Qlarify: Recursively Expandable Abstracts for Directed Information Retrieval over Scientific Papers

TL;DR

Qlarify introduces recursively expandable abstracts to bridge the gap between scientific abstracts and full-text papers. It uses retrieval-augmented generation and a mixed-initiative interface to iteratively expand abstracts with concise, attributed information from full texts, guided by AI-suggested and static questions. Across an interview study (N=9), a field deployment (N=275), and a comparative evaluation (N=12), participants showed improved information foraging, deeper exploration, and higher satisfaction compared to traditional abstract reading or plain QA. The approach demonstrates promising implications for low-effort, just-in-time scholarly reading, while acknowledging challenges around factual accuracy, interface distraction, and potential over-reliance on AI-generated expansions.

Abstract

Navigating the vast scientific literature often starts with browsing a paper's abstract. However, when a reader seeks additional information, not present in the abstract, they face a costly cognitive chasm during their dive into the full text. To bridge this gap, we introduce recursively expandable abstracts, a novel interaction paradigm that dynamically expands abstracts by progressively incorporating additional information from the papers' full text. This lightweight interaction allows scholars to specify their information needs by quickly brushing over the abstract or selecting AI-suggested expandable entities. Relevant information is synthesized using a retrieval-augmented generation approach, presented as a fluid, threaded expansion of the abstract, and made efficiently verifiable via attribution to relevant source-passages in the paper. Through a series of user studies, we demonstrate the utility of recursively expandable abstracts and identify future opportunities to support low-effort and just-in-time exploration of long-form information contexts through LLM-powered interactions.

Figures (9)

• Figure 1: Recursively expandable abstracts allows users to retrieve clarifying information from a broader expansion context (e.g., the full paper) in response to dynamic information needs, forming expansions that grow the abstract fluidly.
• Figure 2: The design space for an expandable abstracts interaction paradigm. Alternatives we explored in Qlarify are highlighted in gray, and those included in the final system are outlined in red.
• Figure 3: Recursively expandable paper abstracts with attribution in Qlarify. Expansions are created on-demand by highlighting text in the abstract or selecting an AI-suggested expandable entity (A), revealing a question palette (B). Selecting a question in the palette prompts an LLM to retrieve relevant clarifying information, presented as a fluid expansion within the abstract (C). Users can drill-down to see evidence for a response in a paper excerpt (D) and within the full paper context itself (E).
• Figure 4: Qlarify's system architecture. Each paper is first processed (Document Processing) and initial expandable entities are extracted from the abstract (Expandable Entity Extraction). When a user asks a question for an expandable entity, Qlarify uses a retrieval-augmented generation approach to form a response and retrieve attribution (Question Answering). Qlarify then suggests expandable entities within the response, allowing recursive expansions.
• Figure 5: Distribution of participants' self-reported ratings within each condition in the comparative evaluation. Participants in the Qlarify condition felt more satisfied with their exploration, more confident in retrieving relevant information from the full paper, more motivated to explore deeply, and a greater desire to use in the future. See Appendix \ref{['sec:comp_eval_survey_questions']} for the precise wording used in the survey questions.