Skip to content

The Trust Quotient (TQ)

Samuel Hawksby-Robinson edited this page Apr 6, 2018 · 23 revisions

Samuel Hawksby-Robinson - https://github.com/Samyoul

Xenia Bogomolec - https://github.com/XeniaGabriela

Introduction

Quantifying the vague and subjective quality of trust is difficult and most attempts focus on a heavily reputation based solution [1], [2], [3]. Reputation may be an element of trust but not wholly trust itself. When approaching the notion of trust the TQ makes a few assumptions:

  • Trust's subjectivity limits measurement to how much the platform trusts an entity, rather than attempting to quantify how much all entities trust each other entity.
  • Trust is made up of multiple facets, some undoubtedly not yet considered in this paper. Considered are
  1. trust in saying the truth
  2. trust in reliability
  3. trust in being able to fulfill a mission
  • Each facet of trust has its own relative value compared to its sibling facets.
  • Some elements of trust will decay over time, giving more weight to trust elements created most recently.
  • By integration reputation as a trust component, we add views of other people and communities on the trustworthiness of a TiiQu member to the trust quotient.

The TQ is the degree to which a TiiQu member can be trusted in a professional context, and although ultimately the TQ is quantified in a hard number value what it really represents is the fuzzy human sense of "I'm pretty sure this person won't cheat me.".

Contents

Trust Context

TiiQu confines the TQ to trust in a professional context only. Private behaviour should not have an impact on the trust quotient. We will examine the potential trust sources accordingly. Like this we ensure privacy of TiiQu members on this level too.

Elements

The TQ is comprised of multiple elements, elements that are derived from asking questions about an entity and their claims.

We need to ask;

  • Identity - Are you who you say you are?
  • Verification - Is what you say about yourself true?
  • Reputation - What do other's think of you?
  • Performance Metrics - How well do you do what you say you do?

As TiiQu matures and the platform is used, further elements will undoubtedly be identified. The algorithm for determining the TQ has been designed with this in mind and is capable of handling very complex configurations involving many different elements.

Most elements, for example Identity, can have many differing sources and many different aspects, but after a certain point you start to experience normalized returns. An example, if we have already confirmed your ID by 7 different sources, your 8th source would likely not impact at all on your overall TQ. This would mirror real life, after someone is pretty sure you are who you say you are, additional information doesn't convince them much more. Identity recources will only have confirmed or not confirmed values. The weight of the values the trust source delivers depends on the kind of verification of the identity of a person.

  1. If the person had to appear personally at the trust source instance and show their passport or identity card, the weight will be be high.
  2. If the identity is verified by an email address and a phone number only, the weight will be low.

Likewise your 10 first project ratings will have a greater impact on your trust quotient averages than the 100th to 110th rating, however more ratings are better than fewer. When considering other aspects of reputation, reference ratings, on-platform ratings, peer reviewed or peer rated activity, an extremely high reputation can still be very powerful, impressive and useful, but up to a certain point it doesn't change how much you trust the person.

However elements of trust, such as reputation, decay over time without constant attention. Meaning that your latest reputation ratings will hold a lot more weight than reputation ratings from 10 years ago. You can't sit on your laurels.

Each element goes towards building up a picture of trustworthiness with no one person able to get a perfect trust quotient, and ever increasing work required to raise your quotient. The underlying elements can be presented as a cumulative rating. Various aspects of trust might play a differenet role in context of the kind of task that has to be fulfilled. But when it comes to trust from a human perspective we only want to know an answer to the question "How much can this person be trusted?". The answer would be something along the lines of "Not much", "A lot", "implicitly", but these are varying degrees of a vague spectrum.

TQ Structure

The TQ is calculated using a modular hierarchical weighted averaging system. As discussed an entity's TQ is comprised of many elements each element that contributes to a TQ is called a TQ Node or a Node. A network of Nodes is called a TQ Tree or simply a Tree.

The TQ Node

All nodes in a TQ Tree are structurally identical giving the overall TQ network several advantages:

  • A TQ rating can be calculated not only for individuals but also for groups of individuals by chaining trees.
  • A TQ tree can be appended to theoretically infinite levels.
  • TQ nodes are agnostic to their related nodes and so can be connected in multiple configurations

Node Fields

Node Behaviour

  • Nodes recursively seek all descendants when calculating their output value.
  • A node's output value is the product of the weighted average of its children and its value, if set.
  • Reputation based sources with an unlimited range use the median value as the upper range, meaning values will calculated as a deviation from the median.

TQ Node Representation

node_id

  • Data Type - Integer
  • Nullable - false
  • Description - The unique machine readable identifier for the node|

name

  • Data Type - String
  • Nullable - false
  • Description - The human readable identifier for the node

parent_id

  • Data Type - Integer
  • Nullable - true
  • Description - The unique machine readable identifier for the node's parent node

value

  • Data Type - Integer
  • Nullable - true
  • Description - The nullable raw value of the node

value_range_from

  • Data Type - Integer
  • Nullable - false
  • Description - The lowest value a node can have

value_range_to

  • Data Type - Integer
  • Nullable - false
  • Description - The highest value a node can have

weight

  • Data Type - Integer
  • Nullable - true defaults to 100
  • Description - The most impact a node's value can have on its parent relative to its siblings

samples

  • Data Type - Integer
  • Nullable - true
  • Description - The minimum number of children nodes a parent node must have before its weighting ratio is normalised

created_at

  • Data Type - DateTime
  • Nullable - false
  • Description - The block datetime the trust node was created at, useful for calculating decay over time.

source

  • Data Type - String
  • Nullable - true
  • Description - A string representing a location or reference for the node's value|

The TQ Tree

Not all TQ Trees will be structured the same, though most will likely have common elements:

  • A single exit node
  • A set of criteria node with the exit as parent
  • A set of source nodes with a criteria node or another source node as its parent

TQ Calculations

Benefits of a good TQ

Aside from the related kudos of having a bigger number next to your name a good TQ translates into additional benefits on the TiiQu platform. These benefits are as follows:

  • Higher voting power during community votes (if you have a TQ of 800 your vote counts twice as much as a rating of 400).
  • Greater share of the monthly Q PoW reward
  • Eligibility to purchase TiiQu equity
  • Greater visibility on the platform

Potential Weaknesses

Favour More Sources Over Fewer

Weakness:

Samples alone are not adequate to determine that more points are always better without seriously penalising TQs with a low number of child nodes. Fairly give a "more sources bonus" calculation for nodes handling reputation and feedback, because 100 5/5s is better than 10 5/5s.

Solution: A method for solving this issue would be to create a separate sub-node that is a sibling of the TQ node. Example simplified structure:

{
    "TQ": {
        "Reputation": {
            "TiiQu Rating": {
                "Points": {
                    "value_range_from": 1,
                    "value_range_to": 5,
                    "weight": 98,
                    "samples": 10,
                    "points": [
                        {
                            "value": 1,
                            "value_range_from": 1,
                            "value_range_to": 5
                        },
                        {
                            "value": 4,
                            "value_range_from": 1,
                            "value_range_to": 5
                        },
                        {
                            "value": 5,
                            "value_range_from": 1,
                            "value_range_to": 5
                        },
                        "etc"
                    ]
                },
                "Count": {
                    "value": 15,
                    "value_range_from": 1,
                    "value_range_to": 10000000,
                    "weight": 2
                }
            }
        }
    }
}

Any node that needs to consider quantity of sources as well the ratings themselves can split the two elements into two node branches belonging to a single node that handles the overarching source. A weight is applied to the rating and count nodes as per the preference for favouring the rating or the count.

Handling Extremely High Deviation From Median

Weakness:

When dealing with a reputation system that doesn't have an upper limit, you apply deviation from the median value. Couldn't a super high deviation from median rating break the system? E.g. stackoverflow of 3000000 against a median of 2600 gives a ratio of 1153.85 against a rating that should be under 100.

Solution:

Deviation from the median allows for ratios to be calculated that are higher than 100. A rating of 100 is in fact 100%, meaning you have no deviation from the median, this also does mean that there is the potential a deviation can be extremely high.

The potential of breaking the system is mitigated by how a TQ node handles child nodes passing a ratio in excess of 100. Nodes will increase the weight of the child node by the proportion that the ratio exceeds 100.

Example: If we take the weakness example of 1153.85. 1153.85 is 11.5385 times greater than 100, a Stack Overflow source node will likely have a weight of 5 (fairly low). Therefore we raise the weight from 5 to 57 (round(5 * 11.5385)), because a node's weight can only be 100. If the proportion that the ratio exceeds 100 by enough to raise the weight above 100, the parent node will default to 100. The latter step is the normalization of the parent node.

Remember an extremely high reputation can still be very powerful, impressive and useful, but up to a certain point it doesn't really change how much you trust the person.

TODO Give details for:

  • Replace diagrams with SVG vectors
  • Add TQ calculations from the PDF document
  • Calculations for element decay over time
You can’t perform that action at this time.