mxaviersmp/market-leads-recommender

App developed for the Codenation Aceleradev Data Science 2020 final project.

About

This app was made using streamlit for the codenation Aceleradev Challenge 2020

It uses a dataset of businesses information to recommend similar businesses based on a portfolio of existing clients

How does it work

A Tf-Idf matrix was created using the dataset features

With this matrix, the cosine similarity is calculated between the given ids and the dataset.

Then, a score is attributed to each example on the dataset by the mean of the similarity scores

Running the app

Create a virtual environment and install the necessary libraries from requirements.txt

Download the data using misc/download.sh

Go to the src directoty

Train the model executing src/train_model.py

Create the geolocations executins src/geolocations.py

To start the app run streamlit run app.py

Exploration

In the notebook folder the is a jupyter notebook exploring how te model mas build.

Data

• estaticos_market.zip → Market data
• estaticos_portifolio{1, 2, 3}.csv → Test sets
• geo.zip → Geolocation of each address from estaticos_market.csv
• recommender.pkl → Trained Recommender model
• features_dictionary.pdf → Description of the features
• download.sh → Bash script to download the csv data
• links.txt → File containing the urls of the files to download
• README.md → Description of the challenge

Scripts

• app.py → Streamlit app file
• geolocations.py → Script created to extract the geolocation from the location on the estaticos_market.csv file
• preprocessor.py → Class responsible for preprocessing the estaticos_market.csv data to be used in the Recommender class
• recommender.py → Class implementing a recommendation system based on text simiarity using tf-idf and cosine distance
• SessionState.py → Class used for persisting user session data on streamlit
• train_model.py → Script user to train the recommender modelr model

Miscelaneous files

• download.sh → Downloads the data
• setup.sh → Configures streamlit for deployr deploy