AdperSense

Josh Campbell
Chris Bell
2/25/2018
HackIllinois 2018

Purpose: Using machine learning, advertisments are anyalized based on data survayed from users.
Users can upload images to the database, and will be given the potential success rate of the advertisemnt, based on a set of data determined by user survays and machine learning.

This specific dataset included contains data of ~900 different advertisements reviewed by 18-24 year-old males. We strongly recommend as many data samples as possible, as it makes the model more accurate.

Data Collection: Advertisements to be tested by users are place in ./images_test/
To pull large quantities of advertisements from a google search, using Fatkun Batch Download Image is recommended
Lines 20 and 22 of UserFiltering.py must be changed to reflect the correct path

Users are prompted with an image, which is closed by the first keystroke The console then prompts for the user to decide if they like the ad, dislike the ad, would like to ignore the ad, or would like to quit the program.
Users can stop the program at any point and it will resume where the user left off upon restarting. Files that are discarded are placed in ./garbage/, files that are liked or disliked are placed in the appropraite folders, currently named ./good_images2/ and ./bad_images2/

After 3 sets of data are collected, ConsensusCalculation.py should be ran to determine which ads are liked or disliked by a majority of the users.
Several paths will need to be changed if you are using a different file structure
The ads that are liked by a majority of users are placed in ./final_good/ while ones that are disliked by a majority are placed in ./final_bad/

Built With

• javascript
• node.js
• python
• react