Twitter’s new method for combating trolls on its service involves showing people less of their tweets during conversations.
The social messaging company said Tuesday it is now analyzing certain behaviors of its users to determine how their tweets should be displayed when they tweet replies to other users during online discussions. Additionally, people’s behaviors will be taken into account to determine whether their tweet should be featured more prominently when other users search Twitter for a particular topic.
The goal is to proactively remove potentially abusive or unbecoming tweets before people react to them and report them to the company.
Twitter (twtr) has been attempting for years to clean up the amount of abuse on its platform. Its latest initiatives involve working with outside researchers to determine what constitutes healthy conversations on its service. The hope is that by understanding what makes a Twitter conversation healthy, the company can promote methods to foster more worthwhile tweets rather than their bellicose counterparts.
Some of the behaviors Twitter is now analyzing for its new initiative include whether a person has confirmed their email address, has signed up for multiple accounts simultaneously, and if those accounts “repeatedly Tweet and mention accounts that don’t follow them,” the company said in a blog post.
“We’re also looking at how accounts are connected to those that violate our rules and how they interact with each other,” Twitter executives wrote in the post. “These signals will now be considered in how we organize and present content in communal areas like conversation and search.”
Twitter said that based on early testing, the company has seen a “positive impact” that’s resulted in “a 4% drop in abuse reports from search and 8% fewer abuse reports from conversations.”
The company said that this is just one of other steps it is taking to weed out abuse.
Get Data Sheet, Fortune’s technology newsletter.
“There will be false positives and things that we miss; our goal is to learn fast and make our processes and tools smarter.”