Tournesol main database

From Tournesol
Revision as of 15:44, 1 May 2021 by imported>Le science4all (Added global statistics and video statistics.)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

The Tournesol main database is the list of all inputs provided by contributors. It is thus an append-only database.

It consists of a list of tuples of the form < update_id, contributor, video_1, video_2, order, quality_criterion, confidence, new_value, context, duration_ms, timestamp >.

The platform also uses other tables to store raw data about contributor information, comments, content reporting, video metadata, user settings, slider changes and video statistics. Tournesol also manages derived databases about contributor updated ratings, contributor scores and Tournesol scores.

The global statistics are computed in RAM.

Note that we are currently updating our platform to implement the principles discussed in this page.


All entries to the database are given an update_id, which is labeled with the contributor's username contributor.

Each entry must then register two video identifiers video_1 and video_2 to be compared. We make sure that the database registers them in alphabetical order, i.e., video_1 < video_2. The value order is equal to true if the user was rating video_1 against video_2, and is false if they were rating video_2 against video_1.

The entry must then specify the quality criterion quality_criterion which was rated, as well as the corresponding confidence level confidence.

The entry must also specify the new_value of the rating. It may report a NaN if the update results from an "ignore" option of the disagreement page or of the inconsistencies page.

The value context corresponds to the page on which the rating was done (rating page, disagreement page, inconsistencies page or video page). If the update results from an "ignore" option, it should also say so: it should then equal disagreement_ignore or inconsistencies_ignore.

The value duration_ms of the time it took for the contributor to provide their rating, starting from the moment they opened the rating page.

The timestamp is the date of the rating update. If the update is a submission (submitted = true), then it corresponds to the time of submission).


By default, the Tournesol database is private.

However, contributors are encouraged to rate videos publicly. If the contributor's privacy setting for video_1 and video_2 are both public, then the entry will be made publicly available.

Each contributor can download all their personal data, public and private, on their user page.