Listing moderation software with machine learning
The problemEvery listing was read by a person before it could go live.
On a sales platform listings arrive faster than a person can read them, and among those that arrive some are misleading. dotenv wrote listing moderation software based on machine learning: it analyses the listing and decides whether to publish it, block it or pass it to an operator.
What moderation has to look at
Three things together, on every listing: the text, the images and videos, and compliance with the rules the platform has set itself. Doing it by hand means one person reading everything, and a platform that grows faster than that person can read.
The software receives the listing and assesses it with machine learning: the correctness of the text, the suitability of images and videos, violations of ethical rules.
The three roads after the assessment
If the listing passes, it is published directly on the platform.
If the software finds an anomaly, publication stops and the platform receives feedback with the analysis of what the system detected.
From there the platform decides, based on preset percentages: above a certain threshold the listing is deleted, below it goes to an operator who looks at it and decides. Today 20% of listings go to an operator; the software decides the rest on its own.
Why the threshold belongs to the client
It is the point where this work shows. How much risk to accept before stopping a listing is not something a software supplier knows: the one who knows is whoever answers for the content published on their own platform.