Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Selenite: Scaffolding Online Sensemaking with Comprehensive Overviews Elicited from Large Language Models

Michael Xieyang Liu, Tongshuang Wu, Tianying Chen, Franklin Mingzhe Li, Aniket Kittur, Brad A. Myers

TL;DR

Selenite presents an AI-powered sensemaking system that uses LLMs as reasoning engines and knowledge retrievers to generate upfront, comprehensive overviews of information spaces, addressing the cold-start problem. It combines global grounding (common criteria and encountered options) with local grounding (in-situ paragraph and page-level summaries) and dynamic next-step suggestions to guide reading. Through intrinsic evaluation, usability study, and a case study, the work shows that Selenite can produce accurate overviews, accelerate information processing, and improve comprehension and learning, while identifying design implications and limitations for future AI-assisted sensemaking tools. The approach demonstrates a practical pathway for scalable, grounded AI assistance in navigating unfamiliar domains and making informed comparisons. The work highlights the potential of retrieval-augmented LLMs and structured interfaces to scaffold reading, with implications for educational and professional contexts where rapid, informed decision-making is critical.

Abstract

Sensemaking in unfamiliar domains can be challenging, demanding considerable user effort to compare different options with respect to various criteria. Prior research and our formative study found that people would benefit from reading an overview of an information space upfront, including the criteria others previously found useful. However, existing sensemaking tools struggle with the "cold-start" problem -- it not only requires significant input from previous users to generate and share these overviews, but such overviews may also turn out to be biased and incomplete. In this work, we introduce a novel system, Selenite, which leverages Large Language Models (LLMs) as reasoning machines and knowledge retrievers to automatically produce a comprehensive overview of options and criteria to jumpstart users' sensemaking processes. Subsequently, Selenite also adapts as people use it, helping users find, read, and navigate unfamiliar information in a systematic yet personalized manner. Through three studies, we found that Selenite produced accurate and high-quality overviews reliably, significantly accelerated users' information processing, and effectively improved their overall comprehension and sensemaking experience.

Selenite: Scaffolding Online Sensemaking with Comprehensive Overviews Elicited from Large Language Models

TL;DR

Selenite presents an AI-powered sensemaking system that uses LLMs as reasoning engines and knowledge retrievers to generate upfront, comprehensive overviews of information spaces, addressing the cold-start problem. It combines global grounding (common criteria and encountered options) with local grounding (in-situ paragraph and page-level summaries) and dynamic next-step suggestions to guide reading. Through intrinsic evaluation, usability study, and a case study, the work shows that Selenite can produce accurate overviews, accelerate information processing, and improve comprehension and learning, while identifying design implications and limitations for future AI-assisted sensemaking tools. The approach demonstrates a practical pathway for scalable, grounded AI assistance in navigating unfamiliar domains and making informed comparisons. The work highlights the potential of retrieval-augmented LLMs and structured interfaces to scaffold reading, with implications for educational and professional contexts where rapid, informed decision-making is critical.

Abstract

Sensemaking in unfamiliar domains can be challenging, demanding considerable user effort to compare different options with respect to various criteria. Prior research and our formative study found that people would benefit from reading an overview of an information space upfront, including the criteria others previously found useful. However, existing sensemaking tools struggle with the "cold-start" problem -- it not only requires significant input from previous users to generate and share these overviews, but such overviews may also turn out to be biased and incomplete. In this work, we introduce a novel system, Selenite, which leverages Large Language Models (LLMs) as reasoning machines and knowledge retrievers to automatically produce a comprehensive overview of options and criteria to jumpstart users' sensemaking processes. Subsequently, Selenite also adapts as people use it, helping users find, read, and navigate unfamiliar information in a systematic yet personalized manner. Through three studies, we found that Selenite produced accurate and high-quality overviews reliably, significantly accelerated users' information processing, and effectively improved their overall comprehension and sensemaking experience.
Paper Structure (62 sections, 1 equation, 5 figures, 7 tables)

This paper contains 62 sections, 1 equation, 5 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Making Sense of Online Information
  4. Tool Support for Reading and Comprehension
  5. Eliciting Knowledge from LLMs
  6. Formative Study & Design Goals
  7. Formative Study
  8. Methodology
  9. Findings
  10. Summary of Design Goals
  11. The Selenite System
  12. Example Usage Scenario
  13. Detailed Designs
  14. [D1] Providing Global Grounding using Common Criteria and Options Encountered
  15. Automatically recognizing topics.
  16. ...and 47 more sections

Figures (5)

  • Figure 1: The main user interface of Selenite, which provides users with a comprehensive overview of the information space in the sidebar (a). When users encounter an unfamiliar topic (b), Selenite offers them a global overview based on commonly considered criteria (c) as well as the options encountered so far (d), helping them develop quick intuitions of the topic. As users read articles that they haven't seen before, Selenite provides local grounding through page-level and paragraph-level summaries and annotations (e), enabling effective comprehension and efficient navigation between the content of their interests. Before leaving a page, Selenite dynamically summarizes users' progress and suggests avenues for finding additional new information (f) in subsequent searches.
  • Figure 2: Main stages and features of Selenite: After the user 1) searches and finds an initial webpage-of-interest to read, Selenite provides: 2) global grounding with a set of common criteria as well as options encountered so far, 3) local grounding with in-situ annotations of criteria per paragraph, and 4) suggestions on what to search for next to gain new information.
  • Figure 3: Selenite enables structured and efficient navigation by criterion through clicking the "previous/next" (shown as "<" and ">") buttons (a), after which Selenite will automatically scroll the page to reveal the previous/next mentioning of the target criterion.
  • Figure 4: When encountering a particularly convoluted paragraph (e.g., the paragraph on the left) with multiple criteria and options that users can't quite absorb in the first pass, they can click the "Analyze" button (a) and leverage the "zoom in" feature that Selenite offers to query for more comprehensive descriptions that clarify which sentences or phrases pertain to which specific criteria and sentiments. Selenite wraps phrases and sentences in colored boxes, with green denoting "positive" (b), red denoting "negative", and grey denoting "neutral" (not shown).
  • Figure 5: Example comparison article on the topic of "best baby strollers." Note that it only contains approximately $1/4$ of the article, which originally includes content about 10 baby strollers, along with other long-form commentaries. The article can be found online at https://www.babygearlab.com/topics/getting-around/best-stroller.