Aug 31 2017 - 01:58pm
Results of Vialogues Study

In my last blog post and presentation I referenced a study, Understanding user behavior in large-scale video-on-demand systems,  by Yu et al. I thought this study had some interesting analyses which could be applied to data from Vialogues. In particular I wanted to look at the relationship between the popularity of a Vialogue measured by the number of page visits and the session lengths or how long people spend on the Vialogue page.


Below is the distribution of Vialogue video lengths.


Video Durations

5 Minutes

10 Minutes

25 Minutes

50 Minutes

Percentage

48.5%

68.2%

84.0%

89.2%


I looked at sessions which remained on the page for less than 85% of the video to focus on the sessions that ended earlier than the length of the video. This accounted for 85.44 percent of the data collected since the release of Vialogues 2.0. Below is the distribution of session lengths.



Session Lengths

5 Minutes

10 Minutes

25 Minutes

50 Minutes

Percentage

95.14%

98.66%

99.88%

~100%


The data seems to show that almost all users have very short 0 to 10 minutes sessions. Meanwhile, less than half of the videos are less than 5 minutes long. This may suggest that users may be exploring the Vialogues site in order to find something interesting.


The plot below shows the relationship between popularity and time spent on page. This plot does not show a very strong correlation. Instead something interesting that this plot shows is that for some Vialogues there may be a few users who spend a lot of time on that page.


Lastly as in the study we plot the cumulative distribution function for four quartiles with 25 percent of the most popular of videos in the first quartile, the next 25 percent in the second and so on.
The results show the opposite as in the study for the Chinese Video-on-Demand service in that more popular Vialogues tended to have higher average session ratios. This is an interesting result because it may be that popular Vialogues may draw in more people because of greater amounts  of discussion. It would be interesting to later look at these relationships with the number of comments.





Posted in: Research|By: Alvaro Ortiz-Vazquez|169 Reads