Finding m-similar users in social networks using the m-representative skyline query
| DOI | https://doi.org/10.1108/IDD-04-2017-0030 |
| Date | 21 August 2017 |
| Pages | 121-129 |
| Published date | 21 August 2017 |
| Author | Kuo-Cheng Ting,Ruei-Ping Wang,Yi-Chung Chen,Don-Lin Yang,Hsi-Min Chen |
| Subject Matter | Library & information science,Library & information services,Lending,Document delivery,Collection building & management,Stock revision,Consortia |
Finding
m
-similar users in social networks
using the
m
-representative skyline query
Kuo-Cheng Ting and Ruei-Ping Wang
Department of Information Engineering and Computer Science, Feng Chia University, Taichung, Taiwan
Yi-Chung Chen
Department of Industrial Engineering and Management, National Yunlin University of Science and Technology,
Yunlin, Taiwan, and
Don-Lin Yang and Hsi-Min Chen
Department of Information Engineering and Computer Science, Feng Chia University, Taichung, Taiwan
Abstract
Purpose – Using social networks to identify users with traits similar to those of the target user has proven highly effective in the development of
personalized recommendation systems. Existing methods treat all dimensions of user data as a whole, despite the fact that most of the information
related to different dimensions is discrete. This has prompted researchers to adopt the skyline query for such search functions. Unfortunately,
researchers have run into problems of instability in the number of users identified using this approach.
Design/methodology/approach – We thus propose the
m
-representative skyline queries to provide control over the number of similar users that
are returned. We also developed an R-tree-based algorithm to implement the
m
-representative skyline queries.
Findings – By using the R-tree based algorithm, the processing speed of the
m
-representative skyline queries can now be accelerated. Experiment
results demonstrate the efficacy of the proposed approach.
Originality/value – Note that with this new way of finding similar users in the social network, the performance of the personalized
recommendation systems is expected to be enhanced.
Keywords Decision support, Social network, Database management, R-tree, Similar-user finding, Skyline query
Paper type Research paper
1. Introduction
Personalized recommendation systems are attracting
considerable attention (Hanze and Junmanee, 2005;Horozov
et al., 2006;Hsieh et al., 2012;Lu et al., 2012), for their ability to
make recommendations (e.g. products and destinations) based
on data obtained from social networks. When User A inputs a
search request, the system first searches for all similar users in the
social network and then makes a recommendation to User A
based on the preferences of these comparable users. For
example, User A is living in Tokyo and wants to make a trip to
Sapporo. But he/she is not sure what local attractions in Sapporo
would be interesting and so enters a relevant search request into
the system. Table I shows how a number of social network users
(including User A) have rated destinations they have previously
visited. The number 10 under Tokyo National Museum for User
A means that User A has visited the museum and rated it at 10.
The “null” rating under Sapporo Beer Museum for User A
means that this user has not previously visited this attraction. The
system then searches for other users similar to User A. It is clear
from the table that Users B and D are more similar to User A
than to Users C, E and F because Users A, B and D all highly
rated the Tokyo National Museum but poorly rated the Tokyo
shopping mall and central park. The ratings of Users C, E and F,
however, were exactly the opposite. The system then determines
which of the four destinations (Sapporo Beer Museum, Sapporo
shopping mall and Sapporo central park) would suit User A
based on how Users B and D have rated these attractions. As
Users B and D rated the Sapporo Beer Museum much higher
than the other three destinations, the system recommends User
A to visit the Sapporo Beer Museum.
Most conventional recommendation systems use cosine
similarity (Lu et al., 2012;Shamir and Tishby, 2010)or
k-means algorithms (Li et al., 2008) to combine the
multi-dimensional data of a user into a single score, and then
search for similar users based on this score. However, this
approach is actually highly illogical – as the dimensions of user
data are independent and unrelated, they should not be
combined into a single indicator (Hou et al., 2015). For
example, Tables I and II are largely similar, with the exception
of Table II having an additional User G. Both Users G and A
The current issue and full text archive of this journal is available on
Emerald Insight at: www.emeraldinsight.com/2398-6247.htm
Information Discovery and Delivery
45/3 (2017) 121–129
© Emerald Publishing Limited [ISSN 2398-6247]
[DOI 10.1108/IDD-04-2017-0030]
This work was supported in part by the Ministry of Science and
Technology of Taiwan, R.O.C., under Contracts MOST
105-2119-M-035-002 and MOST 105-2634-E-035-001. The authors are
grateful to the National Center for High-Performance Computing in
Taiwan for computer time and facilities.
Received 1 April 2017
Revised 5 July 2017
Accepted 7 July 2017
121
To continue reading
Request your trial