Work in Progress

Using Social Network Structures to Predict User Engagement: The Importance of Social Ties in Influencer Marketing (with Remi Daviet, UW Madison and Seungbae Kim, UCLA)

As brands increasingly rely on a larger number of smaller social media influencers for marketing campaigns, predicting their ability to generate engagement (such as likes and comments) becomes essential. In this paper, we argue that the social interaction network of influencers among their peers is a more informative measure for predicting their future ability to generate user engagement than commonly used variables such as the number of followers, post content characteristics, and past engagement. Primarily relying on Instagram data, we perform a deep-learning analysis on the social interaction network of influencers and show that the network structure alone explains a large share of the variations in future engagement, with an out-of-sample predictive accuracy of R2=0.73 for log-likes and R2=0.57 for log-comments, largely outperforming predictions based on any other set of variables. Adding content characteristics, influencer features, and past engagement statistics does not increase the predictive performance substantially. We also show that high-performing influencers form elite sub-communities that may not be central to the larger social network. By documenting the value of social network-based analysis over traditional marketing metrics, our study contributes to the emergent literature on the importance of social ties in the digital environment.

Modeling Co-Engagement Patterns in Brand Information Networks (with Yaxin Cui, Northwestern University and Keran Zhao, University of Houston)

The rise in electronic interactions has made information networks ubiquitous. Correspondingly, research across multiple domains has begun to explore the social and economic value of information networks for business decision-making. While most existing research focuses on descriptive and predictive properties of information networks, statistical analysis of the “generative features” of information networks has largely been overlooked. The objective of our study is to create large-scale brand information networks, from common followership data on Twitter, and to model the generative features of the observed network structures. We propose to employ Exponential Random Graph Models to reveal a mix of network and individual level brand characteristics responsible for the formation of links between brands. Since links between brands arise from the aggregated interest patterns of Twitter users, the ERGM model essentially reveals brand and network characteristics associated with high user co-engagement patterns on social media.

Consequences of Brands Staying Silent in the Wake of Sociopolitical Injustice (with Nooshin Warren, University of Arizona and Yashoda Bhagwat, Texas Christian University)

Managers are reluctant to take sides on hot-button sociopolitical debates, either to stay apolitical, or to avoid alienating opposing stakeholders and risk averse investors. However, nearly 60% of Americans believe it is unacceptable for firms to remain silent on sociopolitical issues. This disconnect between managers and customers challenges conventional wisdom, which dictates that by remaining silent, firms, can “opt out of the conversation” and avoid negative financial consequences. However, in the wake of a sociopolitical injustice, silence may communicate information about the firm’s values and still elicit backlash. For example, consumers were critical of Coca-Cola for not speaking against a voting bill in Georgia perceived to be restrictive. The authors address this conflict by answering the questions: (1) When do consumers notice a firm’s silence on contentious sociopolitical issues? (2) How do consumers react to silent firms in terms of their emotions? and (3) What are the financial consequences of staying silent in terms of brand equity, future sales and firm value? The authors first scrape the Twitter verse to extract the online chatter about a brand’s silence towards a sociopolitical issue. Then they use text analysis and machine learning to categorize consumer sentiment towards the silence. Finally, they investigate the downstream effect of consumer sentiment towards silence on brand and firm value. The findings will provide unique insights to managers trying to navigate an increasingly charged political climate by helping them understand the consequences of choosing not to engage in activism.

Quantifying the creativity of digital platform posts based on temporal evolution of content similarities (with Ruyu Chen, Stanford and Jeremy Yang, Harvard University)

This project proposes an algorithm to quantify the creativity of posts on digital platforms. We use the temporal evolution of content similarities to assign a creativity score for each post in a network, where a post is considered to be creative if it appears as atypical among all prior and contemporaneous posts and becomes typical among all subsequent posts. Leveraging a large-scale Instagram post dataset that includes 1,601,074 Instagram posts by38,113 Instagram influencers with 2,273,578 brand-mentions published between 2013 and2019, we empirically test the algorithm based on deep learning image embeddings. We also propose a few research topics on how creativity affects the performance of influencers (such as sales, engagement, growth, etc.) and the factors that may influence the creativity of creators (such as the network position and network cascading, etc.) for future research.

Social Media and Demand Sensing for Innovative Products/Services: The Case of Movie Box Office Revenue (with Mei Li, University of Oklahoma and Naveen Kumar, University of Oklahoma)

Network based recommender systems for sponsored advertising on Instagram (with Bindan Zhang, Northwestern University and Jeremy Yang, Harvard University

%d bloggers like this: