Can You Trust Online Ratings?
A Mutual
Reinforcement Model for Trust worthy Online
Rating Systems
Abstract:
The average of customer ratings on a product, which we call
a reputation, is one of the key factors in online purchasing
decisions. There is, however, no guarantee of the
trustworthiness of a reputation since it can be manipulated
rather easily. In this paper, we define false reputation as the
problem of a reputation being manipulated by unfair ratings
and design a general framework that provides trustworthy
reputations. For this purpose, we propose Trust-reputation, an
algorithm that iteratively adjusts a reputation based on the
confidence of customer ratings. We also show the
effectiveness of Trust-reputation through extensive
experiments in comparisons to state-of-the-art approaches .
Existing system:
The most common way to aggregate ratings is to use the average
(i.e., to assign the same weight to each rating), which may result
in a false reputation. For example, a group of abusers may inflate
or deflate the overall rating of a targeted product. The existing
strategies avoid a false reputation by detecting and eliminating
abusers. However, abusers cannot always be detected, and it is
possible that normal users may be regarded as abusers.
Consequently, existing strategies can exclude the ratings of
normal users or allow the ratings of abusers to be included in the
calculation of a reputation. existing strategies against shilling
attacks; all are trying to prevent the manipulation of ratings by
abusers. The classification algorithms for detecting shilling
attacks, however, may face situations where malicious users
cannot be detected and/or where normal users are considered as
malicious. As a result, there may be instances when a reputation
is calculated without the ratings of normal users or including the
ratings of malicious users.
Proposed system:
The proposed framework does not require clustering or
classification, both of which necessitate considerable learning
time. Though TRUE-REPUTATION does not require any learning
steps when solving a false reputation, extensive experiments
show that TRUE-REPUTATION provides more trustworthy
reputations than do algorithms based on clustering or
classification. The contributions of this paper are as follows. First,
we have defined false reputation and categorized various real-life
scenarios in which a false reputation can occur. The
categorization of the false-reputation scenarios helps us design
experimental scenarios similar to real-life situations. Second, we
have proposed a general framework to address a false reputation
by quantifying the level of confidence of a rating. The framework
includes TRUE-REPUTATION, an algorithm that iteratively adjusts
the reputation based on the confidence of customer ratings. Third,
we have verified the superiority of TRUE-REPUTATION by
comparing it with machine-learningbased algorithms through
extensive experiments.
Problem statement:
This paper defines the false reputation problem in online rating
systems and categorizes various real-life situations in which a
false reputation may occur. The understanding of why and when a
false reputation occurs helps us establish experimental situations.
In order to solve the false reputation problem, we proposed a
general framework that quantifies the confidence of a rating
based on activity, objectivity, and consistency. The framework
includes TRUE-REPUTATION, an algorithm that iteratively adjusts
the reputation based on the confidence of user ratings. Through
extensive experiments, we showed that TRUE-REPUTATION can
reduce the influence of various RAs. We also showed that TRUE-
REPUTATION is superior to the existing approaches that use
machine-learning algorithms such as clustering and classification
to solve the false reputation problem.
Future Work:
In a further study, we plan to develop an approach to accurately
separate an item score and a seller score from a user rating.
Separating the true reputation of items and that of sellers would
enable customers to judge items and sellers independently.
Implementation Of Modules:
False-Reputation Module:
In an online rating system, it is almost impossible to obtain the
ground-truth data because there is no way of knowing which
users have caused a false reputation in a real-life database. We
artificially establish various situations in which a false reputation
may occur and test the performance of the proposed algorithm in
these situations. In order to claim that the generated situations
are likely to occur in real-life online rating systems, we list various
scenarios involving a false reputation and categorize them
according to the types of user and situations. In this section, we
define dangerous users who cause a false reputation and
dangerous situations leading to a false reputation. Using the
definitions of dangerous users and dangerous situations,
Dangerous users.
Dangerous situations.
False Reputation Scenarios.
TABLE I
FALSE-REPUTATION SCENARIOS
Robustness:
In order to enhance the robustness of recommendation systems,
it is imperative to develop detection methods against shilling
attacks. Major research in shilling attack detection falls into three
categories:
1) classifying shilling attacks according to different types of
attacks.
2) extracting attributes that represent the characteristics of the
shilling attacks and quantifying the attributes
3) developing robust classification algorithms based on the
quantified attributes used to detect shilling attacks
Strategies for improving the robustness of multi agent systems
can be classified into two categories. The first group of strategies
is based on the principle of majority rule. Considering the
collection of majority opinions (more than half the opinions) as
fair, this group of strategies excludes the collection of minority
opinions, viewed as biased, when calculating the reputation.
Unfair Ratings:
The trustworthiness of a reputation can be achieved when a large
number of buyers take part in ratings with honesty .If some users
intentionally give unfair ratings to a product, especially when few
users have participated, the reputation of the product could easily
be manipulated. In this paper, we define false reputation as the
problem of a reputation being manipulated by unfair ratings. In
the case of a newly-launched product, for example, a company
may hire people in the early stages of promotion to provide high
ratings for the product. In this case, a false reputation adversely
affects the decision making of potential buyers of the product.
a reputation based on the confidence scores of all ratings, the
proposed algorithm calculates the reputation without the risk of
omitting ratings by normal users while reducing the influence of
unfair ratings by abusers. We call this algorithm, which solves the
false reputation problem by computing the true reputation, TRUE-
REPUTATION. Our framework for online rating systems and the
existing strategies in multiagent systems serve the same purpose
in that they are trying to address unfair ratings by abusers.
Buyer Modules:
Numerous studies have been conducted to improve the
trustworthiness of online shopping malls by detecting abusers
who have participated in the rating system for the sole purpose of
manipulating the information provided to potential buyers (e.g.,
reputations of sellers and recommended items). Especially in the
fields of multiagent and recommendation systems, various
strategies have been proposed to handle abusers who attack the
vulnerability of the system. In online rating systems, on the other
hand, a buyer can give only a single rating per item. Thus, the
relationship between buyers and items is significantly different
from the relationship between buyers and sellers; as such, the
graph structure of an online rating system is very different from
that of a multiagent system. This paper uses an approach that
considers the relation between buyers and items.
Conclusion:
This paper defines the false reputation problem in online rating
systems and categorizes various real-life situations in which a
false reputation may occur. The understanding of why and when a
false reputation occurs helps us establish experimental situations.
In order to solve the false reputation problem, we proposed a
general framework that quantifies the confidence of a rating
based on activity, objectivity, and consistency. The framework
includes TRUE-REPUTATION, an algorithm that iteratively adjusts
the reputation based on the confidence of user ratings. Through
extensive experiments, we showed that TRUE-REPUTATION can
reduce the influence of various RAs. We also showed that TRUE-
REPUTATION is superior to the existing approaches that use
machine-learning algorithms such as clustering and classification
to solve the false reputation problem. There are more factors
(other than those addressed in this paper) known to be elemental
in assessing the trust of users in the field of social and behavioral
sciences. We plan to study how to incorporate them into our
model to compute the reputation of items more accurately. In the
e-market place such as Amazon.com and eBay.com, buyers give
ratings on items they have purchased. We note, however, that the
rating given by a buyer indicates the degree of his satisfaction not
only with the item (e.g., the quality) but also with its seller (e.g.,
the promptness of delivery).
