Researchers found the platform’s algorithms promoted roles to certain users; company pledges to continue work in removing bias from recommendations