Research

Published work

Cross-Platform Spillover Effects in Consumption of Viral Content: A Quasi-Experimental Analysis Using Synthetic Controls

Haris Krijestorac, Rajiv Garg, and Vijay Mahajan

Published in Information Systems Research (2020)

https://pubsonline.informs.org/doi/10.1287/isre.2019.0897

Decisions Under the Illusion of Objectivity: Digital Embeddedness and B2B Purchasing

Haris Krijestorac, Rajiv Garg, and Prabhudev Konana

Published in Production and Operations Management (2021)

https://onlinelibrary.wiley.com/doi/10.1111/poms.13363

Working Papers

Personality-Based Content Engineering for Rich Digital Media

Haris Krijestorac, Rajiv Garg, and Maytal Saar-Tsechansky

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3366561

In Progress

To Use or Not to Use Samuel L. Jackson? Understanding the Role of “Voice of Machine” in Consumer Decisions

Haris Krijestorac, Rajiv Garg, and Vijay Mahajan

Abstract:

The rise of voice-driven personal assistants such as Amazon Alexa and Apple Siri has created opportunities to promote products through these channels. In response, firms have attempted to boost sales by investing in celebrity voiceovers, as initiated by Amazon’s commissioning of Samuel L. Jackson for Alexa. Yet, it remains uncertain how beneficial these investments are. Moreover, it is unclear whether such benefits can be attributed to the celebrity themselves, i.e., Jackson, or whether there are inherent acoustic properties of a voice, such as its frequency, volume, or inflections, that make it effective. To help firms and voice AI platforms better hone their voice selection, we investigate the role of different voice types on purchase behavior. Based on over one thousand voice samples from celebrities, google voices, and voice actors, we identify clusters of voices based on eight stochastic acoustic properties of voice, including frequency, amplitude, quality, and tempo. Using generated voice samples for both celebrities and non-celebrities corresponding to each identified voice cluster, we explore the effectiveness of both cluster (e.g., authoritative, seductive) and celebrity status on purchase of hedonic and utilitarian goods.

The Role of Emotional Arcs in Generating User Engagement on Social Media Platforms

Reza Alibakhshi, Haris Krijestorac, and Shirish Srivastava

Abstract:

The rise of "story" features on various social media platforms has made it imperative that content creators craft concise, yet well-designed narratives to attract user attention. Such narratives require creators to think beyond one-dimensional content features (e.g., positive or negative sentiment) and focus more on dynamic features (e.g., emotional highs and lows) that truly stimulate an audience and translate attention into engagement. To explore how to craft such narratives, we examine which emotional properties of Instagram posts lead to greater rates of engagement, as measured by like-to-view ratio. Examining the emotional variability and inertia of five key emotions (happiness, surprise, anger, fear, sadness) of Instagram posts from Fortune 500 firms, we find the corresponding emotional arcs of these videos to be strong predictors of their performance. Moreover, our analysis provides prescriptions for content creators regarding which emotional arcs are most effective. These findings contribute to content engineering literature by showing that for rich media such as stories and videos, empirical insights can be extracted that complement intuition and creativity, helping craft an engaging emotional narrative.

The Streisand Effect: Understanding the Effects of Censorship on Information Diffusion Within and Across Platforms

Haris Krijestorac, Matthew Yeaton, and Sri Kudaravalli

Abstract:

With the Internet’s catalyzation of crowd-driven content of potentially dubious veracity, policy makers and online platforms alike struggle to manage the diffusion of content deemed pernicious. Although the legality of such measures is a commonly debated, the effectiveness of de-platforming measures themselves has arguably inspired less inquiry. Indeed, numerous anecdotes illustrate the potential for information suppression in social networks to backfire, rather than serving their intended purpose. One such example includes Barbara Streisand’s hiding of her Malibu beach house in 2003, which allegedly provoked even greater interest from the public. Extending this analogy to the context of online social networks, our research examines how the network position of a censored individual affects the dissemination of their influence, both within the platform in which they are censored, as well as across alternative platforms. Our analysis combines elusive empirical data on aftereffects of censorship, as well as simulation results based on Agent-Based Modelling. Our results offer insights on whether and when online censorship is truly effective at suppressing information dissemination online.

The Effects of Algorithmic Transparency and Personalization on Information Disclosure: A Cross-Cultural Analysis from a Global Field Experiment

Cathy Yang, Xitong Li, Ai-Ting Goh, and Haris Krijestorac

Abstract:

Given the prevalence of algorithms assisting humans in making decisions, it is imperative that companies understand whether and why individuals are willing to adopt algorithmic advice. With the introduction of data protection policies such as GDPR and the California Privacy Act, companies are required to disclose what information is used by an algorithmic advisor, and for what purpose. Thus, firms would benefit from understanding what information individuals are willing to disclose, as so as to better promote their algorithmic advisors. Our study examines two salient dimensions of the privacy-information tradeoff: personalization, or the extent to which data individual-level data about a user is leverages, and transparency, or the extent to which it is disclosed that such information is used. To explore this tradeoff along these dimensions, we conduct a large-scale global field experiment on Facebook in which we manipulate the transparency and personalization components in the context of promoting admissions to a master’s degree program in a highly-ranked business school. We find that high transparency tends to decrease clickthrough, and that this effect is particularly salient under high personalization. We argue that under conditions of high personalization, users feel more vulnerable to disclosing personalized information, and are thus particularly triggered by high transparency, which reminds them that their information is being used. This unique result is somewhat non-intuitive, as transparency does not have the typical effect of increasing user trust by increasing perceived knowledge over how personal data is used.