"Can you believe [1:21]?!"

Abstract

As videos become increasingly ubiquitous, so is video-based commenting. To contextualize comments, people often reference specific audio/visual content within video. However, the literature falls short of explaining the types of video content people refer to, how they establish references and identify referents, how video characteristics (e.g., genre) impact referencing behaviors, and how references impact social engagement. We present a taxonomy for classifying video references by referent type and temporal specificity. Using our taxonomy, we analyzed 2.5K references with quotations and timestamps collected from public YouTube comments. We found: 1) people reference intervals of video more frequently than time-points, 2) visual entities are referenced more often than sounds, and 3) comments with quotes are more likely to receive replies but not more "likes". We discuss the need for in-situ dereferencing user interfaces, illustrate design concepts for typed referencing features, and provide a dataset for future studies.

AQuA: Automated Question-Answering in Software Tutorial Videos with Visual Anchors

“…A notable pattern that emerged in these categories was the tendency of users to reference parts of the video in their comments to provide temporal and spatial context. In particular, we noticed several cases of referring to the visual part of the video, which aligns with previous findings on general videos [74]. In contrast to this prior work [74], we observed this practice in software tutorial videos, in which it can be particularly evident since they feature visual demonstrations through screen sharing.…”

“…In particular, we noticed several cases of referring to the visual part of the video, which aligns with previous findings on general videos [74]. In contrast to this prior work [74], we observed this practice in software tutorial videos, in which it can be particularly evident since they feature visual demonstrations through screen sharing.…”

“…Since these types of questions have direct relevance to video content, a notable trend that emerged from our analysis was the frequent references to video in these questions, sometimes explicitly citing timestamps, which echoes findings from prior research on referencing behavior in comments on a variety of videos [74]. It implies that our answer pipeline design should account for what in the video a question was about and the context of the tutorial when the question was posed.…”

“…This resulted in 633 questions out of the 5944 comments. Similar to Yarmand et al [74], the lead author performed thematic coding of the set of 633 questions and iteratively discussed with the other authors to validate the codes and resolve any conflicts. After finalizing the codes, we grouped them into four main categories, reflecting the overarching themes of the questions.…”

“…Referencing specific audio or visual content within a video is a common practice during video interactions [59,74] suggests that the ability to easily refer to a part of a video enables a range of different applications, including enhanced engagement in live streams [12,72]. The ability to refer to parts of a video fosters a clear understanding of what others are discussing and facilitates pinpointed feedback or areas of confusion.…”