preprocessing.ipynb

- data preprocessing using nltk
- tokenizing data
- removing stopwords and punctuation
- stemming data
- saving data to preprocessed_flipkart_data

Executing the final step in the above mentioned notebooks will create and save two files 'clean_flipkart_data' and 'preprocessed_flipkart_data' to the notebook directory. You can choose to skip this step since these files are already available in data directory. Subsequent notebooks and scripts also use data already present in the data directory

3. cosine_similarity model

- extract features from preprocessed data
- apply cosine similarity
- create function to get recommendations

src

• recco.py

- creates command-line arguments to obtain file name, product name, and the number of recommendations required
- use command-line arguments to execute recommendation system through CLI 

How to use:

The CLI uses three arguments:

1. -f or --file_name obtains the file containing required data. The CLI uses the preprocessed_flipkart_data to provide recommendations. Hence ../data/preprocessed_flipkart_data must be passed as the -f argument when executing the CLI
2. -p or --prod_name obtains the name of the product for which we require recommendations
3. -n or --number obtains the number of recommendations required. This is an optional argument for which the default value is 5.

Code example:

python recco.py -f ..data/preprocessed_flipkart_data -p "Ladela Bellies" -n3

Executing this code returns 3 product recommendations for "Ladela Bellies".

Thank you for visiting my repository.