Reading the Room: Steerable Live Comment Clustering to Improve Speaker-Audience Understanding

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Speakers constantly undergo audience understanding checks where their communication is predicated on an updated perception of the audience at hand. This ability to “read the room” is critical for successful presentations, but there are many possible “rooms” of audiences, and many ways a speaker may want to “read” their responses. Status quo interaction methods flatten these audience understanding checks onto one lens—i.e., via the first raised hands or the most-upvoted comments—but what if a PhD student instead wants to monitor excitement about lines of future work, an educator wants to prepare for areas of student confusion, or a researcher wants to examine critiques from external fields? To aid speakers in their unique goals, we introduce Refractor, a tool that supports speakers through audience understanding checks via speaker-specified text inputs called lenses . For example, a speaker can specify an “urgent questions” lens during a live presentation, and the system will display a high-level synthesis of time-critical audience misunderstandings. Then, after the talk, the speaker can apply an “interdisciplinary critiques” or “suggestions to expand work” lens to explore the audience’s perspectives on future work. To instantiate Refractor, we introduce Steerable Live Clustering (SLiC), an algorithm that uses speaker-specified lenses to rapidly prioritize and thematically group text comments at interactive speeds. Via a series of live deployments, we observe that participants used Refractor to pursue a myriad of directions for audience understanding checks, such as brainstorming new directions, prioritizing important feedback, and triaging relevant questions in a timely manner.