The San Francisco company said Tuesday that it will demote tweets that "distort and detract from the public conversation" by taking into account behavioral signals about those users' behavior on Twitter.
The move is a global one, and is created to deal with behaviour that doesn't breach Twitter's usage policies, but likewise doesn't add anything either.
To better manage trolls on its platform, Twitter says it's beginning to analyze the unique behavior of individual accounts.
This includes people who have repeatedly tweeted at users who do not follow them, those who have often been blocked by other members of the site, or those who set up multiple accounts at once. According to the announcement, Twitter will also be "looking at how accounts are connected to those that violate our rules and how they interact with each other".
Within that, some of those were breaking the rules and we take action on those, but there's a lot that aren't and within that grouping there is a number of accounts that are engaged in behaviors that really distort and detract from the experience that people have in those areas.
Twitter says less than 1 percent of accounts on the platform make up the majority of accounts flagged by user complaints, and a majority of troll tweets don't violate its policies, despite having a large and negative impact on users. However, that didn't mean it wasn't affecting the conversation on the platform.
The company will hide certain replies from conversations and search results in a bid to "improve the health of the conversation" on Twitter. The firm has come under fire in recent years for how it handles reports of abuse by other users and has made a series of changes aimed at addressing that.
"Our work is far from done", said Harvey and Gasca.
According to Twitter's Safety account, these "signals" can be identified by an algorithm, and are tied to behavior, not the content of the tweets themselves.
"We've already seen this new approach have a positive impact, resulting in a 4 percent percent drop in abuse reports from search and 8 percent fewer abuse reports from conversations", it said.
"No. It's important to remember this is about behaviour, not content", she said.
However, the company acknowledged there was more work to do, and warned there would be "false positives" and missed incidents as the technology bedded in.
Twitter acknowledged it expects to make mistakes, and that the system will change, learn and improve over time.