Twitter Tests New Sensitive Content Appeals Process as it Continues to Refine its Detection Efforts

This is good news for Twitter users who believe that the company's visual identification algorithms have unfairly penalized them for suspected content infractions. Twitter is launching a new test of an improved appeal process for sensitive media violations today, to make it easier to challenge such decisions. The procedure will give users a more direct means to respond to content appeals from tweets, allowing them to get a response from Twitter if their content was incorrectly tagged.

On this front, Twitter's image recognition techniques have been the source of some high-profile misclassification problems. Last month, the site was chastised after an image of a critically burned Army veteran saluting in uniform on July 4th was reported as containing "possibly sensitive content". Realistically, Twitter needs to rely on AI as much as possible to limit the spread of hazardous content, and its systems, according to Twitter, are leading to improvements in this area.

According to Twitter's most recent Transparency Report, impressions on violative Tweets - those that break Twitter's rules for whatever reason - accounted for less than 0.1 percent of all total tweet impressions during the period, with the company's AI detection tools helping to significantly reduce exposure over time. Twitter also reported a 322 percent spike insensitive media policy violations, as well as a 194 percent increase in accounts suspended for non-consensual nudity.

These are positive results overall, but greater activity and dependence on automatic, preemptive limitation will lead to more false positives.

This improved appeals process makes it reasonable and should allow Twitter to continue optimizing its detection methods while also catering to users who have been wrongfully penalized as a result.

The new appeals mechanism will be tested with a small sample of users in Mexico, the United States, and Japan, the latter two being Twitter's most popular regions.

You can appeal a Tweet that has been labeled from the Tweet itself (web only) or label notice if you're in the experiment and haven't identified your media as sensitive in settings. It's a solid upgrade that fits in with Twitter's larger push to strengthen user security.