Liu Z Q, et al. Sci China Inf Sci August 2017 Vol. 60 082102:4
service consumers and service pr oviders are movable, so our FCT model is more general than the classic
ones. Mor e over, the trust information is portable when service providers move due to the self-certified
characteristic of our FCT model.
• Service providers are proactive. In our FCT model, service providers proactively release their
mobile service re c ommendation information to nearby potential service consumers through their mobile
devices, which assists service providers to enjoy the trust of more service consumers and improve their
transaction volumes.
• Our FCT model i s context-aware. In our FCT model, the similarities of service type context
and service price context between the potential and previous transactions are calculated so as to ease the
notorious value imbalance attack [25].
• O
ur FCT model contains user preferences. To better characterize the trust, the se rvice rating
consists of various trust aspects with different preference weights, which can be determined by ser vice
consumers. Besides, service consumers can also set multi-attributes according to their requirements, and
the recommendation information which is mismatched with their multi-attributes will b e ignored.
• Our FCT model is of robus tne ss. In our FCT model, we consider four kinds of factor weights,
namely number, time decay, preference and context weights, so as to ease the collusion attack and value
imbalance attack.
• Our FCT model is of high performance. The comprehensive ex periments and analysis demon-
strate that our FCT model significantly surpa sses the CR model in terms of the service recommendation
performance in improving the successful trading rates of honest service providers and reducing the risks
of trading with malicious service providers, as well as the robustness against collusion attack and value
imbalance attack.
The rest of this paper is structured as follows. Sec tion 2 introduces some related work and its limita-
tions. Section 3 presents our FCT model and trust evaluation method in detail. Afterwards, compre hen-
sive experiments and analysis are shown in Section 4, followed by the conclusion in Section 5.
2 Related work
In recent years, LB SR has bee n widely studied in the literature, and lots of trust models have been put
forward. We review some c lassic trust mode ls according to the theory foundatio ns and rese arch e mphases
in them.
Context and user preference a re considered in ma ny studies. To provide personalized recommendations
for mobile tour planning, a novel LBSR model containing various factors (i.e., locatio n, preference, time,
constraint, etc.) was put forward by Yu et al. [10]. Afterwards, Waga et al. [11] proposed a context-aware
L
BSR system based on four factors, namely content, location, time and social network. This system
can provide useful recommendations and relevant items in most case s. Subse quently, Barranco et al. [1 2]
paid attention to on-the-move users a nd brought forward a LBSR system for traveling use rs. This system
incorporates both the speeds and trajectories of user s into context, and it can provide pers onalized serv ice
recommendations according to the current locations and driving speeds of users. Besides, Biancalana et
al. [13] took both context and user preference into conside ration and presented a LBSR system which can
identify the functional and personalized needs of users and provide per sonalized recommenda tions about
the interest points around the current locations of users.
There also exist a lot of researches based on other theories and technologies. Yang et al. [14] broug ht
forward a hybrid LBSR model by integrating the preferences derived from both check-ins and tips with
the sentiment analy sis technology, and they also proposed a social matrix factorization algorithm which
incorporates both social influence and loc ation similarity into location recommendations. In addition, a
novel LBSR mo del based on random walks in a user-space graph was raised by Noulas et al. [15]. This
m
odel combines both s ocial network and location visit frequency, and it outper forms the previous trust
models. Furthermore, Tan et al. [16] came up with a novel preference-oriented approach for location-based
store search based on data mining technology. This approach can efficiently search for the top-k nearby