media a/b voting app

This app allows A/B voting for media (images and videos) inside local directories. It can be used for fast media culling.

• The votes are written to the __voting_db.json file inside the selected folder.
• The app uses an elo-based ranking algorithm, that will output the rankings into the same file.
• It also has a "star" feature to allow boosting the elo of selected file.
• The app only supports Chrome based browsers.

on github

Link: https://adroste.github.io/media-a-b-voting/dist/

The app will work on local files only. Nothing is uploaded or processed remotely.

run locally

1. run npm i
2. run npm run build
3. run npm run preview
4. visit the shown url

local development

run npm run dev

rating algorithms

The app requires a ranking model that can be used to evaluate and rank entities based on pairwise interactions or match outcomes.

This source code has implementations for two ranking models:

1. Davidson Model (extended Bradley-Terry Model):

• A probabilistic model used for pairwise comparisons.
• Extends the Bradley-Terry model by incorporating additional parameters to account for ties or other complexities in the data.

2. Elo Model:

• A rating system originally designed for chess but widely used in other competitive settings.
• Updates player ratings based on the outcome of matches, with adjustments depending on the expected probability of winning.

Findings:

• The Bradley-Terry model (without ties) performs really well. However, the Davidson model is required to add 'tie' votes.
• The Davidson model does not perform as well with sparse data and requires more input data to converge.
• The elo model when trained with shuffled iterations, performs better on less data