Instaseek : find your influencers on Instagram

Manual

This project aims to detect genuine and organic influencers on Instagram. The model is data-based and uses a machine learning approach :

• Simple metrics such as followers, following, media count, mean engagement rate, post frequency, etc.
• Comment quality score: is the audience engaged ?
• Metrics over the feed images' quality : unity of contrast and colorfulness, redundancy of colors, etc.

Setup

Clone this repository on your local machine. Install Python 3.6

I will later make a pip package out of this project.

Install dependencies using pipenv and Pipfile

• If you haven't installed pipenv yet, pip install pipenv
• pipenv install
• If you encounter dependencies conflicts, pipenv install --skip-lock

Install dependencies using requirements.txt

• pip install -r requirements.txt

Start classification

• python src/__init__.py

Docs

You can check out docs here.

Script files

• annotation_tool.py helped me to annotate influencers streamed in the database.
• classifier.py lets you analyze an Instagram profile and classifies it among inlfuencer/not influencer.
• main.py is the entrypoint. Currently, it lauches streamer.py.
• sql_client.py is the SQL client. It processes and creates SQL requests to the database.
• streamer.py streams Instagram content into the database.
• train.py trains the model with data available in the database.
• user.py processes user infomation and extracts feature for machine learning.
• utils.py gathers all utility functions.

Contact

Valentin Berthelot

valentin.berthelot@imt-atlantique.fr valentin.berthelot5@gmail.com