Audiovisual style of user-generated YouTube videos
Here's the abstract and the data file of my thesis. You can also download the whole text (PDF, in Czech only).
The byproduct of this thesis is a random sample of a few hundred thousand YouTube videos which you can download and use for your own research.
This thesis deals with amateur content on YouTube (www.youtube.com) through the lens of Bordwell and Thompson's ﬁlm theory. Eight distinct categories of audiovisual style are identiﬁed within a sample of 50 videos. These categories are deﬁned and described using eight different case studies. Consequently, the exact proportion of these categories is ascertained by means of a quantitative study of 385 YouTube videos. The study shows that the most frequent forms of expression on YouTube are so-called slideshows and video snapshots. Other categories of audiovisual style that were found among user generated YouTube videos include: videoblog, home video snippet, screencast, document, short ﬁlm and editing remix. Only a small proportion of YouTube videos does not ﬁt in the eight categories. The thesis also includes a treatise about the media signiﬁcance of YouTube, a summary of available texts about YouTube, and an enumeration of factors inﬂuencing audiovisual style of YouTube videos. The thesis further points out the shortcomings of present YouTube research texts and proposes a method that should ensure the most representative sample of YouTube videos possible.
The data file is in SPSS .sav format. You can download it here.
Here are some of the interesting findings of this paper:
- 79 % of YouTube videos are user-generated. The rest is either professional or ripped content.
- 25 % of user videos on YouTube are actually slideshows of static photographs
- 18 % of YouTube videos are "video-snapshots" - short, quick and dirty recordings of everyday stuff
- the average duration of a YouTube video is 4:46
- if the claim that 20 hours of video are uploaded every minute is true, then there are 365,512 videos uploaded every day
Please note: N=385, confidence interval ±5%, confidence level 95%. Also, the sample might not be completely unbiased – see Sampling YouTube.