PageRank for Email: How Trust-Graph Sender Reputation Works

Short answer: Trust-graph email reputation treats your mailbox as a graph of relationships and scores each sender by how close they sit to you in that graph. It uses Personalized PageRank, the same family of math behind Google's original search ranking, seeded on you instead of on the whole web. People you actually correspond with score high. Strangers with no path to you score near zero.

What is trust-graph email reputation?

Trust-graph email reputation is a per-person score, from 0 to 100, for every sender who emails you. It is computed from your own mailbox metadata: who you reply to, how quickly, how often, which threads keep going, who you meet in group threads, and which messages you star or file. None of it reads the content of your mail.

The key word is personal. The score is not a public number shared across everyone. It is yours, computed only from your graph, and it can be completely different from the score the same sender has in someone else's mailbox.

Why PageRank, and why personalized?

Classic PageRank ranks a web page by the links pointing to it, weighted by the importance of the pages doing the linking. Personalized PageRank changes one thing: instead of starting the random walk anywhere, it always restarts at a chosen node, the seed. Everything is measured relative to that seed.

For email, the seed is you. The math looks like this:

π = α·e_u + (1 − α)·Pᵀπ, with α = 0.15

Here π is the trust mass spread across the graph, P is the relationship-weight matrix, and e_u is the restart vector that pins every walk back to you. Because all the probability mass originates at you, π measures how much of your trust reaches each sender.

How the edges get their weight

Trust flows along edges, and the edges are built from metadata signals:

• You replied, and replied fast. Replying is the strongest signal that you value a sender.
• Sustained back-and-forth. A thread that keeps going beats a single exchange.
• Recency. Trust decays over time. A contact from five years ago ranks below one you email this year. Old interactions halve in weight about every twelve months.
• Stars and labels. Explicitly keeping a sender is a clear "I care" signal.
• Co-recipients. People who appear with you and your trusted contacts on real threads gain a little trust, which is how warm intros emerge.

Bad signals push the other way. Mark a sender spam and their score drops to zero, for you only.

Why this beats a single global number

A public reputation score gives every sender one number for everyone. That is easy to game, because fake accounts can farm it and a coordinated pile-on can bury a rival. It is also wrong, because a sender who is noise to the world might be exactly who you want to hear from. A personalized score, seeded on you, avoids both problems. A fake-account ring with no real path to you stays near zero no matter how densely it links to itself.

FAQ

Is this a spam filter? No. A spam filter blocks known-bad senders. Trust-graph reputation ranks the mail that gets through, by who it is from. See spam filter vs trust filter.

Does it read my email? No. It reads headers only, never the subject, body, or attachments.

Where can I see it run? Kith is an open-source implementation of trust-graph email reputation that sorts your Gmail in place. Read how to sort your Gmail by who you actually know.