Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Study of Proof of Reputation Consensus

Implementation in a Recommendation Model


Abhishek Saxena
Mathematics and Mechanics Faculty,
Novosibirsk State University
Novosibirsk, Russia
a.saxena@g.nsu.ru

Abstract— Decentralised ledger systems are ready to be The PoR consensus algorithm may have different variations
absorbed in the present world. Arising with the problems and and parameters depending on its performance but the basic idea
attack against the blockchain environment, not only power is simple. Proof-Reputation "It is calculated as Rj = Sumit (Ri
consumption, but also speed, scalability and unhindered secure * Vij), where the sum of the positive rating or wire transfer to
platform for mass adoption and workabilty is required. Vij can be given as a negative vote or implicit rating, spent or
Henceforth, verification of reputation of the individuals or received. The main property of the algorithm is its iterative
validators over the network becomes a challenging task. nature, so each implicit or explicit vote is weighed or adopted.
An effective content recommendation in modern social media Content recommendation, such as news recommendation,
platforms should benefit both creators to bring genuine has been studied for many years in recommender systems
emoluments to them and consumers to help them get really (RSs). When applying to modern content distribution platforms,
interesting content. In this paper, we propose a model to such as Facebook and Steemit , an effective content
implement liquid democracy (PoR consensus) to Social
Explorative Attention Network (SEAN) for content
recommendation algorithm should consider benefits for both
recommendation [1]. SEAN uses a personalized content content creators and content consumers to help them get really
recommendation model to encourage personal interests driven interesting contents. While more accurate recommendation can
recommendation. Moreover, SEAN allows the personalization improve the consumers’ reading experience, it is regarded as a
factors to aŠ end to users’ higher-order friends on the social healthier content distribution ecosystem that encourages
network to improve the accuracy and diversity of individuals, especially small content creators, to share their
recommendation results. creative contents.
This paper deals with the 2-tier implementation of Proof of
One possible way is to use mechanism design approaches in
Reputation consensus in a recommendation model, their
resistance and vulnerability over the present environment. The SEAN algorithm to give recommendation based on the higher
paper would propose the implementation, usage, challenges, order friends of the particular person [1] .
vulnerability and future requirements of the system to cope up
with the challenge of the reputation consensus. In this paper we will see the implementation of PoR liquid
democracy model in the recommendation model above.
Keywords—decentralised systems, centralised systems, PoR,
SEAN, peer-to-peer, social computing, consensus, reputation. The implementation would be at 2-tier level i.e. the
recommendation based on the higher order friends would be
I. INTRODUCTION rated by the content user (apart from direct rating by the users
of the non-recommended content). This would made sense not
Most discussed problems in blockchain consensus which
only at a personal recommendation betterment for the user, but
can pose security risks, is double spending, if not adequately
also better ratings and wider visibility for quality content
discovered [7]. Also using a large pool as in case of PoW
creators (even if they are new or less rated in numbers) The
consensus, energy crisis becomes the problem. Along with
current work done on the thesis is at the stage of basic
challenges to validate the validators in a consensus always
formulation of the model based on the study of research papers
remains the primary concern as it may reduce the overall attack
on the implementation of the above [1,9].
on the blockchain. To eliminate the weaknesses associated with
the existing consensus algorithm, proof-of-reputation (PoR), Section 2 presents the background. Section 3 Conclusion,
which effectively expresses the idea of ​ ​ the so-called future work and scope.
"liquid democracy" algorithm [9] have been introduced [10].

©2020 IEEE 1
II. BACKGROUND that the platform will attract more clicks or reading actions.
Unfortunately, CF will produce unintended Mathew’s Effects
A. Why Proof of Reputation consensus?
(“The Rich Get Richer”) [20] which will hurt small/new
The purpose of the blockchain system is to support a content creators who may not be able to attract attentions.
transparent and reliable financial ecosystem, unbiased content Although most of the traditional CF based methods are often
management and communication systems, or reliable and called personalized recommendation [19, 21] and can be
efficient shared data storage - in all cases, cheaper, faster and generalized to social networks using social regularization [*],
safer. Since there are decentralized and distributed computer
they are in nature looking at global information and cannot
systems without centralized governance, validating the
solve this problem. There have been a few recommender
reputation of participants is well understood as a problem and
has been studied in many respects [2]. A reliable solution is systems that have studied the effects related to Mathew’s
important for peer-to-peer systems, where each node can Effect like using popularity [*] or quality [*] of the content or
communicate with every other node in the network [3]. The even game theory [*].
standard theoretical framework for approaching such a solution
is the so-called "Byzantine generals problem", where the C. Implementation of Liquid Democracy (PoR) in
number of participants with variable level of credibility has Recommendation Model using SEAN algorithm [1] as
some degree of consensus regarding their decision to enter foundation.
independently are stored in a public ledger, so that the
decisions are informed and beneficial to the whole community The following design of reputation calculations is based on
[4]. Since the reliability level of each node in the system is not previous works on reputation, also known as "social capital" or
known in advance, there is a need to reduce the risk of foreign "karma" [16]
nodes trying to reach consensus in favor of foreign alliances. In A simple computational model of reputation [10] is
today's distributed computing systems based on blockchain calculated as
technology, different consensus algorithms that implement
different types of voting provide a certain number of statistics, Rj = SUMit( Ri * Vijt )
each of which utilizes the quality of the node in the computer where Vijt can be either implicit rating as positive or
network to perform a reliable level of estimation [5]. As negative vote or implicit rating as wire transfer amount, either
discussed in introduction to face these challenges PoR spent or received, for node j being rated, node i supplying the
consensus serves to be the right path. rating and time t. Reputation is calculated for nodes, but it is
PoR compliance [10] is that it leverage recognition over assumed that each node plays a multi-agent social network role,
every user-owned digital asset. This means that fame is more such as a user, user account, software agent or blockchain
important than possession. PoR consensus, works as mining of system. Nodes during their operation, so that their artifacts can
reputation calculated for each node based on its performance be indirectly rated by posts and comments on social networks
over a social platform, communication channel, financial or by messages or tasks in multi-agent systems. The following
transactions or any other platform from which a reputation over five types of ratings are considered periodically on a peer-to-
a greater period of time can be fetched. This make PoR peer basis to calculate reputation, so node i will rank node j
consensus an involvement of every node on the network for the directly or indirectly with financial rating value over time. In,
verifier. Hence decentralised PoR consensus which actually Rj = SUMi (Ri * Fij) + SUMit (Ri * (Oijt + Cijt + Dijt +
work on “Reputation mining” [9] as like PoW, are worked Sijt))
upon and effectively being implemented.
Fij - Node Identifies clear direct ratings based on the fact
that node i trust node j. On traditional social networks,
B. Current Recommendation Models including Steemit, this corresponds to a "follower" link
between two users. Such a rating is only in time, but different
Existing recommendation algorithms lack consideration for moments may occur when this link is created or removed.
balancing both content creators and consumers. Content Typically, it can be 0 or 1, but we can set up a system with a
recommendation methods can be content based or collaborative "five stars" rating or any range of values, including a negative
filtering (CF) based ones. Content based methods [17, 18] rating. (THIS CAN SERVE AS RATING FOR
memorize historical reading/watching content of a user and RECOMMENDATIONS FROM ALREADY FOLLOWING
predict his/her future reading/watching content based on CONTENT CREATORS)
features or similarities of both contents. Such approaches
emphasize on particular topics for a user and may not be able Oijt - Node j refers to the fact that node i is clearly
associated with the reference of voting to the written artwork.
to encourage diversity of recommendation results unless a
On social networks like Facebook, this is commonly called
content consumer actively searches for new topics. CF is
"Like", on Steemit it is called "Vote" and it can be 0 or 1, but
considered as a complementary technique to content based we can imagine values in any range such as YouTube.
approaches for content recommendation [19] as it can oversee Contains negative and positive content ratings as -1, 0 and 1.
the global users’ clicking behaviors and make recommendation (THIS CAN SERVE RATING TO THE RECOMMENDED
based on similar users or similar contents, where similar CONTENT BASED ON HIGHER ORDER FRIEND OR
contents mean contents clicked/read by the same group of users. MUTUAL INTEREST BASIS)
CF usually optimizes based on global behavior information so

2
Cijt - Node i have posted an indirect article about textual III. CONCLUSION, FUTURE WORK AND SCOPE
content, such as a comment on the artifact j wrote, derived
from a positive or negative sentiment text, so it can be anything A. Conclusion
positive or negative, any scale. In the most simplified form, we The present study of implementation of PoR using liquid
can assume that it is 0 or 1, that is, if all the current comments democracy model for recommendation system on a platform
are positive or -1, 0 and 1 if negative comments are considered. like Steemit. Initial study has been performed over the already
(THIS WOULD SERVE AS SECOND LAYER (TIER-2) OF present research paper and articles relevant to above.
RATING FOR CONTENT CREATOR AND FOR FUTURE
The results of this study can help provide giving a balance
RECOMMENDATIONS FROM SAME CREATOR FOR
to content creator environment, along with providing better
THE USER)
recommendations for the content user. It could help specially
Dijt - Node refers to the direct ratings of whether i have new content creators or less famous creators to get wider
posted text content or comments with reference to node j in audience. Not only this, it would widen the user experience
terms of positive or negative emotions derived from the text, also.
such as the context above. For. In the most simplified form, we
B. Future work and scope
can assume that any reference to a node in a text such as a
username is counted as 1, 0 or -1. (THIS WOULD SERVE AS The next phase of study of implementation of PoR
SECOND LAYER (TIER-1) OF RATING FOR CONTENT consensus is to prepare an analytical model for liquid
CREATOR AND FOR FUTURE RECOMMENDATIONS democracy in recommendation system and find the resistance
FROM SAME CREATOR FOR THE USER) and vulnerabilities of performance of this model.
*TIER 1- USER’S PERSONAL TAKE ON CONTENT The future scope of this Recommendation system overall
CREATOR (IN THE SENSE USER IS GETTING DIRECT) can be very beneficial and effective in environments like
YOUTUBE, where the new content creators have to face a alot
*TIER 2- USER’S GENERAL TAKE ON CONTENT of challenges to just get the recognition as they are facing
CREATOR (INDIRECT COMMENT) competition from giants and worldwide accepted creators on
BUT BOTH SERVING AS SECOND LAYER OF the top recommendations. This challenging environment is so
RATINGS FOR CONTENT CREATOR, AS USER TAKES fierce that maximum number of new content creators (even
ANOTHER STEP TO QUOTE OR COMMENT ON THE having enough knowledge or better content) just quit before
CREATOR’S WORK even reaching the monetization of the channel (considering
Youtube)
Sijt - refers to a direct financial rating reflecting the fact
that a financial transaction or fund transfer occurs from node i Reputation system have a very capable foundation to build
to node j over time t. Generally, this is only a positive value in the ecosystem which could change the way how entertainment,
a given currency, but it can also be a matter of treating news or any social media platform offers recommendations. If
transactions or cancellations under the umbrella of smart considered now in the digital society, with the power of
contracts for negative ratings. (THIS CAN SERVE AS THE reputation system, the whole imbalance can be controlled in
HIGHEST RATING OVERALL AND APPRECIATION FOR the coming future.
THE CREATOR, AND MORE RECOMMENTATIONS FOR
USERS FOR THE SAME)
In all this, the full formula for reputation may be the
following, assuming hitting that time T can vary between T1
and T0, where T0 is the time for rating and T1 = T0 - dT, and
dT is the temporary window to take into account accounts,
called “reputation period”.
Flexibility of such reputation scoring system helps in
binding the complexity level according to the environment
required. It can be a social platform reputation system, a
financial blockchain authorization, or more closed centralised
working system where only certain validated nodes are
required.
Moreover, it can be based on calculation performance
required at certain period such as hourly, daily, weekly or
monthly level. (WE CAN ADD FURTHER LEVELS FOR
FEATURES LIKE SHARING, PROMOTING, ETC USING
THE REPUTATION DEPTH MODEL IN THE LIQUID
DEMOCRACY)

3
REFERENCES [8] Marshall C. Pease, Robert E. Shostak, and Leslie Lamport. Reaching
agreement in the presence of faults. J. ACM, 27(2):228–234, 1980.
[9] https://en.wikipedia.org/. Wiki / Representative_Democracy.
[10] https://steemit.com/blockchain/@aigents/proof-of-reputation-as-liquid-
[1] Wenyi Xiao , Huan Zhao , Haojie Pan , Yangqiu Song , Vincent W. democracy-for-blockchain
Zheng , Qiang Yang: Beyond Personalization: Social Content
Recommendation for Creator Equality and Consumer Satisfaction [11] https: // medium .com / @ DebrajG / how-Byzantine General Deleted -
https://arxiv.org/pdf/1905.11900v3.pdf Glance-in-Blockchain-370fe637502 c
[2] Swamynathan G., Almeroth K., Zhao B.: The design of a reliable [12] https://www.fxempire.com/education/article/51-attack-explained-the-
reputation system. Electron Commer Res 10: 239–270, DOI attack-on-a-blockchain-513887
10.1007/s10660-010-9064-y, 31 August 2010, pp.239-270 (2010). [13] Parinya Ekparinya, Vincent Gramoli, Gullaume Jourjon. The Attack of
[3] Gupta M., Judge P., Ammar M.: A Reputation System for Peer-to-Peer the Clones Against Proof of Authority
Networks. NOSSDAV’03, June 1–3, 2003, Monterey, California, USA, [14] https://coincentral.com/sybil-attack-blockchain/
ACM 1-58113-694-3/03/0006 (2003).K. Elissa, “Title of paper if [15] https://github.com/ethereum/EIPs/issues/225
known,” unpublished. [16] Social Network-2017- IEEE.pdf.
[4] Lamport L., Shostak R., Pease M.: The Byzantine Generals Problem. [17] Hongwei Wang, Fuzheng Zhang, Xing Xie, and Minyi Guo. Dkn: Deep
ACM Transactions on Programming Languages and Systems, Vol. 4, No. knowledge-aware network for news recommendation.
3, July 1982, Pages 382-401 (1982).
[18] Xuejian Wang, Lantao Yu, Kan Ren, Guanyu Tao, Weinan Zhang, Yong
[5] Cachin C., Vukolić M.: Blockchain Consensus Protocols in the Wild. Yu, and Jun Wang. Dynamic aŠ ention deep model for article
31st International Symposium on Distributed Computing (DISC 2017), recommendation by learning human editors’ demonstration.
Editor: Andréa W. Richa; Article No. 1; pp. 1:1–1:16, Leibniz
International Proceedings in Informatics, 10.4230/LIPIcs.DISC.2017.1 [19] Abhinandan S Das, Mayur Datar, Ashutosh Garg, and Shyam Rajaram.
(2017). Google news personalization: scalable online collaborative €
ltering.
[6] Igor Barinov, Vadim Arasev, Andreas Fackler, Vladimir [20] Daniel Rigney, editor. The Matthew Effect, How Advantage Begets
Komendantskiy, Andrew Gross, Alexander Kolotov, and Daria Isakova. Further Advantage. Columbia University Press, 2010
POSDAO: Proof of stake decentralized autonomous organization. [21] Jiahui Liu, Peter Dolan, and Elin Rønby Pedersen. Personalized news
Technical report, SSRN, 2019. recommendation based on click behavior.
[7] Ethan Heilman, Alison Kendler, Aviv Zohar, and Sharon Goldberg. [22]
Eclipse attacks on bitcoin’s peer-to-peer network. In 24th {USENIX}
Secur. Symp., pages 129–144, 2015.

You might also like