Cross-Site Virtual Social Network Construction
Chenhao Xie
∗
, Deqing Yang
∗
, Jingrui He
†
, Yanghua Xiao
∗
∗
School of Computer Science, Shanghai Key Laboratory of Data Science, Fudan University, Shanghai, China
∗
Email: {xiechenhao,yangdeqing,shawyh}@fudan.edu.cn, Tel: (86)21-51355555
†
Arizona State University, AZ, USA. Email: jingrui.he@gmail.com
Abstract—
Given the plethora of social networking sites, it can be
difficult for users to browse too many sites and discover social
friends. For example, for a new diabetes patient, how can s/he
find the users with similar symptoms on different dedicated
sites and form supporting groups with them? Since different
sites may use different vocabularies, this problem is challenging
to match users across different sites. To address it, in this
paper, we present a tool to demonstrate how to construct a
virtual social network across multiple social networking sites.
Specifically, it uses bipartite graphs to represent the relation-
ships between users and their posts’ keywords in each site;
it bridges the gap between different vocabularies of different
sites based on their semantic relatedness through concept-based
interpretations; and it uses an efficient propagation algorithm
to obtain the similarity between users from different sites,
which can be used to construct the cross-site virtual social
network.
Keywords-cross-site, virtual social network, semantic match-
ing, graph propagation
I. INTRODUCTION
Social networks have experienced fast growth in the last
decade, such as Twitter and Facebook. They have become
fundamental platforms on which many people maintain
their friendships and share information with others. Be-
sides the generic ones, many dedicated social network-
ing sites have been created to help patients with a spe-
cific type of disease, such as diabetes. Examples include
TuDiabetes (http://www.tudiabetes.org/), Diabetic Connect
(http://www.diabeticconnect.com/), Diabetes Sisters (https://
diabetessisters.org/), etc. On one hand, these dedicated social
networking sites provide diabetes patients rich opportunities
to get exposed to recent developments on this disease and
to form support groups with people suffering from similar
symptoms; on the other hand, once a new user has selected
a social networking site, s/he is likely to stick to the site,
although it could be the case that many users from another
site share a lot of commonalities with this user, thus can
provide many useful tips and suggestions.
To help diabetes patients form support groups across mul-
tiple websites, we have developed a graph-based system for
constructing cross-site virtual social network. It is based on
recommendation techniques across heterogeneous domains
This demo was supported by NSFC (No.61472085, 61171132,
61033010), by National Key Basic Research Program of
China under No.2015CB358800, by Basic research project of
Shanghai science and technology innovation action plan under
No.15JC1400900, and by Shanghai Science and Technology De-
velopment Funds (13dz2260200, 13511504300). Corresponding
author is Yanghua Xiao.
introduced in [1], of which the goal is to recommend items
to users in another heterogeneous domain. In particular, in
our system, we build some bipartite graphs to represent the
relationships between users of each site and the keywords
used in their posts. Then, to bridge the gap between different
vocabularies used by different sites, we infer their semantic
relatedness through concept-based interpretation distilled
from online encyclopedias, such as Wikipedia. Finally, the
similarities between users of two different sites are com-
puted as similarity scores via an efficient graph propagation
algorithm. Such similarity scores can be used to construct
a cross-site virtual social network for the sake of forming
support groups for the diabetes patients.
Our techniques are different from: (1) existing work on
cross-domain recommendation [2] in the sense that we target
heterogeneous domains with barely overlapping feature sets
(vocabularies); and (2) transfer learning [3], [4] across
heterogeneous domains as we aim to build the connections
between users across different sites instead of learning
multiple predictive models.
To thoroughly demonstrate our techniques on constructing
cross-site virtual social network, we have implemented a
graphic tool which will be presented in the following section.
The rest of this paper is organized as follows. In Section
II, we introduce the key techniques used in our system,
followed by introducing the user interface and functionality
of our demonstration tool in Section III. And then we present
our paper’s related work in Section IV. Finally, we conclude
the paper in Section V.
II. KEY TECHNIQUES
In this section, we introduce our graph-based recommen-
dation system based on [1]: we start with some notations,
followed by the introduction of the global similarity and the
efficient computation of the relevance vectors, and finally
discuss semantic matching for bridging the gap between
different vocabularies used by different social networking
sites.
A. Notation
In this paper, for the sake of clarity, we consider 2
different sites. Formally, for the i
th
domain (i=1,2), we use a
bipartite graph G
i
= {V
i
, E
i
} to represent the relationships
between the users of one site and the keywords used in
users’ historical posts, where V
i
is the set of nodes in this
graph, and E
i
is a set of undirected edges. Let n
i
denote the
number of users in the i
th
site, and m
i
denote the number of
keywords. Therefore, V
i
consists of two types of nodes: n
i
user nodes, and m
i
keyword nodes. Let X
i
, n
i
×m
i
, denote