Lill / Spann Influence of Recommender Systems on Consumer Behavior

1. Auflage 2025
ISBN: 978-3-7693-8646-2
Verlag: BoD - Books on Demand
Format: EPUB
Kopierschutz: 6 - ePub Watermark

E-Book, Englisch, Band 17, 190 Seiten

Reihe: Electronic Commerce & Digital Markets

In the dynamic landscape of digital platforms, recommender systems silently guide our decisions, shaping what we watch, read, and buy. But how do these algorithms influence our behavior beyond convenience and personalization? This dissertation delves into the psychological undercurrents of recommender systems, uncovering how subtle design choices trigger behavioral biases that affect consumer preferences, beliefs, and decisions. Through a series of rigorous field experiments, this work explores three critical dimensions of recommender systems: item selection, ranking, and recommendation design. It reveals how phenomena like assimilation and contrast effects, as well as visual cues such as product badges, can subtly yet powerfully steer consumer behavior. Challenging traditional recommender system design priorities, the findings highlight that similarity can sometimes outweigh diversity and that strategic positioning can defy conventional assumptions about user attention. Combining a systematic literature review with empirical evidence from real-world settings, this dissertation offers insights into the interplay between technology and human cognition. It is an essential read for scholars, practitioners, and platform designers seeking to understand - and ethically harness - the behavioral dynamics of recommender systems.

Autoren/Hrsg.

Lill, Markus

Spann, Martin

Leseproben

Article 1:
Behavioral Biases in Recommender Systems - A Systematic Literature Review1

Markus Lill

Behavioral biases are pivotal in recommender systems (RSs) in shaping user interactions and outcomes. Understanding these biases is essential for improving the effectiveness and fairness of RSs. This systematic literature review comprehensively analyzes and categorizes biases within three fundamental dimensions: consumer preferences, beliefs, and decision-making. By analyzing various scholarly articles, this study identifies the prevalence and impact of behavioral biases in the context of RSs. It provides an up-to-date overview and categorization of current research articles analyzing behavioral biases in RSs. Among the dimensions of product quantity and selection, ranking and explanation and reviews of RSs, we explore how the identified biases influence user preferences, distort belief formation, and affect decision-making processes. Furthermore, we outline a research agenda, highlighting potential avenues for future research. This work provides an overview for researchers and practitioners to understand the nuanced interplay between human cognition and technological systems, contributing to understanding how RSs influence consumer behavior.

Keywords: Recommender systems, behavioral biases, consumer decision making, literature review 1

1 Introduction

The proliferation of digital technologies has significantly altered the landscape of consumer interaction. Recommender systems (RSs) have emerged as a fundamental tool in this digital ecosystem, designed to enhance user experience by offering personalized product suggestions (Resnick and Varian 1997). These systems are driven by complex algorithms and vast datasets, enabling them to predict and adapt to individual preferences with increasing accuracy.

Since their inception, RSs have been adopted across various online platforms, including retail websites, streaming services, and social media (Aivazoglou et al. 2020, Schedl et al. 2021). Their primary function is to filter through vast content catalogs and present users with the most relevant options, facilitating decision-making processes and improving user satisfaction (Swaminathan 2003). However, as these systems have grown more sophisticated, they have also introduced challenges and complexities.

One critical area of concern is the influence of behavioral biases on the development and utilization of RSs. Behavioral biases are systematic deviations from rational judgment that influence how users perceive and interact with recommendations (Piramuthu et al. 2012, Teppan and Zanker 2015).

While the technological complexity of RSs continues to evolve, there is a growing recognition in both research and practice that behavioral biases can influence the effectiveness of these systems (Y. Wang et al. 2023). These biases can significantly impact the quality of recommendations, leading to outcomes that may not fully align with user preferences or needs. From a consumer perspective, biases can lead to suboptimal decisions. For instance, users might rely too heavily on the first few recommendations they see (Collins et al. 2018) or select items that confirm their existing beliefs (Schwind et al. 2012) rather than exploring a broader range of options that might better suit their needs.

Furthermore, since user decisions are fed back into the RS, these biases can, in turn, lead to biases in the calculation of new recommendations (Zheng et al. 2023). This feedback loop can augment the problem, potentially resulting in the underrepresentation of some items. This phenomenon is part of a broader concept known as RS fairness (Yalcin and Bilge 2022), where certain products or content may be systematically disadvantaged, affecting the RS’s diversity. Understanding and addressing these issues is crucial for developing more effective and fair RSs.

This review aims to provide a comprehensive overview of identified biases within RSs from research domains in artificial intelligence, computer science, information systems, management, marketing and psychology. By systematically categorizing and analyzing these biases, we can offer valuable insights into their impact and prevalence. This detailed overview will help researchers and practitioners recognize and address the specific biases that affect RS designs. Furthermore, presenting an underlying concept of behavioral biases and clustering them into defined schemes can significantly enhance the understanding of future RS development. This structured approach not only aids in identifying the root causes of biases but also promotes the creation of more bias-aware RSs that cover multiple objectives rather than pure accuracy-optimized recommendations (Zaizi et al. 2023).

The remainder of this systematic literature review is structured as follows: Section 2 provides an overview of RS dimensions and outlines the concept of behavioral bias categorization according to DellaVigna (2009). Section 3 describes our methodological approach for identifying relevant RS literature. Section 4 summarizes the descriptive and main findings of the systematic literature review, structured according to the identified behavioral biases: nonstandard preferences, nonstandard beliefs, and nonstandard decision-making. In Section 5, we propose an agenda for future research. We conclude the review in Section 6.

2 Theoretical Foundation

To establish a foundation for categorizing behavioral biases and the influential dimensions of RSs. we present a comprehensive framework for bias categorization and an in-depth conceptualization of the various RS dimensions. We aim to systematically categorize the different types of behavioral biases and outline the specific characteristics of RSs that contribute to the emergence and reinforcement of these biases.

• Figure 1. Influence of RS Dimensions on Behavioral Biases

This systematic literature review focuses on the RS dimensions and the behavioral biases they induce (cf. Figure 1). Specifically, we analyze how various RS dimensions can potentially lead to specific behavioral biases.

2.1 Recommender System Dimensions

This section analyzes three dimensions that significantly impact the performance and user acceptance of RSs: the quantity and selection of recommended products, the ranking of these recommendations, and the explanations and consumer reviews provided to users. By examining these dimensions, we aim to provide a comprehensive overview of the elements that contribute to the influence of RSs on consumer behavior, offering insights into how these systems can potentially distort user behavior and lead to behavioral biases. See Figure 2 for an illustration of an RS implemented as a horizontal list on the bottom of a product detail page (PDP).

• Figure 2. PDP With a List of Recommendations on the Bottom of the Page

We chose these dimensions because they are fundamental to the user experience and directly influence the effectiveness of RSs. While other dimensions, such as users’ data privacy concerns (Gulsoy et al. 2025). also play significant roles, they fall outside the scope of our current analysis. We focus on these selected dimensions because they represent the primary factors that are directly influenced by the RS design.

2.1.1 Item Quantity and Selection

RSs typically reduce the amount of products available to a more manageable subset, streamlining the consumer’s decision-making process (Adomavicius and Tuzhilin 2005). This reduction involves filtering extensive product catalogs to a reduced selection of items. However, the RS objective is not to minimize the number of items to the smallest possible selection. It is also essential to retain a meaningful number of items for consumers to discover. The term “meaningful” (or sometimes “relevant”) does not always refer to accuracy alone but can encompass other characteristics as well, like product catalog coverage or serendipity (Ge et al. 2010). Hence, no universal heuristic prescribes a fixed number of recommended items for all RS algorithms. Instead, the appropriate number of recommendations depends on the product catalog size, the specific algorithm, and the composition of product characteristics it considers.

However, RSs should filter the product catalog to a set of at least consumption-relevant items. Since the concept of relevancy is broad and includes various characteristics beyond accuracy (Kaminskas and Bridge 2016), other RS metrics have gained attention. For instance, novelty is crucial for discovering the latest items (L. Zhang 2013), particularly in news recommender systems. Diversity broadens the scope of user consumption, potentially opening new categories that can lead to higher profits (Smyth and McClave 2001). Serendipity enhances the entertainment and discovery aspects of RSs, helping to retain high engagement rates among consumers (Kotkov et al. 2016). Therefore, the characteristics of selecting items resulting from RSs vary and significantly influence how consumers interact with them.

2.1.2 Item Ranking

A fundamental task of RS algorithms is to calculate a metric for each product, reflecting the user’s tendency toward item consumption. This can be achieved through various methods, such as estimating consumption probabilities, rating predictions, or assessing the similarity to previously consumed items (Aggarwal 2016).

While it might seem straightforward to rank items in descending order of their...