For centuries, reading has meant scrolling—page by page, line by line. But what if reading could mean navigating a tree?
TreeReader, a new system from researchers at the University of Toronto and the Vector Institute, challenges the linearity of academic literature. It proposes a reimagined interface: one where large language models (LLMs) summarize each section and paragraph into collapsible nodes in a hierarchical tree, letting readers skim, zoom, and verify with surgical precision. The result is more than a UX tweak—it’s a new cognitive model for how scholars might interact with complex documents in the era of AI.
The Problem with PDFs
Despite their dominance, PDFs are a legacy format designed for printing, not understanding. Academic papers are hierarchical by nature—section > subsection > paragraph > figure—but PDFs flatten this into a long scroll. The result? Cognitive overload, especially in literature reviews or when jumping between topics.
Readers often:
- Miss key ideas buried deep in method sections,
- Waste time parsing familiar background info,
- Struggle to verify LLM-generated summaries that lack clear sourcing.
LLM chatbots (like ChatGPT or Elicit) have helped with summarization, but they lack structure-awareness. You can’t easily tell whether a summary comes from the Introduction or the Results, or drill into a specific paragraph’s evidence.
TreeReader’s Core Idea: Hierarchical Summarization + Navigable UI
TreeReader restructures academic papers into an interactive tree, where every node represents a section, paragraph, table, or figure. Each node shows a concise GPT-4o-generated summary, with optional access to the full text.
Interface Design:
| Panel | Function |
|---|---|
| Left | Navigation tree with expandable nodes |
| Middle | Scrollable column of summaries (or full text on demand) |
| Right | Context: figures, source references, or subsection previews |
The summarization is recursive:
- Paragraphs are summarized individually.
- Section summaries are generated based on child paragraph summaries.
- All summaries include source references, shown on hover.
This addresses three key user frustrations identified through formative interviews:
- Too much low-priority information in long papers.
- Unreliable or overly verbose LLM summaries.
- Lack of transparency in AI-generated content.
Does It Actually Work?
In a controlled study, 5 graduate-level CS researchers used TreeReader and a traditional PDF reader to read two scientific review papers. Each participant skimmed for 5 minutes, then deep-read for 25 minutes.
Results (Figure summaries omitted for brevity):
- Skimming: TreeReader outperformed PDFs in all metrics: grasping structure, identifying key ideas, and understanding the paper’s goals.
- Deep Reading: Mixed results. Some users found TreeReader helped navigate complex arguments; others struggled to locate fine-grained details quickly.
- Cognitive Load: TreeReader reduced reported mental effort, frustration, and perceived difficulty.
One participant said: > “I would definitely use TreeReader every day for my initial literature review.”
However, limitations remain:
- No Ctrl+F search.
- Lack of real-world distractions (like multitasking tabs).
- Small user sample (N=5).
Still, the signal is clear: when it comes to information targeting and structural awareness, TreeReader’s approach is compelling.
Implications: From Reading to Sensemaking
TreeReader doesn’t just summarize papers—it changes how we sense-make. By encouraging exploration through selective expansion and providing traceable key points, it turns passive reading into active interrogation.
This aligns with broader HCI trends:
- Tools like ScholarMate, Sensecape, and IdeaSynth are shifting AI interfaces from automation to scaffolded reasoning.
- TreeReader fits this ethos, treating the LLM not as an answer machine but as a structural explainer and cognitive assistant.
Where TreeReader Fits in the AI Productivity Stack
TreeReader’s design suggests a future where:
- Literature review tools start with tree-based distillation, not abstracts.
- Peer reviewers can triage submissions more efficiently.
- Researchers can trace LLM summaries back to source evidence, improving trust.
It’s not hard to imagine TreeReader-like interfaces becoming the frontend for semantic search engines, automated reviewers, or multi-document synthesis agents.
Final Thought
TreeReader doesn’t eliminate the work of reading. It just makes the hierarchy visible—and that, cognitively speaking, changes everything.
Cognaptus: Automate the Present, Incubate the Future.