Exploring user topic influence for group recommendation on learning resources
Feng Wang
§
, Wenjun Jiang
†
, Shuhong Chen
‡
, Dongqing Xie
‡
, Guojun Wang
‡,∗
§
School of Information Science and Engineering, Central South University, Changsha 410083, China
†
College of Computer Science and Electronic Engineering, Hunan University, Changsha 410082, China
‡
School of Computer Science and Educational Software, Guangzhou University, Guangzhou 510006, China
*Correspondence to: csgjwang@gmail.com
Abstract—With the rapid development of online social net-
working services, the recommendation systems are facing new
challenges in recommending resources to a target group of
users. How to make a trade-off between group’s preference and
influencer’s impact is one of important problems, especially in
the group recommendation on learning resources. In this paper,
we propose a User Topic Influence (UTI) model, which fully
exploits user topic influence together with group’s preference
and item content for group recommendation on learning
resources. Based on the UTI model, we mine the topic influence
and group’s preference through statistical inference, then we
develop a parameter learning algorithm through Expectation
Maximization (EM) algorithm. In addition, we propose a
group recommendation algorithm with the consideration of
group’s preference and influencers’ impact. The experimental
results demonstrate that our proposed group recommendation
algorithm performs better than other five alternatives.
Keywords-group recommendation; learning resource recom-
mendation; probabilistic generative model; user influence;
topic model
I. INTRODUCTION
The availability of online social networks’ data promote
the researchers pay attention to the problem of how to
recommend resources to a group of user effectively. The
existing researches have been involved in the fields of
tourism [1], music [2], TV programs [3], movies [4], and
web pages [5], etc. However, there is few research involves
in the learning resources recommendation. To the best of
our knowledge, this paper is the first attempt to explore
social influence in the problem of group recommendation
on learning resources.
As the research object of this paper, learning resources
recommendation has its own characteristics. First, different
to other fields of resource recommendation, users can not
recognize whether the learning resources are useful nor-
mally. So users tend to trust the influencers’ selection in
the learning resources recommendation for the influencers’
authority, expertise or other personal reasons [23]. Second,
users behave different when selecting learning resources as
a group member, they will tend to match their individual
and group preference to the topics of the group [22].
Third, due to the learning group usually pay attention to
specific topics, the recommendation resources selected by
the learning group are topic relevant. Thus, to capture the
Figure 1. Example of user profile analysis and influencers finding process.
measure factors in group recommendation, in this work, we
propose a probabilistic generative model that incorporates
user topic influence, group’s preference and item content.
Figure 1 shows a simple example of group user profile
analysis and influencers finding process. This reflects the
idea of recommending learning resource with the considera-
tion of topic (topic model analysis) and influencer (influence
evaluation).
The key problem in the process of group recommenda-
tion is how to recommend resources to a group of users
effectively, but the group member has different preferences.
Thus, how to make the trade-off among group members’
preference is the first challenging. The existing solutions
contains the memory-based approach and the model-based
approach. The memory-based approach can be further di-
vided into preference aggregation approach [1, 3, 4] and rank
aggregation approach [2, 5, 6]. The aggregation functions
can be based on average (AVG) and least misery (LM) strate-
gies, etc [3, 7]. The model-based approaches are heuristic,
and these methods can explore the interaction between users
[14, 15]. However, the existing algorithms ignore the effect
of influencers in the process of group recommendation.
In this paper, we try to capture the measure factors in
group recommendation on learning resources. In summary,
the main contributions of this work are summarized as
follows:
• We propose a probabilistic generative model, which
aims at recommending learning resources to a target
group, especially in the field of education. The UTI
978-1-5386-0435-9/17/$31.00 ©2017 IEEE