International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
ENHANCING PUBLIC REPUTATION SYSTEMS:
TRUST SCALING TO MITIGATE
VOTER SUBJECTIVITY
Manh Hung Nguyen
Intelligent Computing for Sustainable Development Laboratory (IC4SD)
Posts and Telecommunication Institute of Technology (PTIT), Hanoi, Vietnam
ABSTRACT
In the digital age, the reliability of public reputation systems is increasingly challenged by the subjectivity
of voter assessments. This paper presents a novel public reputation estimation method that leverages a
scaling trust framework to mitigate the influence of individual biases and enhance the accuracy of
reputation scores. We propose a scaling mechanism that adjusts the weight of each voter’s input according
to their trustworthiness, thereby reducing the impact of outlier opinions and fostering a more balanced
representation of public sentiment. The experiment results demonstrate that our method significantly
improves the robustness and fairness of reputation estimations compared to traditional models.
KEYWORDS
Reputation; Trust; Voter subjectivity; E-commerce
1. INTRODUCTION
The relationship between personal trust and public reputation has been a subject of research for
many years (Asiri and Alshamrani [1], Corbitt et al. [2], Dai and Cui [3], Falahat et al. [4], Jeon et
al. [5], Kas et al. [6], Kusuma et al. [7], Oghazi et al. [8], Zloteanu et al. [9]). Understanding how
individual trust translates into collective reputation has significant implications, especially in
contexts where decisions are made based on aggregated opinions, such as in online marketplaces,
social platforms, and review sites. Traditionally, personal trust has been the cornerstone of
reputation systems, with the assumption that if many individuals trust a product, service, or
person, this trust will be reflected in a strong public reputation.
In the digital age, online platforms increasingly rely on reputation systems to facilitate user
interactions, foster trust, and enhance overall engagement. With the rapid growth of ecommerce
and digital interactions, the need for reliable public reputation systems has become more critical
than ever. Consumers increasingly rely on these systems to make informed decisions, and
businesses depend on them to build and maintain trust with their customers. In response,
numerous trust-based public reputation models have been proposed, aiming to harness individual
trust assessments to create a collective reputation score as a kind of representative of the quality
and reliability of a given service. For instances, Xiong and Liu [10] present PeerTrust - a coherent
adaptive trust model for quantifying and comparing the trustworthiness of peers based on a
transaction-based feedback system. Wang and Vassileva [11] propose a trust model which is
based on Bayesian network and a reputation model which is based on recommendations in peer-
to-peer networks. Balaji et al. [12] proposed a reputation model which is calculated from users
feedbacks by using algorithm for weights and ratings computation. Goncalves et al. [13] proposed
DOI: 10.5121/ijcsit.2025.17102 17
International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
two major approaches which are based on public and permissioned blockchains. In the work of
Nguyen and colleges ([14], [15], [16], [17], [18], [19]), trust (and also distrust) is estimated from
the interactions in the past (experience trust), or from the evaluation of others (reputation), or
from both types of trust above. Jain and Singh [20] proposed a trust model based on opinion
dynamics temporal network. Lee et al. [21] presented a trust model based on the judgment of
buyers. Priya and Ponmagal [22] presented a trust model based on reputation. You et al. [23] also
presented trust model based on reputation.
However, despite their widespread adoption, trust-based public reputation models face a
significant challenge: they are inherently dependent on the subjectivity of voters. Differences in
experience and/or subjectivity among voters can lead to reputation scores that are not fully
representative of the actual trustworthiness of the entity being evaluated (Bufacchi [24], Dawson
[25], Gherghina and Marian [26], Moon et al. [27], Kusche [28]). This subjectivity can introduce
inconsistencies and inaccuracies, undermining the effectiveness of public reputation systems and
reducing their reliability.
To address this challenge, this paper proposes a novel approach: a scaling trust-based public
reputation model designed to mitigate voter subjectivity. Our method introduces a scaling
mechanism that adjusts the influence of each voter’s input based on their trustworthiness, as
determined by their past voting behaviour. By scaling the individual votes according to the
consistency and reliability of the voter’s previous assessments, our model aims to reduce the
impact of biased or anomalous opinions and produce a more accurate and balanced public
reputation score.
This paper presents scaling trust mechanism in reputation systems. In which, the scaling trust can
improve the accuracy and fairness of reputation scores, particularly in environments where voter
subjectivity poses a significant challenge. Through this work, we aim to contribute to the on-
going development of more reliable and trustworthy public reputation systems in the digital age.
The paper is organized as follows: Section 2 presents the similarity model. Section 3 presents
some experiments to evaluate the proposed model in some considered factors. Section 4 is the
conclusion and perspectives.
2. TRUST SCALING MODEL
Without loss of generality, we assume that:
• A public community could be considered as a multi-agent system, in which, member
agents are called agent i, agent j.
• There is possibly some transactions between agent i and agent j in the community. After
each transaction k, agent i may vote the service quality of agent j: tkij is called the real trust
of agent i on the agent j over the transaction k. Note that, tkij may differ from tkji for several
i ≠ j.
• Let’s [MIN,MAX] is the normalized interval value of transaction trust, therefore t kij∈
[MIN,MAX] for ∀i, j, k.
• Let’s tmini is the minimal transaction trust value voted by the agent i:
tmini=min{tkij | ∀j,k} (1)
• Let’s tmaxi is the maximal transaction trust value voted by the agent i:
tmaxi =max{ tkij | ∀j,k } (2)
18
International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
• The subjectivity difference of a voter i regarding the normalized interval [MIN,MAX] is:
(3)
The higher this value is, the more different the agent's subjectivity is.
• The Scaled trust of transaction k voted by agent i for agent j is estimated as follow:
(4)
where θ is a subjectivity difference threshold. If the subjectivity difference of a voter is higher
than this threshold, then the scaling of original trust is needed; otherwise, the classical (without
scaling) is applied to calculate the transaction trust of the voter. This subjectivity difference
threshold is possibly considered as a parameter which may influence on the model. It is thus
experimented in the evaluation section.
• The public reputation of agent j is thus estimated as the mean of all scaled trust voted for
agent j:
(5)
The more this public reputation is closed to the MAX value, the better the agent j.
3. EVALUATION
This section presents the evaluation of the proposed model by testing some sensitive parameters
used in the model such as the best threshold of θ, compare to the traditional reputation, and
testing in the case of limited number of transaction.
3.1. Simulated System Setup
In order to evaluate the public reputation by using the proposed scaled trust, we created a
simulated e-commerce system on the GAMA platform[29]. In this system:
• There are many seller agents who sell some products and many buyer agents who buy
some products.
• A product has a real utility value for buyer.
• A transaction occurs when a buyer agent decides to buy a product from a chosen seller
agent. The buyer agent has the right to evaluate the transaction quality (also the product
quality - based on the real utility value of the product) of the seller agent after each
transaction between them. The evaluated value is also called transaction trust.
• The public reputation of a seller agent is estimated from all the transaction trust evaluated
by all of its clients. This public reputation is published for all buyer agents in the system.
• Before making a transaction, a buyer agent chooses the best seller agent based on their
public reputation: The seller agent with the highest public reputation will be chosen.
19
International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
• The higher the real value of bought products that buyer agents obtain, the more efficient the
public reputation method is.
The used value of parameters in the system is listed in the Table 1.
Table 1. Simulated system configuration
Parameters Value
Number of seller 1000
Number of buyer 1000
Average number of product/seller 500
Average number of bought product/buyer 50
[MIN, MAX] [0,5]
3.2. Experiment 1: The best threshold θ
This experiment is conducted to determine the optimal value for the subjectivity difference
threshold (θ) by testing various values of this parameter.
3.2.1. Scenario
The experimental scenario is structured as follows:
• Iteration across θ values: The experiment is repeated for each of the following subjectivity
difference threshold (θ) values: 0%, 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%,
50%:
– For each specified value of θ, the simulated system, as described in Section 3.1, is
executed.
– Single buyer utility calculation: During the simulation, observe and compute the mean
real utility value of all products purchased by each buyer agent. This value is referred to
as the single buyer utility value.
– Overall buyer utility calculation: Next, calculate the average of the single buyer utility
values across all buyer agents in the system, normalized as a percentage. This is called
the overall buyer utility value.
– Repetition and averaging: The above steps are repeated 50 times for each given value of
θ. The mean of the overall buyer utility values from these 50 simulation runs is then
calculated, resulting in the buyer utility value for the specific value of θ.
• Comparison and selection of optimal θ: Finally, the buyer utility values for all the tested θ
values are compared. The θ value that yields the highest buyer utility value is identified as
the optimal subjectivity difference threshold. This optimal value will be utilized in
subsequent experiments.
20
International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
3.2.2. Results
Figure 1. Variation of buyer utility value with several subjectivity difference threshold θ
The results are illustrated in Figure 1. They show that the buyer’s utility value reaches its peak
when the subjectivity difference threshold (θ) is set at 10%. Beyond this point, the utility value
starts to decline. Specifically, the maximum buyer utility value observed is 96.74% at θ = 10%.
This indicates that a 10% threshold is optimal for maximizing buyer utility. Consequently, this
threshold will be applied in the subsequent experiments to ensure the most favorable outcomes.
3.3. Experiment 2: Compare to Classical Reputation
This experiment is conducted to evaluate the effectiveness of the proposed model, which
incorporates public reputation with scaled trust, in comparison to the traditional public reputation
model that does not include scaled trust. The comparison is performed across various system
configurations, each with different ratios of anomaly subjectivity buyer agent.
3.3.1. Scenario
The experiment follows this scenario:
• Iteration across anomaly subjectivity ratios: The experiment is conducted for each ratio of
anomaly subjectivity among buyer agents: 0%, 10%, 20%, 30%, 40%, 50%, 60%, 70%,
80%, 90%, For each ratio, the following steps are repeated:
– System initialization: Initialize the system with the specified ratio of anomaly
subjectivity buyer agent.
– System execution with two reputation methods:
• Classical method (without scaling trust): In this method, the reputation of a seller agent is
calculated directly as the mean of all original transaction trust values given by buyer agents
who purchased products from that seller.
This is done using the formula:
(6)
• Proposed method (with scaling trust): In this method, the reputation of a seller agent is also
calculated using the formula from Equation 5. However, this time it incorporates scaling
21
International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
trust, with the subjectivity difference threshold (θ) set at 10%, as determined to be the
optimal value in the first experiment.
– Single buyer utility calculation: During the simulation, observe and compute the mean
real utility value of all products purchased by each buyer agent. This value is referred to
as the single buyer utility value.
– Overall buyer utility calculation: Next, calculate the average of the single buyer utility
values across all buyer agents in the system, normalized as a percentage. This is called
the overall buyer utility value.
– Repetition and averaging: The above steps are repeated 50 times for each given value of
anomaly subjectivity ratio. The mean of the overall buyer utility values from these 50
simulation runs is then calculated, resulting in the buyer utility value for the specific
value of anomaly subjectivity ratio.
• Comparison of methods: Finally, the buyer utility values obtained from both the classical and
proposed methods are compared across all tested ratios of anomaly subjectivity among
buyer agents. For any given ratio, the method that produces the higher buyer utility value is
considered the superior approach.
3.3.2. Results
Figure 2. Comparison the buyer utility value between two methods (withoutand with scaling trust) on many
ratio of anomaly subjectivity buyer agent
The results are illustrated in Figure 2. They reveal that when there is 0% anomaly subjectivity
among buyer agents in the system, the buyer utility values obtained from both methods show no
significant difference. However, as soon as the anomaly subjectivity ratio reaches 10% or higher,
the buyer utility value for the method incorporating scaling trust consistently exceeds that of the
method without scaling trust. This trend indicates that the proposed method becomes increasingly
advantageous as the proportion of anomaly subjectivity among buyer agents increases.
3.4. Experiment 3: Comparison when there is Few Data
Since reputation depends on the past experiences of voters, it may not be accurate if there have
been too few transactions involving those voters. Therefore, the objective of this experiment is to
evaluate the proposed model compared to the classical public reputation model (without scaled
trust) under conditions where there is limited transaction trust data from voters in the past.
22
International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
3.4.1. Scenario
The experiment is taken with the following scenario:
• Iteration across average transactions per buyer: The experiment is repeated for cases where
each buyer has, on average, purchased 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 products. This means
that the system contains, on average, the corresponding number of transaction trust records
for each buyer.
– System initialization: Initialize the system with the specified ratio of anomaly
subjectivity buyer agent.
– System execution with two reputation methods:
• Classical method (without scaling trust): In this method, the reputation of a seller agent is
calculated directly as the mean of all original transaction trust values given by buyer agents
who purchased products from that seller. This is done using the formula 6.
• Proposed method (with scaling trust): In this method, the reputation of a seller agent is also
calculated using the formula from Equation 5. However, this time it incorporates scaling
trust, with the subjectivity difference threshold (θ) set at 10%, as determined to be the
optimal value in the first experiment.
– Single buyer utility calculation: During the simulation, observe and compute the mean
real utility value of all products purchased by each buyer agent. This value is referred to
as the single buyer utility value.
– Overall buyer utility calculation: Next, calculate the average of the single buyer utility
values across all buyer agents in the system, normalized as a percentage. This is called
the overall buyer utility value.
– Repetition and averaging: The above steps are repeated 50 times for each given number
of transaction trust. The mean of the overall buyer utility values from these 50
simulation runs is then calculated, resulting in the buyer utility value for the specific
number of transaction trust.
• Comparison of methods: Finally, the buyer utility values obtained from both the classical
and proposed methods are compared across all tested number of transaction trust. For any
given ratio, the method that produces the higher buyer utility value is considered the
superior approach.
3.4.2. Results
The results are shown in Figure 3. They reveal that when the average number of transaction trust
records per buyer is fewer than 3, the classical reputation method yields a higher buyer utility
value than the proposed method. However, when the average number of transaction trusts exceeds
3, the buyer utility value for the method incorporating scaling trust consistently surpasses that of
the classical method.
23
International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
Figure 3.Comparison the buyer utility value between two methods (withoutand with scaling trust) on
different small value of transaction trust
In summary, the proposed model could perform similarly to, or less effectively than, the classical
public reputation method when there are no anomaly subjectivity voters in the system or there is a
limited amount of transaction trust data from voters. However, in scenarios where anomaly
subjectivity voters are present or there are sufficient amount of transaction trust data, the public
reputation method with scaling trust proves to be more reliable. It offers better support for voters
in identifying the most trustworthy partners.
4. CONCLUSIONS
This paper proposes scaling trust framework which offers a significant advancement in enhancing
the accuracy and fairness of public reputation systems in the face of subjective voter assessments.
By dynamically adjusting the weight of voter inputs based on trustworthiness, our method
effectively minimizes the influence of biased or outlier opinions. The results from experiments
validate the robustness of this approach, demonstrating its superiority over traditional models in
producing more reliable and equitable reputation scores.
REFERENCES
[1] Ahmad Yahya Asiri and Sultan S. Alshamrani. Performance evaluation of a b2c model based on
trust requirements and factors. Scientific Programming, 2021(1):9935849, 2021.
[2] Brian J. Corbitt, Theerasak Thanasankit, and Han Yi. Trust and e-commerce: a study of consumer
perceptions. Electronic Commerce Research and Applications, 2(3):203–215, 2003. Selected Papers
from the Pacific Asia Conference on Information Systems.
[3] Qi Dai and Xiaolin Cui. The influence and moderating effect of trust in streamers in a live streaming
shopping environment. JUSTC, 52(2):6, 2022.
[4] Mohammad Falahat, Yan-Yin Lee, Yi-Cheng Foo, and Chee-En Chia. A model for consumer trust
in e-commerce. Asian Academy of Management Journal, 24(Supp. 2):93-109, Oct. 2019.
[5] Hyeon Gyu Jeon, Cheong Kim, Jungwoo Lee, and Kun Chang Lee. Understanding e-commerce
consumers’ repeat purchase intention: The role of trust transfer and the moderating effect of
neuroticism. Frontiers in Psychology, 12, 2021.
[6] Judith Kas, Rense Corten, and Arnout van de Rijt. Trust, reputation, and the value of promises in
online auctions of used goods. Rationality and Society, 35(4):387–419, 2023.
[7] Linda Kusuma, Sri Rejeki, Robiyanto Robiyanto, and Lala Irviana. Reputation system of c2c
ecommerce, buying interest and trust. Business: Theory and Practice, 21:314–321, 04 2020.
[8] Pejvak Oghazi, Stefan Karlsson, Daniel Hellstr¨om, Rana Mostaghel, and Setayesh Sattari. From
mars to venus: Alteration of trust and reputation in online shopping. Journal of Innovation &
Knowledge, 6(4):197–202, 2021.
24
International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
[9] Mircea Zloteanu, Nigel Harvey, David Tuckett, and Giacomo Livan. Judgments in the sharing
economy: The effect of user-generated trust and reputation information on decision-making
accuracy and bias. Frontiers in Psychology, 12, 2021.
[10] Li Xiong and Ling Liu. A reputation-based trust model for peer-to-peer ecommerce communities.
IEEE International Conference on E-Commerce, pages 275 – 284, 07 2003.
[11] Y.Wang and J. Vassileva. Trust and reputation model in peer-to-peer networks. In Proceedings
Third International Conference on Peer-to-Peer Computing (P2P2003), pages 150–157, 2003.
[12] Penubaka Balaji, O. Nagaraju, and D. Haritha. Commtrust: Reputation based trust evaluation in
ecommerce applications. In 2017 International Conference on Big Data Analytics and
Computational Intelligence (ICBDAC), pages 318–323, 2017.
[13] Maria Jose Angelico Gon，calves, Rui Humberto Pereira, and Marta Alexandra Guerra Magalhaes
Coelho. User reputation on e-commerce: Blockchain-based approaches. Journal of Cybersecurity
and Privacy, 2(4):907–923, 2022.
[14] Manh Hung Nguyen. A distrust model to detect faulty sensor in an IOT network. In Proceedings of
The 19th IEEE - RIVF International Conference on Computing and Communication Technology,
pages 53–58, 12 2023.
[15] Manh Hung Nguyen and Dinh Que Tran. A combination trust model for multi-agent systems.
International Journal of Innovative Computing, Information and Control, 9(6):2405–2420, 2013.
[16] Manh Hung Nguyen and Dinh Que Tran. A trust-based mechanism for avoiding liars in referring of
reputation in multiagent system. International Journal of Advanced Research in Artificial
Intelligence (IJARAI), 4(2):28–36, 2015.
[17] Manh Hung Nguyen and Dinh Que Tran. A trust model for new member in multiagent system.
Vietnam Journal of Computer Science, 2(3):181–190, 2015.
[18] Dinh Que Tran and Manh Hung Nguyen. Modeling trust in open distributed multiagent systems.
East-West Journal of Mathematics, Special issue for Contribution in Mathematics and Applications
III:98–108, 2010.
[19] Manh Hung Nguyen. Hybrid Deep Learning and Distrust Model for Fault Detection in IoT
Networks. International Journal of Science and Research (IJSR), p.166-170 , V.13(11), 2024. ISSN:
2319-7064
[20] Eeti Jain and Anurag Singh. Trust- and reputation-based opinion dynamics modelling over temporal
networks. Journal of Complex Networks, 10(4), 06 2022.
[21] Suk-Joo Lee, Cheolhwi Ahn, Kelly Minjung Song, and Hyunchul Ahn. Trust and distrust in
ecommerce. Sustainability, 10(4), 2018.
[22] Priya S and R.S. Ponmagal. Trust based reputation framework for data center security in cloud
computing environment. In 2023 7th International Conference on Computing Methodologies and
Communication (ICCMC), pages 1041–1047, 2023.
[23] Xinli You, Fujun Hou, and Francisco Chiclana. A reputation-based trust evaluation model in group
decision-making framework. Information Fusion, 103, 2024.
[24] Vittorio Bufacchi. Voting, rationality and reputation. Political Studies, 49:714–729, 02 2001.
[25] Stephen Dawson. Poll wars: Perceptions of poll credibility and voting behaviour. The International
Journal of Press/Politics, 29(1):206–226, 2024.
[26] Sergiu Gherghina and Claudiu Marian. Election campaign and media exposure: explaining objective
vs subjective political knowledge among first-time voters. Journal of Contemporary Central and
Eastern Europe, 32(1):37–53, 2024.
[27] Shin-Il Moon, Yunjin Choi, and Sungeun Chung. “this unfavorable poll result for my candidate
doesn’t affect me but others”: Third-person perception in election poll coverage. Asian Journal for
Public Opinion Research, 11(4):274–303, 11 2023.
[28] Isabel Kusche. Private voting, public opinion and political uncertainty in the age of social media.
Zeitschrift fur Soziologie, 51(1):83–98, 2022.
[29] Patrick Taillandier, Benoit Gaudou, Arnaud Grignard, Huynh Quang Nghi, Nicolas Marilleau,
Philippe Caillou, Damien Philippon, and Alexis Drogoul. Building, composing and experimenting
complex spatial models with the GAMA platform. GeoInformatica, 23:299–322, 04 2019.
25
International Journal of Computer Science & Information Technology (IJCSIT) Vol 17, No 1, February 2025
AUTHOR
Manh Hung Nguyen received B.E in Computer Science(CS) at PTIT in 2004,
M.Sc. in CS at the Institute Francophone International (IFI) in 2007, and Ph.D in CS
at the University of Toulouse, France, in 2010. He is currently working as an
associate professor at the Faculty of Computer Science, at The Posts and
Telecommunications Institute of Technologies (PTIT), Hanoi, Vietnam. His
domains of interest are: Artificial Intelligence, Multi-agent system, Modelling and
simulation of complex system, Machine learning
26
