How to cluster?

[tttp]

This is another tricky one:

for each message, we need at least to be able to assign it to a specific campaign

but it's probably/possibly useful to classify it further, for instance the language of the message.

should we add some sentiment analysis? or to help understand nuances?

for instance let's imagine there is a big campaign running about climate change.
We might imagine that the supporters might decide to highlight one of the specific issues:

1. climate resilience: fund infrastructure work to prevent flooding/improve hospitals...
2. climate justice: help nations that are going to be particularly affected by climate change
3. tax the polluters and oil companies

This is likely for post MVP, but having some sentiment analysis might be useful? politicians might want to have different replies?

unless great idea, let's assume one category/campaign.

anyway, how for each message identify how close/different from a campaign? cosine is simple, is this good enough?

should we regularly adjust the "campaign reference vector" to be the average of all the messages that are clustered into the campaign?

should we let the politicians (their assistant) to classify emails where we aren't sure? (either to add them to an existing campaign or to identify a new campaign?

might be post MVP too. not sure