Toward Order-of-Magnitude Cascade Prediction

Abstract

Abstract-When a piece of information (microblog, photograph, video, link, etc.) starts to spread in a social network, an important question arises: will it spread to "viral" proportions -where "viral" is defined as an order-of-magnitude increase. However, several previous studies have established that cascade size and frequency are related through a power-law -which leads to a severe imbalance in this classification problem. In this paper, we devise a suite of measurements based on "structural diversity" -the variety of social contexts (communities) in which individuals partaking in a given cascade engage. We demonstrate these measures are able to distinguish viral from non-viral cascades, despite the severe imbalance of the data for this problem. Further, we leverage these measurements as features in a classification approach, successfully predicting microblogs that grow from 50 to 500 reposts with precision of 0.69 and recall of 0.52 for the viral class -despite this class comprising under 2% of samples. This significantly outperforms our baseline approach as well as the current state-of-the-art. Our work also demonstrates how we can tradeoff between precision and recall.

Understanding and forecasting lifecycle events in information cascades

“…In our work, we denote this social network information by an undirected network G D =(V D , E D ) where V D denotes the individuals involved in the historical diffusion process and an edge e ∈ E D denotes that information has been shared between a pair of individuals ignoring the direction of propagation. This is similar to our previous work [1] and will be described in detail in Section 4. We denote the network produced by the participants of a cascade C by…”

“…The diffusion network mentioned in Section 3.2 G D =(V D , E D ) is created by linking any two users who are involved in a microblog reposting action within the period May 1, 2011 and August 31, 2011. Similar to most social networks, this network also exhibits a power law distribution of degree [1]. Table 2 shows the statistics of the diffusion network and the corpus of cascades used in our experimental study.…”

“…For building the diffusion network, we use the dataset provided by WISE 2012 Challenge 1 as has been previously used in [1]. The dataset provides us with user data and the reposting information of each microblog along with the reposting times which enables us to form the cascades for each microblog separately.…”

An empirical evaluation of social influence metrics

“…This group of measurements take into account the various parameters that are part of a microblog cascade. There has been many studies in the area of predicting the cascades including Bakshy et al [25] , Cheng et al [13] and more recently Guo et al [8]. Unlike our study, there hasn't been many attempts to utilize the cascade parameters in predicting retweet behavior.…”

“…See [12] for a review of some of this work. There has also been related work on predicting cascades [13], [14], [8] which are more focused on determining if a trend in social media exceeds a certain size. That said, some of the ideas from these approaches, such as structural diversity [5] are examined here (though this paper is focused on a different problem).…”