Mapping the Design Space of Teachable Social Media Feed Experiences (2401.14000v2)

Abstract: Social media feeds are deeply personal spaces that reflect individual values and preferences. However, top-down, platform-wide content algorithms can reduce users' sense of agency and fail to account for nuanced experiences and values. Drawing on the paradigm of interactive machine teaching (IMT), an interaction framework for non-expert algorithmic adaptation, we map out a design space for teachable social media feed experiences to empower agential, personalized feed curation. To do so, we conducted a think-aloud study (N=24) featuring four social media platforms -- Instagram, Mastodon, TikTok, and Twitter -- to understand key signals users leveraged to determine the value of a post in their feed. We synthesized users' signals into taxonomies that, when combined with user interviews, inform five design principles that extend IMT into the social media setting. We finally embodied our principles into three feed designs that we present as sensitizing concepts for teachable feed experiences moving forward.

• The paper maps the design space of teachable social media feeds, proposing a framework using Interactive Machine Teaching to empower users to teach algorithms for aligning content curation with their values.
• Through a think-aloud study with 24 participants, the research identified key account-based and content-based signals users employ to evaluate social media content, highlighting a gap with current engagement-focused platforms.
• The study synthesizes findings into five design principles and proposes three novel feed designs (Exploded UI, Multi-Feed Curriculum, Finite Feeds) to enable more teachable and user-aligned social media experiences.

Insights into Teachable Social Media Feed Design

The paper "Mapping the Design Space of Teachable Social Media Feed Experiences" by Feng et al. addresses the nuanced challenge of aligning social media feed curation with user values and preferences. Grounded in the paradigm of Interactive Machine Teaching (IMT), the authors propose a framework aimed at enhancing user agency in social media interactions by providing users with tools to "teach" algorithms how to curate content that reflects their unique values and preferences.

Context and Methodology

Digital feeds on platforms like Instagram, Mastodon, TikTok, and Twitter are algorithmically curated, often reducing individual agency in favor of broad, platform-wide content strategies. The authors argue that this approach leads to a personalization paradox where users feel their individual values are not represented. The paper leverages IMT—a paradigm where non-experts teach machine learning algorithms task-relevant concepts—to explore how social media feeds can be rendered more teachable. The authors conducted a think-aloud paper involving 24 participants, examining the signals users utilize to assign value to social media content. This qualitative approach unearthed the dimensions along which users evaluate posts, resulting in taxonomies of account-based and content-based signals.

Findings

The empirical paper produced several taxonomies that reveal the multi-faceted and nuanced approach users take when evaluating social media content. For account-based features, the original poster's identity was a significant factor, with characteristics such as personal connection, admired traits, and topical alignment bearing importance. Content-based taxonomies were more varied, with topic alignment to personal interests being particularly valued. The participants' feedback highlighted a significant gap between current platforms' engagement-focused curation strategies and users' nuanced evaluative criteria, signalling a need for more granular feedback mechanisms.

Design Principles

The paper synthesizes its findings into five design principles for teachable feed experiences:

1. Situate Teaching Language within the Feed: This principle encourages embedding teaching interactions directly within the feed to reduce disruption in user experience and leverage existing representations as teaching mechanisms.
2. Be Available, but Not Intrusive: The design should balance availability of teaching affordances without overwhelming users, respecting the context-dependent nature of social media interactions.
3. Embrace Feed Multiplicity: Facilitating multiple feeds within a platform allows users to categorize and contextualize content, enhancing the relevance of personalized experiences.
4. Seek Structured and Unstructured Feedback: Integrate mechanisms for capturing both structured (e.g., UI buttons) and unstructured (e.g., natural language) feedback to more accurately reflect user preferences.
5. Enable Teaching and Evaluation at Varying Timescales: Supporting both immediate feedback and long-term evaluation empowers users to continuously refine feed behaviors over time.

Proposed Designs

To practically embody these principles, the authors propose three novel feed designs:

1. Exploded UI Views: This design provides a granular, feature-specific view for content evaluation, allowing users to express preferences about post attributes directly within the feed.
2. Multi-Feed Curriculum Organization and Seeding: Users can organize preferences into curricular folders, with seeds generated from these preferences to streamline the creation of focused, personalized feeds.
3. Purposefully Finite Feeds with Natural Language Feedback: Incorporating finite content stacks allows users to provide reflective, natural language preferences, merging the immediacy of direct manipulation with longer-term evaluative interactions.

Implications and Future Directions

The proposed framework offers the potential to redefine social media interactions by enhancing user agency and aligning content curation more closely with individual values. However, the authors acknowledge ethical concerns such as privacy and hyper-personalization, suggesting that increased transparency and thoughtful moderation mechanisms could mitigate potential risks.

Additionally, while these proposed designs offer a robust starting point, further empirical validation is necessary to evaluate their efficacy in real-world applications. Future work could explore how these designs operate in various media types or translate to beyond social media, such as news or personalized content platforms.

In essence, this paper moves forward the discourse on empowering users through algorithmic transparency and teachable systems, offering practical insights for the development of socially accountable algorithms in digital environments.