Confluence: Conformity Influence in Large Social Networks

Conformity is a type of social influence involving a change in opinion or behavior in order to fit in with a group. Employing several social networks as the source for our experimental data, we study how the effect of conformity plays a role in changing users’ online behavior. We formally define several major types of conformity in individual, peer, and group levels. We propose Confluence model to formalize the effects of social conformity into a probabilistic model. Confluence can distinguish and quantify the effects of the different types of conformities. To scale up to large social networks, we propose a distributed learning method that can construct the Confluence model efficiently with near-linear speedup.

Our experimental results on four different types of large social networks, i.e., Flickr, Gowalla, Weibo and Co-Author, verify the existence of the conformity phenomena. Leveraging the conformity information, Confluence can accurately predict actions of users. Our experiments show that Confluence significantly improves the prediction accuracy by up to 5-10% compared with several alternative methods.

This work has been accepted in KDD 2013. If you use data sets provided or refer to the model, please cite:

@INPROCEEDINGS{Tang:13KDD,
     AUTHOR = "Jie Tang and Sen Wu and Jimeng Sun",
     TITLE = "Confluence: Conformity Influence in Large Social Networks",
     BOOKTITLE = "KDD'13",
     YEAR = {2013},
     pages = {347-355}
}

The full paper can be downloaded here[pdf].

Conformity is the act of matching attitudes, beliefs, and behaviors to group norms [8]. The phenomenon of conformity could occur in small groups or the whole society, as a resultant of peer influence or group pressure. Conformity can have either good or bad effect depending on the situation. For example, it helps form and maintain the social norms, and helps prevent acts that are perceptually dangerous. Conformity can be influenced by various factors such as individual status, peer influence and group pressure. Therefore there is a clear need for quantitative methods for measuring conformity from different aspects, so as to understand the complex dynamics in social networks.

Conformity was first studied by psychologists through interviews with small groups of participants [17]. In economics, Bernheim [3] found that sometimes people are willing to conform simply because they recognize that departure from the social norm may impair their status. Bernheim further proposed a theory of social conformity and presented a model to describe the conformity process. However, due to the lack of real data, he only studied the model from the theoretical aspect. With the rapid proliferation of online social networks such as Facebook, Twitter, and Flickr, it becomes feasible and also very necessary to conduct an in-depth investigation of the conformity problem on real large social networks. In practice, the effect of conformity has been also observed in online social networks. For example, Bond et al. [4] reported results from a randomized controlled trial of political mobilization messages delivered to 61 million Facebook users. They found that when one is aware that their friends have made the political votes, their likelihood to vote will significantly increase. Bakshy et al. [2] also found that when their friends click an ad, they will be more likely to click the same ad.

From a broader viewpoint, conformity can be seen as a special type of social influence. There are a bulk of studies on social influence analysis. These studies can be roughly classified into three categories: influence testing [1, 9, 20], influence quantification [12, 13, 23, 32], and influence maximization models [6, 18]. However, most of the works focus on qualitative study of social influence. With an exception, Tang et al. [32] presented a Topical Affinity Propagation (TAP) approach to quantify the topic-level social influence in large networks. However, they do not distinguish the effect of peer influence and group conformity.

To address the above challenges, we formally define the problem of conformity influence analysis in social networks and categorize conformity into individual conformity, peer conformity, and group conformity. We propose an Confluence method to formalize the effects of social conformity into a probabilistic factor graph model. The Confluence method can distinguish and quantify the effect of the different types of conformities. We test the proposed method on four social networks, i.e., Flickr, Gowalla, Weibo1, and Co-Author. Our experimental analysis verifies the existence of conformity in all the four social networks. We further apply the Confluence method to model and predict users’ online behavior. To scale up to large social networks, we propose a distributed learning version of Confluence that can leverage parallel computing to significantly reduce the computational time to nearly linear speedup (∼ 9× with 16 computer cores). Figure 1 shows the prediction accuracy in terms of Area Under Curve (AUC) on the four networks. It can be clearly seen that by incorporating the conformity factors, the prediction performance can be significantly improved up to 5-10% compared to various baseline methods.

Figure 1: Action prediction accuracy (AUC) of different methods by considering the effect of conformity in four networks.

 [Download]

Our goal is to design a unified model to capture users’ action dynamics and model conformities from different levels. We propose Confluence, a conformity-aware factor graph model. To handle real large networks, we develop a distributed model learning algorithm.

We evaluate the proposed method on four different genres of networks: Flickr, Gowalla, Weibo, and Co-Author. Table 2 lists statistics of the four networks.

• Flickr is a photo sharing network. Users on the site can share photos and add comments to other photos. Flickr users can also create and join different groups. The data set spans the period from Apr. 1st, 2012 to Jun. 16th, 2012. We define the action as adding a comment to a specific photo. Thus the action space includes all photos on Flickr. To avoid the sparsity problem, we remove those photos with less than 5 comments. This results in 144,627 unique actions. We try to study how users’ commenting actions conform to the other users in the network.
• Gowalla is a location-based social network, where users share their locations by checking-in. The data was from [7] and all checkins of these users over the period of Jul. 10th, 2010 - Jul. 29th, 2010. The action in this data set is defined as whether a user checks in some location (indicated by hashtag or location ID). Thus the dimension of the action space is the number of available locations. We also remove those locations with less than five check-ins and finally obtain 218,811 unique actions. Our goal is to study whether the users will conform to their friends’ check-in behavior.
• Weibo is the most popular microblogging service in China. We collected a complete network between 1,700,000 users and all the tweets posted by those users between Sep. 28th, 2012 and Oct. 29th, 2012. The action is defined as whether a user posts a message on a specific topic (indicated by hashtag). We choose the ten most popular topics in 2012 and study how users conform to each other in the network on discussing those topics. We aim to study how users conform to each other on discussing those topics.
• Co-Author is a network of authors. The data set, crawled by Arnetminer.org [33], is comprised of 737,690 CS authors and 2,416,472 co-authorships over 1975 - 2012. Based on the publication venues, authors are categorized into different domains such as Data Mining, Artificial Intelligence, Computer Graphics, etc.4 The action is defined as whether an author will publish a paper in a specific domain. Thus, in total we have 200 unique actions. Our goal is to study how an author conforms to the other authors on choosing the publication venue.

We apply the learned model for action prediction and evaluate its performance in terms of Precision, Recall, F1-Measure, and Area Under Curve (AUC).

Jie Tang, Sen Wu, and Jimeng Sun. Confluence: Conformity Influence in Large Social Networks. In Proceedings of the Ninteenth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD'13). pp. 347-355. [PDF] [Slides_PPT] [Slides_PDF] [Poster] (Oral Presentation)