The Impact of In-Feed Recommendation on Consumers' Impulse Buying Behavior on Short Video Platforms

Introduction

With the rapid evolution of algorithm-driven digital platforms (Li, 2024), short video applications such as TikTok and Kuaishou have reshaped the way users consume information and make purchase decisions (Liu & Khong-khai, 2024; Xing et al., 2025). Information flow recommendation—powered by real-time user profiling and content delivery algorithms—has transformed passive content browsing into an immersive, personalized shopping experience (Koç, 2023; Zhang et al., 2025). In this context, impulse buying behavior has become increasingly prevalent and strategically significant (Gao et al., 2022; Li et al., 2023; Ma et al., 2024), prompting both academic and industrial interest in understanding its psychological and technological underpinnings (Kaur & Sharma, 2024). This study aims to investigate how algorithmic recommendations on short video platforms influence consumers' impulsive decision making, offering a novel and comprehensive framework for capturing these dynamics.

Prior studies have approached impulsive buying from various perspectives, including affective engagement (Bai & Bai, 2024; Gong & Jiang, 2023), content characteristics (Shao, 2024), and user motivation (Febrilia et al., 2024). The stimulus–organism–response (SOR) model has been widely adopted to explain media-driven behavioral reactions (Li & Guenier, 2024; Xia et al., 2024), while the elaboration likelihood model (ELM) has provided valuable insights into the dual-route nature of information processing (Yu et al., 2025). More recently, emerging works have begun incorporating algorithmic factors such as perceived personalization and transparency (Mingyang Li et al., 2022; Ram et al., 2024), acknowledging that user responses are increasingly shaped by system-level signals (Braganza, 2025; Dianxia & Yin, 2024). However, three key limitations remain:

• •

  The interaction between cognitive elaboration and emotional immersion is often oversimplified.

• •

  Algorithmic trust is rarely modeled as an integrated and dynamic influence.

• •

  Post-decision outcomes, such as purchase regret, are frequently overlooked.

Addressing these gaps presents several challenges. First, impulsive behavior in algorithmic contexts is not merely a function of emotional arousal but a consequence of complex, real-time information processing. Second, algorithmic cues often serve dual roles—as both enablers of trust and triggers of skepticism—which require fine-grained modeling. Third, impulse behavior is inherently multi-dimensional, involving cognitive intention, behavioral execution, and post-action evaluation, yet few models explicitly disentangle these stages. These challenges call for a framework that can integrate psychological, content-based, and algorithmic perspectives in a cohesive and testable structure.

To tackle these challenges, we propose a new model—SOR-ELM, which builds on the SOR paradigm by incorporating dual-route elaboration from ELM, emotional flow experience, and algorithmic perception. The core premise of our method is that short video-induced impulsivity emerges from the interaction of three dimensions: content signal strength (credibility, diagnosticity); user processing mechanisms (central and peripheral elaboration, flow) and algorithmic perception (personalization, transparency, trust). Moreover, the model uniquely incorporates post-purchase regret as a behavioral feedback loop, completing the decision–response–evaluation cycle.