More and more teachers in HKU are interested to produce instructional videos. However, they often encounter difficulties in producing effective and engaging videos by themselves, particularly in areas such as scripting and storyboarding. In this project, we want to analyze the relationship between linguistic properties (e.g. tone and coherence) of video scripts and engagement properties (e.g. average percentage viewed) of videos. Through analysis, we hope we can propose data-supported pedagogies to produce instructional videos, for engaging and effective online/blended learning experiences. To facilitate the analysis, we use deep-learning cognitive tools (IBM Watson Tone Analyzer) for determining high-level and sentence-level linguistic metrics. Currently, we have analyzed 21 videos (70 minutes in total) from an online course. 12 emotional/language/social tone metrics have been determined for analysis. Preliminary results show that tentative tone drives video retention. In other words, scripts with thought provoking instructions stimulate students to watch the whole instructional video.

In the future, we will conduct more in-depth analysis. In particular, we will analyze the scripts of the remaining 486 videos, with more linguistic metrics and relevant engagement metrics. Predictive analysis for storyboard screening will also be designed for script pre-screening.

Slides: https://docs.google.com/presentation/d/1FYLmCasd-Gm8PLXJJ7E4VLdlLQiTUOLBl59dln5iz64/pub?start=false&loop=false&delayms=5000