eigentrust 0.2.0

What is EigenTrust?

EigenTrust turns local trust ("alice trusts bob") into a global reputation score for every peer. Trust spreads from a few seed peers you already trust, so fake accounts that only vouch for each other get almost nothing. With no seeds it behaves like PageRank.

It is the algorithm from the EigenTrust paper (Kamvar, Schlosser, Garcia-Molina, 2003), used for reputation in peer-to-peer and decentralized networks, Sybil and spam resistance, and ranking accounts or contributors by who vouches for them.

Try it: eigentrust.jenyadoesapps.com, a live playground in 10 languages that can also rank 250,000 peers in the browser.

Rust library

[dependencies]
eigentrust = "0.2"

use eigentrust::{eigentrust, PreTrust, TrustEdge};

fn main() -> Result<(), eigentrust::EigenTrustError> {
    let local_trust = [
        TrustEdge::new(0, 1, 2.0), // peer 0 trusts peer 1 with weight 2
        TrustEdge::new(0, 2, 1.0),
        TrustEdge::new(1, 2, 1.0),
        TrustEdge::new(2, 0, 1.0),
    ];
    let pre_trust = [PreTrust::new(0, 1.0)]; // peer 0 is the seed

    let result = eigentrust(local_trust, pre_trust)?;
    for (peer, score) in result.ranking() {
        println!("{peer}: {score:.4}");
    }
    Ok(())
}

• eigentrust_with_options takes EigenTrustOptions for alpha (default 0.5), the convergence threshold and the iteration limit.
• Results come back as TrustScores: one score per peer, summing to 1, plus the iteration count and residual.
• Errors are a typed EigenTrustError.
• Optional features:
  • csv adds eigentrust::csv::Network for named peers from CSV.
  • parallel runs the iteration on all cores with rayon.
• There are no required dependencies beyond log, and the same code builds for wasm32.

The documentation on docs.rs covers the exact input rules and convergence behavior. See also examples/.

Command line

cargo install eigentrust-cli
eigentrust localtrust.csv pretrust.csv [alpha]

alice,0.6666666865348816
bob,0.3333333134651184

It prints peer,score for every peer with a non-zero score, highest first. Errors go to stderr with exit code 1.

Browser (WebAssembly)

import init, { run } from './pkg/eigentrust.js'

await init()
const enc = new TextEncoder()
const result = JSON.parse(run(enc.encode('alice,bob,2\nbob,carol,1\n'), enc.encode('alice\n'), 0.5))
// { Ok: [["alice", ...], ["bob", ...], ["carol", ...]] }  or  { Err: "..." }

• ./build.sh builds pkg/ and a multithreaded pkg-parallel/ from the wasm crate.
• The multithreaded build needs a cross-origin isolated page; see demo/vercel.json.
• demo/worker.js runs the engine in a Web Worker and picks the right build.

Input rules

• Weights: finite and non-negative, default 1. Zero means no trust. Negative trust (distrust) is rejected.
• Normalization: each truster's weights are scaled to sum to 1, and so is pre-trust. With no pre-trust, every peer starts equal.
• Duplicates: a repeated edge or seed keeps its last weight.
• Convergence: at α = 0 some networks oscillate instead of converging. The engine stops after 10,000 iterations with an error.
• CSV: standard CSV, so quoted fields ("Smith, J"), a header row, spaces, CRLF and a UTF-8 BOM are fine. Errors name the file and line.

Performance

Random trust graphs, 10 links per peer, CSV parsing included.

Results are reproducible bit for bit, with or without threads.

Repository layout

Development

cargo test --workspace --all-features   # tests, including doc tests
cargo run --example basic               # library example
./build.sh                              # WASM builds, copied into demo/
python3 -m http.server -d demo          # run the playground locally
git config core.hooksPath .githooks     # once per clone

License

Licensed under either of Apache License 2.0 or MIT, at your option.