“They’ve Over-Emphasized That One Search”: Controlling Unwanted Content on TikTok's For You Page

Research Background and Issues

• What problems or challenges did the authors identify?
  The authors focused on users' dissatisfaction with TikTok's For You Page (FYP) recommendation content, particularly the issue of algorithms persistently recommending content that users have explicitly expressed disinterest in. This phenomenon is referred to as "algorithmic persistence," where users repeatedly attempt to resist certain content, but the recommendation system continues to provide it.

• Why is this issue important?
  This issue is critical to the user experience and the long-term sustainability of recommendation systems. While TikTok's personalized recommendations are often praised for their accuracy, in certain cases, such recommendations can lead to user frustration, mental health concerns, and even prompt users to abandon the platform.

• Research Motivation and Related Work
  This study aims to explore users' folk theories and how they address mismatched algorithmic recommendations. Previous research has examined how users understand algorithms through folk theories, but few studies have focused on users' efforts to resist and interact with algorithms, as well as the specific manifestations of such failed interactions.

Solutions

• What methods or solutions did the authors propose?
  The authors did not propose direct modifications to the algorithm but conducted qualitative interviews to document how users employ strategies to cope with recommended content that does not align with their interests. These strategies include quickly skipping disliked content, using the "filter keywords" feature, reporting or blocking accounts, and actively engaging with preferred content.

• What is innovative about this solution?
  The innovation lies in introducing the concept of "algorithmic persistence," which describes an algorithm's insensitivity to users' rejection responses. This expands existing research on algorithm-user adaptation and provides new considerations for algorithm design.

• What are the implementation steps and key techniques used?

  • Data Collection Methods: Qualitative interviews with current and former TikTok users to discuss their perceptions and coping strategies regarding recommended content.
  • Data Analysis Methods: Grounded theory coding techniques were employed to iteratively refine theories and themes to identify patterns and concepts.
  • Systematic comparison and thematic coding were used to explore the interaction between user behavior and platform response mechanisms.

Research Findings

• What specific findings were obtained?
  The study revealed that algorithms often fail to respond to users' folk theories and behavioral strategies. This phenomenon may be due to technical blind spots in the algorithm or reflect a "rigid" understanding of user interests. Users employed various strategies, such as using the "not interested" button, skipping content, and actively following preferred content, but these efforts were largely ineffective. Additionally, "algorithmic persistence" emerged as a significant factor impacting users' platform experiences.

• What advantages does it have compared to existing solutions?
  The study's uniqueness lies in introducing the concept of failed interactions between users and algorithms within the context of human-computer interaction. This provides new insights for improving recommendation system design. Through qualitative research, the study offers a deeper understanding of how users resist recommended content and fail to achieve desired outcomes.

• What are the experimental or evaluation results?
  Some user strategies (e.g., keyword filtering) proved effective in the short term, but in the long term, these strategies could not completely eliminate unwanted content. Even when users took more extreme measures, such as blocking or reporting, they struggled to fundamentally change their FYP.

• Limitations and Future Directions

  • Limitations: The study's sample size was small, based on only 14 interviews, which may limit the generalizability to all TikTok users or other platforms. Additionally, as TikTok's algorithm evolves, the findings may not apply to future system versions.
  • Future Directions: The authors suggest conducting larger-scale quantitative experiments or algorithm audits to further validate the prevalence and impact of "algorithmic persistence." They also recommend exploring similar phenomena on other social platforms to assess the generalizability of this concept.

By uncovering the constrained user behavior strategies and mismatches with algorithms on TikTok, this study deepens the understanding of recommendation system design and user behavior interaction, while encouraging similar exploratory research on other platforms.