The Tournesol dataset: Which videos should be more largely recommended?

Lê-Nguyên Hoang, Romain Beylerian, Julien Fageot, Louis Faucon, Aidan Jungo, Adrien Matissart, Nathaël Noguès

Keywords: Recommendation, Ethics, Preferences, Human

TL;DR: We publish a dataset of human judgments on which videos should be more largely recommended by algorithms.

Abstract:

This paper introduces the Tournesol public dataset, which was collected as part of
the online deployed platform https://tournesol.app. Our dataset contains a
list of 1,116,318 comparative judgments of YouTube videos by 8,804 users of the
Tournesol platform. 263,668 of these judgments were about which video should
be more largely recommended, while the remaining evaluate secondary criteria
like content reliability, topic importance and layman-friendliness. The dataset also
exports information about users’ pretrust statuses and vouches. It is published
at https://api.tournesol.app/exports/all under ODC-By license. The
data is currently used by Tournesol to make community-driven video content
recommendations to over 6,000 users.

Dataset URL: https://api.tournesol.app/exports/all

Primary Area: AL/ML Datasets & Benchmarks for social sciences (e.g. climate, health, life sciences, physics, social sciences)

Submission Number: 1405