Yes, relay groups can be relevant. I image a "mainnet" group (the actual Twitter) of relays that sync together; clients treat them as unique subject optimizing band/cpu usage in the read/write operations. Then there will isolate (or small group for redundancy) relays that can be clustered manually by the user or with an algorithm by language, community, content typology, etc