310 likes | 482 Views
Comparing Presentation Summaries: Slides vs. Reading vs. Listening. Liwei He, Elizabeth Sanocki Anoop Gupta, Jonathan Grudin Collaboration and Multimedia Group Microsoft Research. Motivation. Multimedia presentations are being archived for on-demand access University courses
E N D
Comparing Presentation Summaries:Slides vs. Reading vs. Listening Liwei He, Elizabeth Sanocki Anoop Gupta, Jonathan Grudin Collaboration and Multimedia Group Microsoft Research
Motivation • Multimedia presentations are being archived for on-demand access • University courses • Corporate training and seminars • Effective summarization and skimming can help users utilize time better
Video Skimming Techniques • Time compression • 1.5 – 2.5 saving factor at most • Video summary • 2.5+ saving factor is possible
Previous Summarization Study • Compared 4 video summary techniques • 1 by authors of the presentation • 3 by computer algorithms • Pre- and post quizzes and subjective ratings • More details in our paper in ACM Multimedia 99 • “Auto-Summarization of Audio-Video Presentations”
Auto Summary Study Results • All four summaries improve quiz scores • Human-generated summary is significantly better than computers • No difference among computer-generated summaries • Overall, all are appreciated by subjects
Questions Raised • What about other forms of summary? • Amount of information from slides? • Skimming text transcript vs. watching video? • Transcripts with key points highlighted vs. video summaries?
Experimental Design (1) • 4 summarization techniques • PowerPoint slides only • Raw text transcripts • Transcripts with key points highlighted • Author-generated video summaries
Experimental Design (2) • To compare summarization techniques • Objective measure: quiz score improvement before and after each summary • Subjective measure: user ratings • 4 talks chosen from Microsoft training site • Original presenters wrote quiz questions
Experimental Design (3) • 24 Microsoft employees were subjects • 32 question pre-quiz • 4 summaries, each of a different type • 8 quiz questions after each summary • Summary types and talks are counter-balanced within each subject
Quiz Score Improvement (1) • Plot by summary types
Quiz Score Improvement (2) Video-summary > highlight-text ? p = 0.087
Quiz Score Improvement (3) Highlight-text and video-summary are significantly better than (at p<0.001) slide-only and raw-text
Quiz Score Improvement (4) Plot by presentations
Subjective Ratings (1) Table by summarization techniques
Subjective Ratings (2) Highlight-text and video-summary are significantly better than (at p=0.01) slide-only and raw-text
Subjective Ratings (3) Highlight-text and video-summary are not significantly different (at p=0.05)
Subjective Ratings (4) Raw-text, highlight-text, and video-summary are significantly better than (at p=0.05) slide-only
User Comments • 13 out of 24 like video-summaries • “It is more enjoyable listening and seeing the presenter.” • 11 prefer highlighted transcripts • “I liked having the option of being able to get more detailed info when I need it.”
Conclusions • Effective summary techniques are key • Slide-only does not work well for most talks • Authoring style makes a difference • Raw text transcript is hard to read • Human produced summaries work better
Conclusions (cont.) • What to do? • For authors: tools to generate summaries • For users: interactive and intelligent video browser
Compare with AutoSum Study (1) • Current study and auto summary study are comparable • 4 talks and quiz are the same • Both have author-generated summary • Slides are shown in all conditions for both • Evaluation methods are the same
Compare with AutoSum Study (2) AutoSum Study Current Study
Compare with AutoSum Study (3) Subject ratings (AutoSum vs. Current)
Compare with AutoSum Study (4) A* in AutoSum consistently > A
Compare with AutoSum Study (3) Slide-based summary (S) > slide only (SO)