Do you ask about how to build Twitter if it didn’t exist, or how to build something that people would switch away to?
The №1 issue to overcome in the latter is to achieve Twitter’s network effect fast, so the first goal would be to design for easy bridging, connecting or aggregating.
For the former, people open the page to see content they yearn for. That depends on their interests. If you only show content from their follows, there is too little; if you show content from the whole platform, their interest will be drowned in things they don’t care about.
One easy trick is to show messages from people you follow with high probability, and messages from people they follow with lower probability, and so on. But unfortunately, you eventually need a stronger recommendation algorithm, which can show viral tweets (like the “dress color” illusion from 2015) and suggest people that tweet about your interests. This is the main reason people come back to the app every day.
The replies feature is actively radioactive. I’d recommend only showing replies from people you’ve approved, and putting other replies in a separate menu without a notification counter.
On decentralization, unless you add a hefty seasoning of cryptography and replication, it does not make sense. Beyond fragmentation, the risk of servers going down or in malicious access to private information or forgery is better addressed in a centralized fashion. Trust is easily lost.
Besides, the tech becomes a no-brainer with a central system. Go for a relational database and index as things take steam. Start synchronous, then move to async batch jobs.
The №1 issue to overcome in the latter is to achieve Twitter’s network effect fast, so the first goal would be to design for easy bridging, connecting or aggregating.
For the former, people open the page to see content they yearn for. That depends on their interests. If you only show content from their follows, there is too little; if you show content from the whole platform, their interest will be drowned in things they don’t care about.
One easy trick is to show messages from people you follow with high probability, and messages from people they follow with lower probability, and so on. But unfortunately, you eventually need a stronger recommendation algorithm, which can show viral tweets (like the “dress color” illusion from 2015) and suggest people that tweet about your interests. This is the main reason people come back to the app every day.
The replies feature is actively radioactive. I’d recommend only showing replies from people you’ve approved, and putting other replies in a separate menu without a notification counter.
On decentralization, unless you add a hefty seasoning of cryptography and replication, it does not make sense. Beyond fragmentation, the risk of servers going down or in malicious access to private information or forgery is better addressed in a centralized fashion. Trust is easily lost.
Besides, the tech becomes a no-brainer with a central system. Go for a relational database and index as things take steam. Start synchronous, then move to async batch jobs.