a flexible and distributed system for deriving, and interacting with, computational trust
TrustNet is a complete trust system that can be incorporated as a ready-made software component for e.g. distributed ledger technologies, or in a traditional client-server model, and which provides real value for impacted users by way of automating decision-making and actions as a result of assigned trust scores.
TrustNet has support for:
- managing different trust areas,
- discarding statements from distrusted nodes, and
- computing the subjectively most trusted nodes, as seen from a particular node.
TrustNet is the result of over 1 year of research into systems for transitive and computational trust. Read the 103-page report, or the shorter blog post.
See the examples/
folder for short snippets on how to use the system.
If you are unfamiliar to the terminology used in this repository, see the Concepts section, below.
TrustNet can be used either directly, or through the TrustHandler
class. The TrustHandler
has an extended API for managing trust across disparate trust areas (such as moderation
and music recommendations
, for example).
See the API section below for the available methods.
const { TrustNet } = require("../")
const distrusted = ["eve", "mallory"]
const trustAssignments = []
trustAssignments.push({ src: 'alice', dst: 'bob', weight: 0.25 })
trustAssignments.push({ src: 'alice', dst: 'carole', weight: 0.8 })
trustAssignments.push({ src: 'carole', dst: 'david', weight: 0.8 })
trustAssignments.push({ src: 'david', dst: 'carole', weight: 0.8 })
trustAssignments.push({ src: 'carole', dst: 'alice', weight: 0.8 })
trustAssignments.push({ src: 'bob', dst: 'eve', weight: 0.80 })
trustAssignments.push({ src: 'eve', dst: 'mallory', weight: 1.0 })
trustAssignments.push({ src: 'mallory', dst: 'eve', weight: 1.0 })
const trust = new TrustNet()
const root = process.argv.length > 2 ? process.argv[2] : "alice"
trust.load(root, trustAssignments, distrusted).then(() => {
let mostTrusted = trust.getMostTrusted()
console.log(`${root}'s most trusted:`, mostTrusted)
console.log("all trusted", trust.getAllTrusted())
})
TrustNet operates on trust assignments issued by nodes, which are typically peers in a peer-to-peer system like Cabal or Secure Scuttlebutt.
A trust assignment has a trust source, a trust target (or destination), and a trust weight. It is typically issued within a trust area. See the table below for descriptions of these terms.
Term | Description |
---|---|
Trust Source | The issuer of a trust assignment. |
Trust Target | The target of a trust assignment; the entity being trusted. |
Trust Weight | The amount of trust assigned to trust target. |
Trust Area | For example: book recommendations, moderation capabilities |
The trust assignments TrustNet operates on are of the following form:
{
src: 'alice',
dst: 'bob',
weight: 0.25
}
The trust weight weight
is a float defined on the range 0.0
- 1.0
, while src
and dst
are strings representing unique user identities.
TrustNet takes the trust assignments, and a trust root with id rootid
, and derives a trust graph. The trust graph is basically the subjective view of the trust assignments from the trust root.
Using this trust graph, and the trust root, we follow the trust assignments stemming out from the root and iteratively calculate trust ranks. The goal is to end up with a subset of the unique identities, represented in the trust assignments, which are regarded as the most trusted nodes, as seen from the passed-in trust root.**
If you want to know more about TrustNet, consider reading:
If you want to fund the development of TrustNet, or other independent research, sponsor me on GitHub or elsewhere.
Use either TrustNet
, for a single trust area, or TrustHandler
if you are managing multiple trust areas:
const { TrustNet, TrustHandler } = require("trustnet")
const tnet = new TrustNet() // or..
const handler = new TrustHandler(["moderation", "music recommendations"])
Default value of 0.50
. Sets the threshold determining which trust assignments are regarded as too low for a accurate trust computation (see meta-ranking trust strategy in paper or article).
NOTE As long as there exists at least one direct trust assignment, i.e. issued from the trust source rootid
, with a trust weight > opts.threshold, the computation will continue as expected.
Returns a promise. The promise is resolved when the trust computation has finished, and the final trust ranks (based on the passed-in trust assignments) have been determined.
rootid
is the id you are viewing the trust graph from, the trust source,assignments
is a list of trust assignments of form{ src, dst, weight }
. Wheresrc
anddst
are strings representing a unique id, and weight is a float in the range0.0
-1.0
.distrusted
is an optional list containing ids to discard from the final trust graph, before performing trust computation
Synchronous, requires having run load()
at least once.
Returns the ids of the most trusted nodes, as seen from the trust source passed into load()
.
Synchronous, requires having run load()
at least once.
Returns the ids of all nodes with a non-zero ranking, as seen from the trust source passed into load()
.
Synchronous, requires having run load()
at least once.
Returns a mapping of id -> trust rank
of the most trusted nodes, as seen from the trust source passed into load()
. Only returns non-zero mappings.
Also allows passing in areas as an object mapping trust area to its opts, see TrustNet
for which opts are applicable.
Example:
const { TrustHandler } = require("trustnet")
const handler = new TrustHandler({ "moderation": { threshold: 0.75 }, "recommendations": { threshold: 0.25 }})
Returns a promise. The promise is resolved when the trust computation has finished, and the final trust ranks (based on the passed-in trust assignments) have been determined.
Same as TrustNet.load
except it takes an area
, representing the trust area.
area
name of the trust area to perform trust computations for. If the trust area is being added for the first time, a call toTrustHandler.add(area)
will be issued.
Synchronous, requires having run load()
or loadAll()
at least once.
Same as TrustNet.getMostTrusted
except it takes an area
, representing the trust area.
Returns the ids of the most trusted nodes, as seen from the trust source passed into load()
.
Synchronous, requires having run load()
or loadAll()
at least once.
Same as TrustNet.getAllTrusted
except it takes an area
, representing the trust area.
Returns the ids of all nodes with a non-zero ranking, as seen from the trust source passed into load()
.
Synchronous, requires having run load()
or loadAll()
at least once.
Same as TrustNet.getRankings
except it takes an area
, representing the trust area.
Returns a mapping of id -> trust rank
of the most trusted nodes, as seen from the trust source passed into load()
. Only returns non-zero mappings.
Returns a promise. The promise is resolved when the trust computation has finished for all passed in trust areas.
areaMapping
An object mapping a trust area to the parametersload
requires
Example:
const rootid = areaMapping[area].rootid
const assignments = areaMapping[area].assignments
const distrusted = areaMapping[area].distrusted
Adds a new trust area to TrustHandler
area
the name of the trust areaopts
optional object containing options to pass toTrustNet
Gets the TrustNet instantiation for the passed in trust area.
area
the name of the trust area.area
should have been added previously duringTrustHandler
's instatination or via a call toadd()
,load()
, orloadAll()
.
Returns a list of all the currently registered trust areas.
Removes a trust area.
Note: Current implementation of TrustNet
is lacking a .destroy()
method. PR adding it welcome.
If you want to fund the development of TrustNet, or other independent research, sponsor me on GitHub or elsewhere.
TrustNet is available for dual-licensing. All the code in this repository is licensed as AGPL3.0-or-later
. If AGPL3 does not work for you, or your organization, contact cblgh-at-cblgh dotte org
to purchase a more permissive usage license.
If your project is a not-for-profit project, the permissive license will likely be available at very low-cost :)