Can you be more precise about how those algorithms recreate influences and echo chambers ?