Reputation Evaluation Model
The simple computational model of reputation in the formula described above can be complex to account for multiple ratings and can be constrained by respective parameters, as discussed further. In the following discussion, we will discuss the reputation of a node, and each node can be thought of as a user, a user account, a software agent, or any role played by a multi-agent social network such as a blockchain system. We will also refer to artifacts generated by nodes during their activities, so they can be indirectly rated by rating their artifacts, such as posts and comments in social networks, or messages or tasks in a multi-agent system. The following five ratings for reputation calculations are considered on a time-constrained peer-to-peer basis, so node i ranks node j by explicit or implicit, direct or indirect financial rating values at time t.
Fij — identifies an explicit direct rating of the fact corresponding to node i trusting node j. In traditional social networks, including Steemit, it corresponds to a “follower” link between two users. There may be only one such rating at time t, and there may be different moments when creating or removing this link. Typically, it is 0 or 1, but we can imagine a system with a “five stars” rating or any range of values, including possible negative ratings.
Oijt — Corresponds to an explicit indirect rating, indicating that node i explicitly voted for the artifact created by node j. In a social network like Facebook, this is often referred to as “like”. In Steemit it is called “voting”, it can also be 0 or 1, but we can also imagine any range of values, such as Youtube. There are negative and positive content ratings such as -1, 0 and 1.
Cijt — Represents an implicit indirect rating, corresponding to the fact that node i creates textual content, such as post comments about artifacts authored by node j, extracting positive or negative emotions from the text, so this can be any positive or negative value . Any size. In an oversimplified form, we can assume that there may be 0 or 1, which means that all existing comments are considered positive, if any.
Dijt — represents an implicit direct rating, corresponding to the article content of node j, such as a post or comment, in the context of positive or negative emotions extracted by node i in the text, as in the above case. In an oversimplified form, we can assume that any node in the reference text (such as a username) will count as 1 and 0.
Sijt — represents a direct financial rating, reflecting the fact that there is a financial transaction or transfer of funds from node i to node j at time t. Typically, it may be just a positive value for a given currency, but a cancellation or return of a trade under the protection of a smart contract may be considered a negative rating.
All of these, say, the complete formula of reputation can be as follows, assuming that time t can vary between T1 and T0, where T0 is the time of rating evaluation and T1 = T0-dT, and dT is the time window to consider because , called the “reputation period.”
Rj = SUMi(Ri * Fij)+ SUMit(Ri *(Oijt + Cijt + Dijt + Sijt))
In addition to the time range of ratings in the DT window, there may be other ranges of other criteria, such as a subset of ratings that match a particular standard and what we call a “reputation basis”, as follows:
A rating associated with a particular topic P, called a “reputation domain” (such as a tag in Steemit)
Specify the rating within node set S, called the “credit audience” (same as the user in Steemit)
Furthermore, due to the recursive nature of reputation calculations, there may be a parameter N indicating the maximum depth of graphics traversal when performing the calculation, which may be referred to as “credit depth.”
Then, it may be possible to implement a formula to filter any mutual rankings to eliminate cheating so that for any type of rating, there may be a “non-reverse” version, taking only the difference balance time of the mutual ranking transactions made during the given time period, so only Calculate asymmetric non-reciprocal ratings.
R’j = SUMi(Ri * Fij)+ SUMi(Ri * SUMt((Oijt-Ojit + Cijt-Cjit + Dijt-Djit + Sijt-Sjit)))
Alternatively, it can be rewritten with a difference ranking, such as O’ijt, C’ijt, D’ijt, and S’ijt as follows.
R’j = SUMi(Ri * Fij) + SUMi(Ri * SUMt((O’ijt + C’ijt + D’ijt + S’ijt))), O’ijt = Oijt — Ojit, etc.
The above formula can also be complex to avoid a closed reputation loop where L is used as the maximum number of links in the loop to detect and subtract the rating.
Each rating may have its own scaling and may need to provide scaling for the entire reputation assessment, so there may be more complex formulas to account for the scaling of all original ratings and final reputation.
R’’j = X1(Y1(SUMi(Ri * Z1(Fij)))) + Y2(SUMi(Ri * Z2)(SUMt((W1(O’ijt)+ W2(C’ijt)+ W3(D’) Ijt) )+ W4(S’ijt)))))))
In the above, X1 is the scaling function of the entire reputation, Y1 is the scaling function of its temporally unrelated part, and Z1 is scaling the time-independent rating itself. In addition, Y2 is zooming in on the time limit of the reputation Z2 score is scaling its peer score, and W1, W2, W3 and W4 are scaling the time-constrained score — original or difference. Each function can be a non-modified original value or implement a scaling factor for linear scaling, or non-linear, which can limit higher values. These functions can operate on a so-called “reputation basis”, keeping values within a controlled range, such as 0.0–1.0 or 0–100.
Finally, since recursive reputation calculations involving multiple ratings may be computationally expensive, it may need to be performed periodically, hourly, daily, weekly or monthly, so there should be a time period T indicating the duration of the time period.
Obviously, for a particular kind of blockchain, different kinds of rankings can be used. For example, for Steemit or Golos, we can use all of this — some examples are given at https://steemit.com/@aigents. For most blockchains, only financial information is processed and text content or explicit ratings are not supported, such as Bitcoin and Ethereum, only financial ratings apply. The above formula can be simplified separately.
On the other hand, in some blockchain applications, more information may be attracted to verify reputation, so along with KYC information (https://en.wikipedia.org/wiki/Know_your_customer), it may be used from the side Reputation information obtained — chain and out-of-chain social networks.
GitHub: https://github.com/Bitconch/BUS
Web:www.bitconch.io
Twitter:https://twitter.com/bitconch
Telegram : https://t.me/BitconchEnglish

