2 Related Work
With the abundance and fast growth of online services available, how to rank services
effectively has been a pressing problem which attracts lots of researchers doing valuable
and interesting research [4, 5]. Related work in this area can be classified into two main
streams [4]: ranking method based on attribute where the ranking is only related to the
features of services, such as sales, price, reputation and ratings; and personalized ranking
method.
(1) Attribute-Based Ranking. Kai Hwang [6] combines the services attributes such as
sale price, quantity ordered, delivery time, seller trust, and service quality to ranks
sellers. S.N. Junaini [5] proposes a framework consisting of the usability factors
such as simplicity, attractiveness, effectiveness, service image, site information,
service details and so on to rank online services. Services are ranked based on each
considered service feature, using text analysis techniques to extract condensed
information from massive customer reviews in [7–9]. Some people also select
services in accordance with the rank of reputation values [10]. Jiliang Tang obtains
the user preference from users’ reviews with changes over time, thereby compute
the trust values of services and rank the services. EBay
1
computes the reputation of
the service provide through collecting the feedback information from users after
each transaction. Feedback information provided by users includes positive rating,
neutral rating and negative rating. The reputation value is the result that the number
of total positive ratings minus the number of total negative ratings. Amazon
2
also
uses the feedback information to compute the reputation value. The only
difference
is the reputation value is the result of the average of all ratings. However, all of the
researches mentioned above, whether based on the ranking of service feature or
based on the ranking of reputation, involve the users’ feedback information, but
ignore the fact that the feedback information are given by customers actually
incomparable. Moreover, some users maybe provide dishonest opinions. As a
result, the result of ranking is subject to manipulation.
(2) Personalized Ranking. In [11], services are ranked based on the users’ own pref‐
erences and also on the information in the different search engines about the serv‐
ices. Ghose et al. [12] proposes a ‘utility-preserving’ ranking strategy from an
economic perspective which takes multi-dimensional preference and customer
heterogeneity into consideration. In [13], a new personalized service ranking
method is proposed based on estimating consumer information search
benefits
and
considering the uncertainty and confidence. All of these works ignore the relation‐
ship between different services, although some works take the users’ preference
into account.
Different
from these works, the approach proposed in our paper
processes the ratings data firstly, rather than use it directly. Additionally, the
approach takes the relationship among different services into account.
1
http://www.ebay.com/.
2
http://www.amazon.cn/.
Ranking Online Services by Aggregating Ordinal Preferences 43