Bonsai: Intentional and Personalized Social Media Feeds

TL;DR

The paper addresses the misalignment between engagement-focused social feeds and users' long-term goals by introducing Bonsai, a four-stage framework (Planning, Sourcing, Curating, Ranking) that enables intentional, personalized feeds via natural-language prompts. It combines an LM-driven planning and curation pipeline with a Weighted Borda Count ranking across relevance, recency, and popularity, implemented on Bluesky and designed to be platform-agnostic. In a multi-week, 15-participant field study, Bonsai helped users discover new content and filter out toxicity, while also highlighting the cognitive effort required and the demand for actionable transparency and tighter feedback loops. The findings suggest intentional feedbuilding is feasible and valuable, guiding future designs toward transparent, user-driven administration of feed content and hybrid approaches that balance control with usability and system learnability.

Abstract

Social media feeds use predictive models to maximize engagement, often misaligning how people consume content with how they wish to. We introduce Bonsai, a system that enables people to build personalized and intentional feeds. Bonsai implements a platform-agnostic framework comprising Planning, Sourcing, Curating, and Ranking modules. This framework allows users to express their intent in natural language and exert fine-grained control over a procedurally transparent feed creation process. We evaluated the system with 15 Bluesky users in a two-phase, multi-week study. We find that participants successfully used our system to discover new content, filter out irrelevant or toxic posts, and disentangle engagement from intent, but curating intentional feeds required more effort than they are used to. Simultaneously, users sought system transparency mechanisms to effectively use (and trust) intentional, personalized feeds. Overall, our work highlights intentional feedbuilding as a viable path beyond engagement-based optimization.

Figures (7)

• Figure 1: Positioning of different social media feeds along two dimensions: intentional vs. emergent and generic vs. personalized. These axes are introduced in this paper as an analytic framing and do not reflect platform categories. Twitter's old chronological feed is generic and intentional; Reddit’s "Hot" feed is generic and emergent; TikTok’s "For You" feed is personalized and emergent. Bonsai appears in the remaining quadrant as a system for constructing feeds, not a platform.
• Figure 2: Key components of the Bonsai interface: (A) natural language input to describe feed purpose, (B) source selection interface with suggested feeds, lists, and accounts, (C) preference management panel for including or limiting topics with importance levels, and (D) ranking presets balancing relevance, recency, and popularity.
• Figure 3: Bonsai's generalizable architecture for intentional feeds, composed of four components: Planner (translates natural language intent into an initial configuration), Sourcer (retrieves candidate posts from feeds, accounts, or hashtags), Curator (scores posts against user-stated preferences), and Ranker (orders content by relevance, recency, and popularity, as determined by the user).
• Figure 4: Three representative participant user journeys using Bonsai: low effort, low reward (top row); low effort, medium reward (middle row), high effort, high reward (bottom row). The top two participants (P10 and P7) represent perspectives of people who spent minimal effort to improve their feeds. P11 (bottom) represents a participant who had prior experience using other feed builders, and invested time to optimize their Bonsai feeds and reap high rewards.
• Figure 5: Feed management interfaces in Bonsai. (a) Feed configuration page, where users customize their sources, set inclusion and exclusion prompts, and select ranking styles. (b) Manage feeds page, where users can view, activate, duplicate, and edit multiple feeds from a single account.