When E-Commerce Personalization Systems Show and Tell: Investigating the Relative Persuasive Appeal of Content-Based versus Collaborative Filtering

Abstract

In the e-commerce context, are we persuaded more by a product recommendation that matches our preferences (content filtering) or by one that is endorsed by others like us (collaborative filtering)? We addressed this question by conceptualizing these two filtering types as cues that trigger cognitive heuristics (mental shortcuts), following the heuristic-systematic model in social psychology. In addition, we investigated whether the degree to which the recommendation matches user preferences (or other users’ endorsements) provides an argument for systematic processing, especially for those who need deeper insights into the accuracy of the algorithm, particularly in product categories where quality is subjective. Data from a 2 (algorithm type: content vs. collaborative filtering) x 3 (percentage match: low vs. medium vs. high) x 2 (product category: search vs. experience) + 2 (control: search and experience) between-subjects experiment (N = 469) reveal that for experience products, consumers prefer content-based filtering with higher percentage matches, because it is perceived as offering more transparency. This is especially true for individuals with high need for cognition. For search products, however, collaborative filtering leads to more positive evaluations by triggering the “bandwagon effect.” These findings have implications for theory pertaining to the use of artificial intelligence in strategic communications and design of algorithms for e-commerce recommender systems.

Notes

1 We ran a separate set of analyses treating the manipulation-check item (perceived percentage matching), rather than the manipulated percentage matching, as the independent variable and found the same results (for experience products that have high percentage match, individuals high in NFC prefer recommendations via content-based filtering rather than collaborative filtering, whereas those low in NFC prefer collaborative filtering over content-based filtering). As a result, we decided to retain the use of manipulated percentage matching as the independent variable, as that would afford us causal, rather than simply correlational, inferences.

Additional information

Notes on contributors

Mengqi Liao

Mengqi Liao (MA, Pennsylvania State University) is a PhD student, Donald P. Bellisario College of Communications, Pennsylvania State University.

S. Shyam Sundar

S. Shyam Sundar (PhD, Stanford University) is James P. Jimirro Professor of Media Effects and Co-director of the Media Effects Research Laboratory, Donald P. Bellisario College of Communications, Pennsylvania State University.