Social Knowledge Graphs

Knowledge doesn’t emerge in isolation. Every idea builds on something else. It’s shaped by discourse, validated by interaction, and made legible through repetition, critique, and contrast. A Social Knowledge Graph reflects this by linking both information and interpretation—ideas, and the people negotiating them.

We treat engagement — likes, replies, shares — not as popularity signals, but as interpretive moves. When someone shares an article and adds a quote or comment, they’re not just amplifying content — they’re framing it. Those frames matter. They turn raw information into situated knowledge.

This framework builds on ideas from learning theory and cognitive science, particularly the concept of the “learning zone” or “zone of proximal development.” It also echoes political communication models like the Overton window, which describes the range of ideas the public is willing to consider at any given time.

We structure information discovery across a conceptual gradient:

An effective system nudges people along this path — offering just enough unfamiliarity to learn, without straying into irrelevance. The Social Knowledge Graph supports that journey by surfacing content within a user’s cognitive range.

Whether a user starts from a person, a post, a hashtag, or a headline doesn’t matter. What matters is the system’s ability to follow the thread — through content, context, and community.

A Social Knowledge Graph should be source-agnostic—able to follow any relevant thread regardless of entry point:

Rather than separate people and content, the graph interweaves them — highlighting the mutual construction of knowledge through discourse, reaction, and resonance.

At the core of a Social Knowledge Graph are the relationships that animate it — not just between nodes, but between perspectives. These relationships fall into three overlapping dimensions that interact to shape the flow of meaning:

Social → Social

This layer models interpersonal structure. It includes direct relationships like followers and friends, but also inferred associations through conversation, collaboration, or shared community involvement. These links help surface how knowledge travels socially.

Social → Content

This is where engagement becomes data. A user replying to, quoting, or annotating a piece of content is more than just a signal — they’re shaping its context. By modeling these interactions explicitly, we trace how content is interpreted, reshaped, and recirculated through the network.

Content → Content

Not all connections are social. Some content relates semantically, temporally, or contextually to other content. This includes explicit references (citations, links), as well as inferred ones (topic similarity, co-mentions). These relationships form the graph’s semantic backbone.

What makes the Social Knowledge Graph unique is how these dimensions intersect. A single quote tweet, for example, might link two users, reframe a piece of content, and introduce a new topic — all in one move. The graph tracks those intersections, mapping not just who’s talking, but how the discourse evolves.

A graph isn’t static. Its structure reflects not just what exists, but what matters—now. That’s why growth and pruning aren’t separate processes, but two sides of temporal relevance.

We grow the graph by ingesting new material — not only tweets or posts, but the content behind them. We parse linked articles, extract key sections, and annotate them semantically. This ensures the graph captures substance, not just signals. A link to an article becomes not just an edge, but a doorway to deeper structure.

But signals evolve. That’s where pruning comes in.

We evaluate signal strength based on how relevance shifts over time — not by volume alone, but by how and where attention propagates. Relevance can be:

Rather than treat information as static, we treat the graph as a living ecosystem. It updates as discourse evolves. We prune stale or isolated nodes while strengthening connections that continue to carry interpretive weight.

In this way, the graph becomes less of a data warehouse — and more of a relevance engine.

To support semantic traversal and personalized discovery, we introduced two node types: Trail and Crumb.

• Trail nodes represent thematic paths through the graph — clusters of pieces, concepts, and interactions that reflect emergent topics or user-specific journeys.
• Crumb nodes log user interaction with specific parts of the graph — like a highlight, a visited concept, or a referenced entity. These Crumbs make user paths legible, allowing us to reconstruct and visualize interest flows.

What makes this structure useful is its ability to not just track exploration, but to enable sharing. When a user explores a concept, highlights sections, or jumps between ideas, they are leaving behind a meaningful trail. And those trails can be picked up by others.

This makes Trails more than just an interface — they become shareable maps of thought. You can share not just a link, but a path: “Here’s how I came to understand this.”

This is one of the ways Trails attempted to realize the foundational ideas behind Metasphere—a spatial, decentralized framework for navigating knowledge.

Metasphere proposed moving beyond the limits of hierarchical categorization toward something closer to world-building: spatializing concepts as landforms, where density, distance, and connection reflect the semantic terrain of our interests. Trails builds on that vision, structuring knowledge not in folders or files, but as navigable, lived experiences—grounded in exploration and shaped by interpretation.

Example: a miniature Social Knowledge Graph

Below is a small, concrete example that combines content granularity (Piece → Chunk → Span) with social context (TwitterAccount), and then layers in interpretation and navigation (Crumb + Trail).

This is intentionally small, but it shows the core idea: knowledge isn’t just “content nodes and edges.” It’s content plus the social and interpretive moves that give it meaning.

In practice, the graph gets richer as you add more people, more pieces, and more cross-links between concepts. But the primitives stay the same: granular content structure, canonical entities/concepts, and interaction traces that can be replayed and shared.

The full graph schema spans from top-level publications (what we call Pieces) to atomic mentions (Spans) and conceptual overlays (Concepts and Trails). Here’s how the layers break down:

This hierarchy allows us to:

• Break down longform content without flattening it
• Track which parts of a source are actually cited or shared
• Build recommendations not just by keyword match but by shared conceptual footprint

External Link Ingestion

A large part of our content curation effort goes into reading what people reference. When someone shares a URL, we fetch the article, extract the structure, identify important passages, and connect them back into the graph. This ensures that the graph represents substance, not just surface.

Personalized Trails

We also log user interactions with specific entities, concepts, and content fragments using Crumb nodes. These allow us to form personalized “trails” through the graph based on what someone actually explores, not just what they click. It’s a feedback mechanism that respects agency while generating rich metadata.

By layering these components — from raw fragments to social overlays — we build systems that make information not only explorable, but also context-aware and adaptive over time.

Most recommendation systems treat engagement as a shortcut to relevance. We treat it as a window into interpretation.

The Social Knowledge Graph isn’t just about seeing who said what. It’s about seeing how knowledge flows—who reshaped what, who connected which ideas, and how those ideas traveled. It’s a way of turning social media into something closer to an archive, or a map — not just a feed.

A Social Knowledge Graph isn’t just a data model — it’s a way of seeing. It pushes us to treat interpretation as data, and relationships as more than links. Meaning is never static. It’s relational, social, and always in motion.

When you design knowledge systems, this model invites you to shift your focus — from what’s popular to what’s meaningful, from who has reach to who carries resonance. It helps you ask better questions:

• What shapes the context of this idea?
• How are different communities assigning meaning?
• Which interactions deepen understanding, and which flatten it?

Graphs encode structure, but also values. The more intentionally we build them, the more powerful they become as tools — not just for navigation, but for reflection.