Notes

However, new consumers usually have limited or virtually no direct interaction or relationship with other consumers in the context of social commerce because of the strong community structure in social networks. This observation implies that new entrants often serve only the small communities (cliques) of direct acquaintances. A review is the subjective perspective of a consumer regarding his/her experiences in community activities and may merely represent his/her individual preference and opinion. Furthermore, trusted friends may not have similar tastes. Consequently, users who increasingly rely solely on their acquaintance communities for information readily encounter collective community bias [37]. In this case, the objectivity of reviews is not fully guaranteed under subjective environments.
The assessment of information credibility is a more challenging problem in social media than in conventional media because the username (virtual identity [25]) is the sole piece of information on its source. Social media is characterized by a lack of strong governance structures [37]. As a form of social governance, reputation mechanisms are among the most successful and widespread incentive mechanisms on the Internet. A reputation system uses a specific method (e.g., averaging,
social network-based, game-theoretic, probabilistic-based, or belief-based) to compute the reputation values for a set of objects (e.g., users, goods, or services) within a community based on rating data collected from others [21,43].
Social voting (user voting) is a simple but widely used reputation mechanism that has been applied to indicate the usefulness of reviews and the popularity of reviewers. Thus, the cognitive load of users is reduced. However, cheating behaviors such as vote-buying, vote-exchanging, and fake news reduce the reliability of voting results [29]. In addition, social voting mechanisms suffer from various types of bias, including the imbalance vote bias (users tend to value others’ opinions positively rather than negatively), winner circle bias (reviews with many votes attract much attention and therefore accumulate votes disproportionately), and early bird bias (the first reviews to be published tend to obtain more votes, but newly posted reviews are most likely to receive no votes or only a few votes) [31,46]. Without proper measures, the reputation mechanism obtains and produces unreliable information.
User credibility is generally addressed by solving the information credibility problem in social media. The veracity and objectivity of opinions are mainly influenced by the honesty and volume of the opinion contributors or so-called ‘‘raters’’ who share their experiences or opinions with others. In line with this information, different approaches have been proposed to determine and filter out deceptive information provided by raters [8,11,52]. The approaches that trust public information typically assume that the majority of raters are honest, whereas those that depend only on personal observations may fail in settings where in the observations of raters are inadequate [40]. Social network-based methods [16] that extract trust values from social links among users can improve the performance of reputation mechanisms based solely on the information obtained from limited personal observations or from unknown users. However, these links may not always be availableand ar e typically sparse in social commerce, especially for new users [57]. Furthermore, malicious users can still exploit the perceived social connections among users to easily publicize misinformation in social commerce if social links are available [1,7]. Some incentive-based proposals [13,27] are usually set up based on the rational behavior model of economics theory.
These proposals have a sound logical foundation but do not seriously consider user behavior strategies (such as collusion or bad-mouthing). These problems are intrinsic in the processes of obtaining genuine feedback about actual interactions and
estimating reputations accurately.
To circumvent the aforementioned problems, this paper proposes a new graph-based comprehensive reputation (CR) model for social commerce based on consumer behavior and psychological theory to improve the veracity and objectivity of opinions and to enhance consumer trust in (product and opinion) providers even in the presence of malicious and new users. This study is also motivated by the power of graph theory in data representation. The main salient features of the model proposed in this study are listed below.
The (activity) history of opinion contributors and their social network interactions instinctively generate a social context for opinions (e.g., reviews and votes) given the limited direct interactions among numerous consumers in the social commerce
context [32]. All of the data representation methods used in existing trust and reputation systems are applicationspecific and therefore limit the data inputs and representations that can be employed [19,43]. This study models the social context of opinions as users’ activities and social relationship networks (UASRNs in the present paper) through a graph-based data representation model. The proposed reputation model fully exploits the social context of opinions and incorporates both the behavior and social relationship reputations of a user to combat the scarcity of first-hand
information.