Mapping the Design Space of Teachable Social Media Feed Experiences

Title of the Paper

Mapping the Design Space of Teachable Social Media Feed Experiences

Paper Information

• Subject Area: Human-Computer Interaction (HCI), Social Media Algorithms, and User Experience Design
• Keywords: Social Media, Interactive Machine Teaching (IMT), Personalized Recommendations, User Agency, Feed Design, Multimodal Design

Research Background and Issues

• What problems or challenges did the authors identify?

  • The lack of transparency in personalized recommendation algorithms on social media (the "black box" problem) leads to a reduced sense of control over the content in users' feeds.
  • Many platforms design algorithms in a "one-size-fits-all" manner, neglecting cultural differences and personalized needs among users.
  • Current user input mechanisms (e.g., likes, blocking) fail to capture users' complex and nuanced multidimensional content preferences.

• Why is this issue important?

  • Social media has become a critical tool for social interaction and information acquisition in modern society, and its algorithms significantly influence users' information reception, behavior, and social cognition.
  • User agency is a fundamental psychological need, and designing experiences that enhance users' sense of control and personalization can improve satisfaction, trust, and healthy engagement.

• Research Motivation and Related Work

  • Interactive Machine Teaching (IMT) has emerged as a research hotspot in recent years, emphasizing the guidance of algorithmic models (learners) through user input (teachers). However, this framework has yet to be widely applied to social media feed design.
  • Existing research focuses on the transparency of recommendation systems and user behavior modeling, but few studies directly address how to effectively capture user input and integrate it into content recommendations.

Solutions

• What methods or solutions did the authors propose?

  • Introduced the concept of teachable feeds tailored for social media scenarios.
  • Proposed five design principles based on IMT:
    1. Embedded Teaching Language: Enable teaching functionality within the feed interface rather than in an external environment.
    2. Non-Intrusive Design: Teaching functionality should be available at any time without disrupting browsing behavior.
    3. Multi-Feed Experience: Allow users to create multiple topic-specific feeds to organize content.
    4. Support for Structured and Unstructured Feedback: Combine intuitive interactions (e.g., buttons) with natural language input.
    5. Support for Multi-Timescale Teaching and Evaluation: Allow seamless switching between short-term adjustments and long-term optimization.
  • Proposed three actionable design patterns: decomposed UI views, course-based multi-feed management, and constrained feed stacks with natural language feedback.

• What is innovative about this solution?

  • Introduced the findings of Interactive Machine Teaching (IMT) into the task of social media content recommendation, preliminarily defining the relevant design space.
  • Simulated how teaching language can help capture and refine user preferences and guide algorithmic improvements.
  • Proposed multidimensional user experience optimization strategies within the framework of user agency.

• Implementation Steps and Key Techniques

  • Experiment: Conducted a "think-aloud" study with 24 social media users across four platforms (Instagram, Mastodon, TikTok, Twitter) to collect signals on how users evaluate feed content.
  • Data Extraction and Analysis: Classified signals using coding methods and summarized account-based and content-based features and evaluation criteria.
  • Teaching Language Construction: Designed new algorithm teaching interfaces using existing UI components and interaction patterns (e.g., direct clicks and natural language).
  • Practical Validation: Designed and simulated various application scenario interfaces incorporating the proposed teaching language.

Research Outcomes

• What specific results were achieved?

  • Proposed a cross-platform evaluation framework based on user signals, covering account features (e.g., account activity frequency and the user's relationship with the author) and content features (e.g., alignment with interests, information richness, and emotional response).
  • Developed five principles that extend the IMT framework for social media, focusing on user experience, algorithm design, and interaction feedback dimensions.
  • Created three prototype designs based on these principles: decomposed views, course-based multi-feed management, and constrained stacks with natural language feedback functionality.

• What advantages does it have compared to existing solutions?

  • Contemporary social media algorithms rely on implicit user data, which can lead to algorithmic bias or user confusion. In contrast, this design emphasizes explicit user input to optimize recommendations through user guidance.
  • Supports transparency and user customization: the causal link between user preferences and system behavior is demonstrated and reinforced.
  • Provides multi-timescale adjustments, addressing the limitations of traditional "set-and-forget" tools.

• What were the experimental or evaluation results?

  • The experiment revealed significant differences in the signals users employ to evaluate feed content. Overall, "account features" (e.g., personal relationships) and "content features" (e.g., topic consistency) were key evaluation factors.
  • Most users, regardless of platform, exhibited a universal desire for self-causality and a sense of participation.

• Limitations and Future Directions

  • Participant Diversity: The study sample may not fully represent global user groups, necessitating validation across more cultural and platform contexts in the future.
  • Prototype Evaluation: The current three types of teachable feed designs lack user testing and ecological evaluation.
  • Extension Exploration: Further research is needed to apply these designs to video-dominated platforms (e.g., TikTok) or non-social media domains (e.g., news feeds).
  • Collaborative Teaching Potential: Future studies could explore the potential of multi-user shared and collaborative "teaching curricula" for community-based content recommendations.

Through this research, the authors established a new interaction model for social media, providing concrete guidance and initial technical exploration for integrating personalized recommendations with user experience.