Predicting the influence of users’ posted information for eWOM advertising in social networks

https://doi.org/10.1016/j.elerap.2014.10.001Get rights and content

Highlights

  • We defined the “influence score” of a post.

  • We proposed some potential predictive features from our own investigation.

  • We considered two scenarios for developing predictive models.

  • We conducted an empirical evaluation to evaluate the proposed features and models.

Abstract

Many social network websites have been aggressively exploring innovative electronic word-of-mouth (eWOM) advertising strategies using information shared by users, such as posts and product reviews. For example, Facebook offers a service allowing marketers to utilize users’ posts to automatically generate advertisements. The effectiveness of this practice depends on the ability to accurately predict a post’s influence on its readers. For an advertising strategy of this nature, the influence of a post is determined jointly by the features of the post, such as contents and time of creation, and the features of the author of the post. We propose two models for predicting the influence of a post using both sources of influence, post- and author-related features, as predictors. An empirical evaluation shows that the proposed predictive features improve prediction accuracy, and the models are effective in predicting the influence score.

Introduction

Social network websites (SNWs) provide users with a convenient and efficient online platform to share evaluations (reviews) of products with their contact groups. Because of the rapid proliferation of SNW participation, companies have been actively exploring the use of the vast amount of product reviews posted by SNW users to develop sustainable electronic word-of-mouth (eWOM) advertising strategies. Many companies – including, for example, Geico, Dell, and eBay – have been investing heavily in using social networks to influence consumer purchasing decisions (Kumar and Mirchandani, 2012, Evans and McKe, 2010). Empirical evidence has been established recently by academic researchers, which show the significant influence of online product reviews on consumer purchase decisions (Dellarocas et al. 2007). A survey by Gartner (2013) shows that “content creation” is the most important task for an e-marketer, and a survey by PowerReviews Inc. (2011) indicates that user-generated consumer reviews have the most influence on purchasing decisions among all the advertising contents (tools) on SWNs.

In order to utilize user-generated product reviews (posts) to develop eWOM advertising strategies, Facebook, the largest SNW, has launched several tools to engage users in sharing their own product ratings and reviews (Harris and Dennis 2011). For example, a service is offered to allow e-marketers to use the posts shared by Facebook users to automatically generate advertisements. PowerReviews has developed a sophisticated and aggressive service to allow post writers to combine their profile data and posts on e-commerce sites (Wonham 2010). When the service is used, a post can appear on the writer’s Facebook Wall as well as in his or her friends’ newsfeeds. It is expected that the trend of using self-generated product reviews in online advertising will expand rapidly in the future.

A practice of this nature relies on three postulations: (1) posts (information) shared by a “friend” could be very influential because they come from a trustworthy source and from first-hand experience; (2) certain “friends” are more influential than others; and (3) the quality and contents are important to the influence of a post (Yu et al. 2011). In order to develop effective advertising strategies in SNWs, it is essential to identify factors contributing to a successful eWOM process and to develop sound methods for predicting the influence of posted information.

In general, the influence of a post is determined jointly by the features of the post, such as contents and time of creation, and by the features of the author of the post. Although there is a rich literature on influence in social networks, these two sources of influence have been defined and studied individually. Existing studies on author-related features focus on identification of influential users (influencers) or opinion leaders (Bakshy et al., 2011, Cha et al., 2010, Kim and Han, 2009, Kiss and Bichler, 2008, Li et al., 2010, Li et al., 2011, Li and Du, 2011), and studies on post-related features investigate methods for identifying important predictive features and effective models for predicting the influence of the information shared among users (Adamic et al., 2008, Bian et al., 2009, Cao et al., 2011, Hong et al., 2011, Ratkiewicz et al., 2010, Suh et al., 2010, Yu et al., 2011). Furthermore, the influence of a post is often measured by the number of users who respond favorably to the post, such as “like” clicked counts (Yu et al. 2011), browser counts (Ratkiewicz et al. 2010), and forward counts (Hong et al. 2011).

In order to ensure the effectiveness of using a post in advertising on SNWs, we need a model for predicting the post’s influence with a satisfactory level of accuracy. We discuss three components associated with developing an effective predictive model: the definition of the target variable, the selection of predictors, and the selection of a functional form for linking the target variable and predictors.

It is evident that the influence of a post would be greatly undervalued if the variation in influence among users is not considered, when it receives a favorable response from a smaller number of influential users, or vice versa. To define the target variable of the model, which can adequately reflect the influence of a post, we first assign a weight to each user according to the number of users linked to him or her. Then, we define the “influence score” of a post as the sum of the weights of the people who respond favorably to the post.

In order to identify predictors for the predictive model, we compile a list of potential predictive features from two sources. The first is a comprehensive literature review to identify influencers and predict the influence of the contents of a post. The second source is our own investigation, based on the definition of the influence score. Using these predictive features, we propose two predictive models. The first is multiple-regression analysis, a useful and popular statistical tool. It is known that regression is based on a linear pattern for describing the relationship between the target and predictors. To allow for more flexible patterns, we also propose an ensemble classification model based on five techniques, including four data mining methods – Neural Networks, Decision Trees (C5.0), Naive Bayes, and Support Vector Machines (SVM) – and a classical statistical method, Logistic Regressions. It should be noted that, in the second model, the influence score is divided into several levels (intervals). We formulate the problem as a multi-label classification problem (as in Adamic et al., 2008, Hong et al., 2011, Yu et al., 2011), and propose a framework to predict the post’s influence level. To improve the overall accuracy of prediction, we exploit an ensemble model that aggregates the predictions made by multiple classifiers. From the evaluation of the two predictive models, we identify important predictive features and develop models for accurately predicting the influence of a post.

The remainder of this paper is organized as follows. In Section 2, we review the related literature, and, in Section 3, we present the definition of the influence score, predictive features, and the two predictive models. An empirical analysis of the predictive features and evaluations of the two predictive models are given in Section 4. A conclusion with a discussion for future studies is given in Section 5.

Section snippets

Literature review

We review the literature separately on finding influencers on SNWs and assessing the influence of the contents of a post. We identify five categories of predictive features for influence on SNWs; namely, content, authorial, temporal, exogenous, and topological. We provide a detailed literature review of content and authorial features and introduce briefly the three remaining types of features. These features, along with additional features proposed in this study, will be used to develop the

Predictive models

In this section, we present two models for predicting the influence score of a post. Since the target, the influence score, is continuous, multiple-regression analysis is an appropriate method for developing a prediction model and analyzing the relationship between the features and influence score. Regression analysis is a useful and popular tool for investigating the relationship between the response variable and a set of potential predictors, but it is limited in the (linear) pattern used to

Evaluation

We conduct an empirical evaluation to investigate (1) the relationship between the predictive features and the influence score, and (2) the effectiveness of the features in predicting the influence score of a post. For the evaluation, we selected a Facebook user and recorded all the posts from her contacts (friends). In the data preparation step, we eliminated advertisements and the posts artificially generated by computer software. As a result, we obtained, in total, 510 posts from 31 friends.

Discussion

In this study, we consider a situation where posts shared by users on SNWs are used in eWOM for advertising. We propose two models for predicting the influence of a post using both sources of influence, post- and author-related features, as predictors. The major results, contributions, limitations, and future research follow.

Acknowledgement

This study was supported in part by National Science Council of Taiwan, ROC, under Grant NSC-101-2410-H-004-013-MY2.

References (27)

  • D. Evans et al.

    Social Media Marketing: The Next Generation of Business Engagement

    (2010)
  • Gartner, 2013. Gartner’s Social Marketing Survey 2013. Accessed from...
  • Hong, L., Dan, O., Davison, B.D., 2011. Predicting popular messages in twitter. In: Proceedings of International...
  • Cited by (41)

    • Identification of influencers in online social networks: measuring influence considering multidimensional factors exploration

      2021, Heliyon
      Citation Excerpt :

      The following section 3.1.1-3.1.3 in this section will exhaustively show the measurement of each factor and each dimensional influence. As illustrated by Figure 2, information-based factors are used to estimate the contribution of users' posts, which includes the following four factors related to the text content [45]: the “popularity of the information”, the “type of information”, the “relevance of the message with respect to the topic”, and “users’ similarities”. The following Eqs. (1), (2), (3), and (4) show the measurement of these factors.

    • Customers' need for uniqueness theory versus brand congruence theory: The impact on satisfaction with social network sites

      2020, Journal of Business Research
      Citation Excerpt :

      Few studies have examined the role of the SNS on consumer engagement with brands (e.g. Algesheimer, Dholakia, & Hermann, 2005; Shih, 2009; Thorbjørnsen, Supphellen, Nysveen, & Pedersen, 2002; Wetsch, 2012). Much of the literature on online consumer psychology has focused on understanding consumer online brand relationships (e.g. Hayes, Luoma, Bond, Masuda, & Lillis, 2006; Morgan-Thomas & Veloutsou, 2013), online shopping attitudes (e.g. Hausman & Siekpe, 2009; Overby & Lee, 2006), and e-word of mouth (e.g. Chen, Tang, Wu, & Jheng, 2014; Okazaki & Taylor, 2013). However, the impact of consumer psychology and SNSs' brand similarity on customers' satisfaction with SNSs (e.g. Facebook) lacks examination.

    • Examining the influence of customer-to-customer electronic word-of-mouth on purchase intention in social networking sites

      2019, Asia Pacific Management Review
      Citation Excerpt :

      However, the mechanism by which eWOM transfers from customer to customer over SNSs, influencing consumption behavior, remains unclear. This study thus examines eWOM in C2C-sponsored SNSs (C2C eWOM), as derived from a set of activities, including reading posts and conducting peer communication with recommenders, with many studies only focusing on one of these issues (Chen, Tang, Wu, & Jheng, 2014; Wang et al., 2012). Although C2C eWOM within social media remains outside the control of the firm, a more nuanced view of such customer-driven content will provide firms with valuable data to better analyze customer demand (Larson & Watson, 2011).

    • Estimation and maximization of user influence in social networks

      2019, International Journal of Information Management
    View all citing articles on Scopus
    View full text