Meetic Implements AI Content Moderation

Meetic has conducted a new case study review into how to protect users against online romance scams.

Meetic is a major European dating site and, like all dating sites, it faces challenges around user security. The company felt that it needed to provide a safe online environment in a cost efficient way.

One solution it found was to move towards implementing a larger proportion of automated content moderation. The challenge for this, it reports, was to ensure that it could be carried out without putting user safety at risk.

Meetic has been working with Besedo to try and learn more about how artificial intelligence can help to keep its high quality standards.

In order for the companies to investigate how much of the content could be automated at what quality, and before Meetic committed to a long-term engagement, Besedo’s proof of concept was outlined in the following steps:

  1. Training the engine – one year of content and moderation decisions were used to train the engine. It then compared the quality of manually moderated vs automatically moderated content. 94% of the automated decisions matched the human decisions.
  2. Refining data – Besedo refined the dataset by identifying the human decisions which had a high chance of being erroneous. Supervisors then reviewed the decisions, and 44% of the human decisions made were deemed invalid.
  3. Comparing automated and human decisions – Once the knowledge base was reliable, production data was automatically compared against human decisions. The automatic decisions which didn’t match the human decisions were then reviewed again by supervisors. The moderation efficiency was then calculated by both automatic and manual moderation.

The results discovered that 74.8% of content could be automatically moderated with a 99% accuracy level. The manual moderation then turned out to have a 97.9% accuracy level.

In conclusion, Meetic has been using Besedo for one year and have managed to streamline their moderation team and cut costs dramatically.

The automation rate has also increased over the year as the machine is constantly learning, and has now reached an increased 90% automation on moderation decisions – boosting its growth even further.

Read the full report here.