Source-linked AI summary
CrossWeave: Bridging Perspectives Across Online Communities with a Dual-Pane Design
Fei Fang, Reva Hirave, William Jurayj, Yuqi Li, Brian Lu, Tarik Metin, Tsugunobu Miyake, Kateryna Morhun, Yash Permalla, Kenan Rustamov, Allen Shen, Haojun Shi, Prabhav Singh, Xiheng Tom Wang, Kevin Xu, Qingcheng Zeng, Jiayi Zhang, Daniel Khashabi, Andrew Perrin, Tiziano Piccardi, Ziang Xiao, Jason Eisner
TL;DR
Social media discussions often remain within disjoint threads, amid concerns about division and misinformation. CrossWeave augments the feed with diverse related posts and ways to explore other threads. The paper presents this as a design effort for broader understanding and more constructive participation, while noting technical challenges for live use.
Problem
Method
Results
The live demo poses questions about trust and usability, perspective broadening, informed positions, and constructive dialogue rather than reporting answers.
Source: §4 · p. 4Takeaways & Limitations
CrossWeave is presented as an effort to connect isolated threads to broader viewpoints, arguments, and users, in the hope of supporting mature understanding and more constructive participation.
Source: §6 · p. 5The demo’s quadratic, offline side-pane construction cannot support live traffic efficiently, and vector-based retrieval may struggle especially with relevant non-replies.
Source: §5 · p. 4
Abstract
from arXiv · showhide
Social media systems typically display conversations among already familiar contributors, which can be predictable and one-sided. In civic discourse, this design narrows discussion, reinforces divides, and distorts the perception of public opinion. To encourage cross-community engagement, we present CrossWeave, an AI-powered bridging system that augments the standard social media feed. As the user reads a post, CrossWeave surfaces diverse relevant posts from other threads in a side pane and highlights the connections. Users are invited to venture out of their echo chamber, explore a broader range of views and arguments, and ``click across'' to engage with their authors. When they do, CrossWeave facilitates constructive posting, not only by showcasing relevant past content but also by simulating possible reactions as the user drafts a post.
1 Introduction
CrossWeave addresses concerns that social media contributes to division, incivility, and misinformation while remaining popular and likely to persist. It augments the feed with a curated side pane of diverse relevant posts and highlighted connections, aiming to broaden users’ information diets.
1 Introduction
- Social media is widely viewed as contributing to political division, less civil discourse, and greater vulnerability to misinformation, while remaining popular and likely to stay.The authors frame mitigation as a challenge that must also appeal to users and companies.Source: §1 · p. 1
- CrossWeave adds a curated side pane of relevant posts from across platforms without intervening in the main feed.LLMs make automatic identification of diverse relevant content and highlighted connections possible; the paper focuses on the UI concept and prosocial goals.Source: §1 · p. 1
- Figure 1 shows related posts, highlighted connections, and real-time draft feedback with simulated community reactions.Hovering links a side-pane span to the focused post, while contribution types are color-coded.Source: Figure 1 · p. 2
2 Design Principles
CrossWeave is designed to broaden discussion without replacing familiar social media conversations. Its side pane exposes users to other views and communities, supports exploration and engagement, and is intended to reward useful contributions.
2 Design Principles
- Social media organizes discussions into disjoint threads with different readerships, a pattern the authors associate with echo chambers and limited discussion range and quality.CrossWeave augments rather than replaces this familiar design because existing social circles serve user needs, and users may not seek new information or conversation partners.Source: §2 · p. 2
- A side pane is intended to make users peripherally aware of other views and communities and draw them into exploration and engagement.The authors frame it as an addition to users’ existing social media display.Source: §2 · pp. 2–3
- CrossWeave aims for both informational broadening and social bridging by showing additional perspectives and letting users jump to original threads to reply.The pane is a conduit rather than a referee: its posts are human-authored, and it is customized to the current context rather than personalized to a user.Source: §2 · p. 3
- CrossWeave aims to strengthen incentives for useful posts by allowing a rebuttal to misinformation to appear beside similar claims in other threads.The authors note that users ideally would be notified and commended when their work is shown elsewhere; that feature is on their to-do list.Source: §2 · p. 3
3 Design Elements
CrossWeave pairs a focused conversation with a ranked side pane of related posts, curated for relevance, added value, and diversity. Its reading and posting features help users explore connections, navigate contributions, and consider how a draft may be received.
3 Design Elements
- The interface keeps a focused post and its conversation in the main pane while a permanent side pane lists related posts that users can open in their original threads.The focused post is the one most recently clicked or scrolled to.Source: §3 · p. 3
3.1 Side Pane Curation
- Side-pane posts are ranked to balance relevance, added value, and diversity, so they address a focused post while adding non-redundant evidence, arguments, counterarguments, solutions, or reframings.The authors suggest that successive posts should add new topics, comments, and contribution types rather than repeat familiar talking points.Source: §3.1 · p. 3
3.2 Enhanced Reading Interface
- Highlighted spans show why a side post relates to the focused post, and hovering links corresponding spans and their different comments on the same topic.Highlights are color-coded by contribution type; users can also hover over or filter around topics in the focused post.Source: §3.2 · p. 3
- Users can filter posts by contribution type, drill down into a topic block, or use the same reading features to survey diverse direct replies.Unusual-view ranking risks misleading users about prevalence; the authors suggest favoring popular views when other factors are equal and using prevalence cues such as tooltips.Source: §3.2 · pp. 3–4
3.3 Enhanced Posting Interface
- While drafting, users can draw on related posts, receive real-time AI feedback, and preview three AI-simulated community replies before submission is enabled.The feedback acknowledges draft contributions and suggests ways to strengthen their articulation; the preview is meant to prompt consideration of how the post might be perceived.Source: §3.3 · p. 4
4 Data and Goals for the Live Demo
The live demo focuses on U.S. news conversations and asks users to consider whether CrossWeave supports trustworthy, broadening, and constructive civic discourse. Its questions concern usability, perspective, defensible positions, and dialogue with political opponents.
4 Data and Goals for the Live Demo
- The demonstration focuses on U.S. news and includes over 300 reader comments harvested from Reddit, Fox News, and The New York Times.Source: §4 · p. 4
- The demo invites users to assess whether CrossWeave feels trustworthy and usable and whether the side pane genuinely broadens their perspectives.Source: §4 · p. 4
- The demo asks whether the side pane and posting feedback could support informed, defensible positions and constructive dialogue with political opponents.Source: §4 · p. 4
5 AI Challenges
CrossWeave’s AI challenges center on retrieving broadening content fairly and efficiently: relevance, added value, and diversity require close judgment, while the current implementation is offline and quadratically expensive.
5 AI Challenges
- The demo classifies post contributions and scores relevance and added value, then uses maximal marginal relevance to penalize redundancy in subtopic and contribution type.A secondary pipeline induces subtopic labels for depth-first drilldown.Source: §5 · p. 4
- The current demo is quadratically expensive and offline, so live traffic would require efficient on-demand side-pane construction.A proposed workflow preprocesses reusable posts, indexes their key points, then reranks retrieved posts using a fast scoring function that considers diversity.Source: §5 · p. 4
- Assessing relevance, added value, and distinctness requires close reading and structured evaluation, yet remains subjective, value-laden, and potentially gameable.The authors identify fair assessment as important for avoiding inadvertent viewpoint suppression and loss of user trust.Source: §5 · p. 4
- Finding non-replies that challenge an implicitly assumed standpoint seems especially difficult for vector retrieval and reranking, even when an LLM could recognize their relevance in context.Source: §5 · p. 4
6 Conclusion
CrossWeave reimagines social media with AI-enabled prosocial features, aiming to connect isolated threads to a broader landscape of viewpoints, arguments, and users. The design hopes to support more mature understanding and more constructive participation.
6 Conclusion
- CrossWeave aims to connect isolated social media threads with a broader landscape of viewpoints, arguments, and users to support more mature understanding and constructive participation.Source: §6 · p. 5