Start Date
12-17-2013
Description
Social network-based prediction, more specifically targeting friends and contacts of existing customers, has proven successful in various domains like retail banking, telecommunications, and online advertising. However, little is known about for what types of product categories and brands social network-based marketing is especially effective at predicting brand engagement, both in absolute terms and compared to demographic targeting or collaborative filtering. In this work, we compare the performance of a social network-based recommendation engine against a product network-based recommendation engine of the kind used in collaborative filtering. We do so over 700 brands and 223,000 consumers a novel data set collected from Twitter. We compare the performance of the two approaches by product and user features. Preliminary results indicate that the variance in performance within and across methods is related to differences in brand and user popularity as well as brand audience.
Recommended Citation
Hill, Shawndra; Benton, Adrian; and den Bulte, Christophe Van, "WHEN DOES SOCIAL NETWORK-BASED PREDICTION WORK? A LARGE SCALE ANALYSIS OF BRAND AND TV AUDIENCE ENGAGEMENT BY TWITTER USERS" (2013). ICIS 2013 Proceedings. 16.
https://aisel.aisnet.org/icis2013/proceedings/KnowledgeManagement/16
WHEN DOES SOCIAL NETWORK-BASED PREDICTION WORK? A LARGE SCALE ANALYSIS OF BRAND AND TV AUDIENCE ENGAGEMENT BY TWITTER USERS
Social network-based prediction, more specifically targeting friends and contacts of existing customers, has proven successful in various domains like retail banking, telecommunications, and online advertising. However, little is known about for what types of product categories and brands social network-based marketing is especially effective at predicting brand engagement, both in absolute terms and compared to demographic targeting or collaborative filtering. In this work, we compare the performance of a social network-based recommendation engine against a product network-based recommendation engine of the kind used in collaborative filtering. We do so over 700 brands and 223,000 consumers a novel data set collected from Twitter. We compare the performance of the two approaches by product and user features. Preliminary results indicate that the variance in performance within and across methods is related to differences in brand and user popularity as well as brand audience.