A Study on Individual Information Disseminating Behavior on Social Networking Sites (SNS)

在線社交網站上的用戶信息轉發行為研究

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
  • Kin Keung LAI (Supervisor)
  • Ye LU (Supervisor)
  • Ping Hu (External person) (External Supervisor)
  • Kin Keung LAI (External person) (External Co-Supervisor)
Award date11 Jun 2018

Abstract

Social networking sites(SNS) are conducive for information to go viral because numerous users are using SNS to receive and share information conveniently and efficiently. Therefore, SNS have become an important venue for businesses to diffuse information. For marketers, how to promote individuals on SNS to share information is an important question. However, There is a lack of a thorough understanding of such behavior. Prior studies tend to add features indiscriminately into the prediction model without examining the relevance of these features. This not only increases data collection cost, may lead to curse of dimensionality, but hinders us from understanding which factors are actually dominating an individual’s retweeting behavior. In this thesis, we first rank the relative importance of features by various feature selection algorithms and identify a subset of dominating features. And then we examine determinants of individual retweeting behavior from the perspective of an information receiver and construct a conceptual model based on the Elaboration-Likelihood Model (ELM). More than 60 million Twitter posts are used to verify the model. In the end, we focus on the relationship between topical relevance and individual retweeting behavior. The main innovation of this thesis is as follows:

1. We verify that individual information dissemination behavior on SNS is not random at all and propose a new feature selection algorithm which considers both relevance and redundancy of features. Using this algorithm, we pick out the dominating features which have an influential impact on individual dissemination behavior on SNS. Our research find that Among the most dominating six features, topical relevance and social tie strength are the most important factors, followed by #mention (@), #URL, retwttimes, and #hashtag. However, author-related factors are of the lowest importance and almost negligible. Comparison experiments show that under SVC or logistic regression, using dominating features can even improve the prediction performance to some extent. By picking out dominating features, this research not only reduces the cost of collecting features, helps us better understand individual forwarding behavior, but also makes sure that the prediction performance will not deteriorate. Therefore, the curse of dimensionality is avoided effectively.

2. Based on the above work, we carry out a comprehensive investigation of individual retweeting behavior. From the perspective of an information receiver, we consider all involved aspects of a social communication process. Based on ELM, we propose a conceptual model of individual retweeting behavior on SNS and then verify this model using Twitter data. In this model, topical relevance and information richness (#URL, #hashtag) belong to the central route as both factors require effortful elaboration. Social tie strength, informational social influence and other factors belong to the peripheral route as these factors do not require individuals to scrutinize the message arguments and allow them to make quick decisions. Analysis results show that both routes have significant effects on individual retweeting decisions. Among them, topical relevance, social tie strength and value homophily are the most important ones, followed by information richness, #mention and informational social influence. Author-related factors such as source trustworthiness have trivial impacts. Besides, we validate that social tie strength partially mediates the effect that value homophily has on individual retweeting behavior. This study expands the application area of ELM and offers at least one explanation for the contradictory findings about the effect of homophily on individual sharing behavior.

3.We investigate three types of moderators that moderate the effect of topical relevance on individual retweeting decisions, including individual characteristics, characteristics of tweets, and interpersonal relationships. A hierarchical linear model is employed to testify the moderating effects using Twitter panel data. The comparative experiment and robustness test show the superiority and stability of the hierarchical model. We find out that the impact of topical relevance is stronger for individuals with larger number of followers. However, the moderating effects of individual cumulative experience and gender are not significant. Users who tend to produce longer original tweets are more likely to expend more cognitive effort and considers more about topical relevance when making retweeting decisions. Besides, the effect of topical relevance is stronger for shorter time intervals, that is, for active individuals on SNS. When a tweet comes from a followee with similar tastes and preferences, the individual is prone to rely on the peripheral route and thus the impact of topical relevance is weaker. However, the impact of topical relevance is stronger for tweets coming from strong ties. This research expands the ELM theory and deepens our understanding of the impact that topical relevance has on individual forwarding behavior. The research reveals that this impact will be influenced by individual characteristics, characteristics of tweets and interpersonal relationships, and can provide guidelines on how to take advantage of topical relevance in online marketing.

    Research areas

  • Individual retweeting behavior, Elaboration-Likelihood Model, Social networking sites, Feature evaluation, Information dissemination, Twitter