Deep Exogenous and Endogenous Influence Combination for Social Chatter Intensity Prediction

Abstract

Modeling user engagement dynamics on social media has compelling applications in market trend analysis, user-persona detection, and political discourse mining. Most existing approaches depend heavily on knowledge of the underlying user network. However, a large number of discussions happen on platforms that either lack any reliable social network (news portal, blogs, Buzzfeed) or reveal only partially the inter-user ties (Reddit, Stackoverflow). Many approaches require observing a discussion for some considerable period before they can make useful predictions. In real-time streaming scenarios, observations incur costs. Lastly, most models do not capture complex interactions between exogenous events (such as news articles published externally) and in-network effects (such as follow-up discussions on Reddit) to determine engagement levels.To address the three limitations noted above, we propose a novel framework, ChatterNet, which, to our knowledge, is the first that can model and predict user engagement without considering the underlying user network. Given streams of timestamped news articles and discussions, the task is to observe the streams for a short period leading up to a time horizon, then predict chatter: the volume of discussions through a specified period after the horizon.ChatterNet processes text from news and discussions using a novel time-evolving recurrent network architecture that captures both temporal properties within news and discussions, as well as influence of news on discussions. We report on extensive experiments using a two-month-long discussion corpus of Reddit, and a contemporaneous corpus of online news articles from the Common Crawl. ChatterNet shows considerable improvements beyond recent state-of-the-art models of engagement prediction. Detailed studies controlling observation and prediction windows, over 43 different subreddits, yield further useful insights.

Social-Media Activity Forecasting with Exogenous Information Signals

“…As shown in previous work [7], [8], online topics of discussion are often correlated with different exogenous signals over time. For example, in the context of the Venezuela political crisis, Horawalavithana et al [7] show that Twitter discussions related to the announcement of Juan Guaidó as interim president were dictated by particular external events and news articles reports while Twitter discussions related to Venezuela's former president, Chávez, were strongly correlated with Reddit activity.…”

“…They found news carry different predictive power based on the nature of the entities under study. Dutta et al [8] predict the volume of Reddit discussions in a future short-time horizon leveraging the text from news and initial set of comments using a recurrent neural network architecture. Shrestha et al [19] used a deep learning model to forecast the number of retweets and mentions of a specific news source on Twitter using the network structure observed in the day before the predictions.…”

“…Our solution takes into account different sources of information exogenous to the social media platform whose activity we model, such as news or datasets of protests and armed conflicts. Moreover, from extensive data analysis [7], [8], we are well aware that online topics are often stimulated by different exogenous signals over time, and no single exogenous source will be reliably mirrored in the activity of one particular topic on a social media platform. Our modeling approach reflects these data-driven observations.…”

Nipping in the Bud: Detection, Diffusion and Mitigation of Hate Speech on Social Media

“…Based on these observations, we proposed a model which, given a tweet and its hate markers (ratio of hateful tweets and hateful retweets on the user's timeline), along with a set of topical and exogenous signals (news title in this case), predicts which followers of the said user are likely to retweet hateful posts. The motive behind using exogenous signals is to incorporate the influence of external events on a user's posting behaviour [Dutta et al 2020]. Interestingly, we observed that existing information diffusion models that do not consider capturing any historical context of a user or incorporate exogenous signals perform comparably on non-hateful cascades but fail to capture diffusion patterns of hateful users.…”

Incomplete Gamma Integrals for Deep Cascade Prediction using Content, Network, and Exogenous Signals

“…Self-exciting point processes [7,8] were also employed as generative models. Recently, exogenous influence has been incorporated [9,10]. Neural methods, particularly graph embedding-based techniques, are quickly becoming popular [11,12].…”