"I Need to Find That One Chart": How Data Workers Navigate, Make Sense of, and Communicate Analytical Conversations

TL;DR

A design probe is developed that supplements analytical conversations with structured elements and affordances to support re-visitation and communication of analytical conversations and discusses design implications to support re-visitation and communication of analytical conversations.

Abstract

Conversational interfaces are increasingly used for data analysis, enabling data workers to express complex analytical intents in natural language. Yet, these interactions unfold as long, linear transcripts that are misaligned with the iterative, nonlinear nature of real-world analyses. Revisiting and summarizing conversations for different contexts is therefore challenging. This paper investigates how data workers navigate, make sense of, and communicate prior analytical conversations. To study behaviors beyond those supported by standard interfaces (i.e., scrolling and keyword search), we develop a design probe that supplements analytical conversations with structured elements and affordances (e.g., filtering, multi-level navigation and detail-on-demand). In a user study (n = 10), participants used the probe to navigate and communicate past analyses, fulfilling information needs (recall, reorient, prioritize) through navigation strategies (visual recall, sequential and abstractive) and summarization practices (adding process details and context). Based on these findings, we discuss design implications to support re-visitation and communication of analytical conversations.

Figures (9)

• Figure 1: SyncSense extracts and presents elements of an analytical conversation at varying levels of abstraction. The Outline panel provides a high-level overview of the conversation elements, the Detail Explorer panel displays additional details about each element, and the Annotated Conversation panel shows the raw conversation annotated with each component. These three synchronized panels are connected visually and interactively through the structured elements of an analytical conversation (§\ref{['sec:adding_structure']}).
• Figure 2: The Outline panel offers a visual overview of the conversation by displaying components (i.e., threads, speech acts, insights, and artifacts) as vertically arranged icons following the dialogue turns (A). Hovering over icons reveals additional details (B). Users can filter by speech acts, insights, and artifacts which update the outline view and the contents in the Detail Explorer(C).
• Figure 3: Detail Explorer and Annotated Conversation panels. The Detail Explorer panel presents a structured, mid-level view of the conversation (A). Threads can be expanded to reveal individual turns and elements (A1). Icons with content counts summarize nested elements (A2). Clicking the To Turn button navigates to the corresponding turn in the Annotated Conversation panel (A3). The Annotated Conversation panel displays the full raw conversation, annotated with speech act tags and artifact icons to support provenance and traceability (B).
• Figure 4: Summary Authoring Process in the Authoring Panel. Users can drag items from the Detail Explorer or Annotated Conversation panels into the drop container within the Authoring panel (A). Within the drop container, the items can be rearranged by dragging to adjust the order or nesting structure (B). The contents are serialized in the post-drop editor, where users can utilize an LLM to reformat the summary based on specified parameters such as length, technical detail, and formality (C). Finally, the LLM-generated output is displayed in the post-LLM editor so users can make any final edits.
• Figure 5: Analytical Conversation Elements from Session 1 (Sec. \ref{['sec:dataAnalysisSession']}). Across participant conversations, all speech acts and artifacts types were observed. Except for Extreme, all types of insights appeared. The top chart shows the average number of turns each element was involved in, while the bottom chart shows the average count of unique insights per conversation.