Social Proof: The Impact of Author Traits on Influence Detection
Sara Rosenthal, Kathy McKeown · 2016
It has been claimed that people are more likely to be influenced by those who are similar to them than those who are not.In this paper, we test this hypothesis by measuring the impact of author traits on the detection of influence.The traits we explore are age, gender, religion, and political party.We create a single classifier to detect the author traits of each individual.We then use the personal traits predicted by this classifier to predict the influence of contributors in a Wikipedia Talk Page corpus.Our research shows that the influencer tends to have the same traits as the majority of people in the conversation.Furthermore, we show that this is more pronounced when considering the personal traits most relevant to the conversation.Our research thus provides evidence for the theory of social proof. IntroductionThe psychological phenomenon of social proof suggests that people will be influenced by others in their surroundings.Furthermore, social proof is most evident when a person perceives the people in their surroundings to be similar to them (Cialdini, 2007).This tendency is known as homophily.One manner in which people can be similar is through shared author traits such as the demographics age (year of birth), gender (male/female), and religion (Christian/ Jewish/ Muslim/ Atheist), as well as political party (Republican/Democrat).In this paper, we explore the impact of social proof via author traits in detecting the most influential people in Wikipedia Talk Page discussions.We present an author trait detector that can detect a suite of author traits based on prior state-of-the art methods developed for individual author traits alone, and use it to classify individuals along four author traits: age, gender, religion, and political party.We train the classifier using automatically * Work completed as graduated student at Columbia University labeled or prior existing datasets in each trait.Our classifier achieves accuracy comparable to or better than prior work in each demographic and political affiliation.The author trait classifiers are used to automatically label the author traits of each person in the Wikipedia Talk Page discussions.An influencer is someone within a discussion who has credibility in the group, persists in attempting to convince others, and introduces topics/ideas that others pick up on or support (Biran et al., 2012; Nguyen et al., 2013b).We use supervised learning to predict which people in the discussion are the influencers.In this paper we use the demographics and political affiliation of the authors in the Wikipedia Talk Page as features in the classifier to detect the influencers within each discussion.This is known as situational influence.In contrast, global influence refers to people who are influential over many discussions.It is important to explore situational influence because a person can be quite influential in some Wikipedia Talk Page discussions but not at all in others.We show that social proof and homophily exists among participants and that the topic of the discussion plays a role in determining which author traits are useful.For example, religion is more indicative of influence in discussions that are religious in nature such as a discussion about the Catholic Church.In the rest of this paper we first discuss related work in influence detection.We then describe our author trait classifier, related work, and the datasets used to train the models.All of our datasets are publicly available at http://www.cs.columbia.edu/˜sara/ data.php.Next, we discuss the Wikipedia Talk Page (WTP) dataset and how they were labeled for influence.Afterwards we discuss our method for detecting influence, the experiments and results.Finally, we conclude with a discussion of the impact of author traits on influence detection.