Y. Borghol et al. Performance Evaluation 68 2011

Citation

Youmna Borghol, Siddharth Mitra, Sebastien Ardon, Niklas Carlsson, Derek L. Eager, Anirban Mahanti: Characterizing and modelling popularity of user-generated videos. Perform. Eval. 68(11): 1037-1055 (2011)

Online version

[1]

Summary

This is a paper proposes a model for the dynamics of YouTube videos popularity based on the data set they collected over 8 months. Specially it claims the peak view of an individual video follows a certain distribution called Time-to-peak distribution. Based on it they divide the view into three phases namely before, at or after peak. Finally a three-phase evolution model is brought forward to explain the dynamics of video views for newly-uploaded videos.

Data set

1.1 million videos metadata collected at weekly level over 8 months in the following two ways: 1) sampling from the recently-uploaded videos (29,791 videos collected) and 2) sampling using keyword search (1,135,253 videos collected).

There are potential two drawbacks in their collecting methods. First collecting views in a weekly manner would discard some important information about the popular videos such as viral videos. According to Google's paper [2], many videos reach their peek-view within a week and receive 25% social view in their first uploaded day.

Secondly, about 97% videos metadata were collected by keyword search which is significantly biased towards the current #views and the age of the response videos. The ranking algorithm often favors the newly updated and popular videos given the similar keyword relevance. In other words, the data set may not well represent a random proportion of Youtube videos (In contrast the data set used in [3] derived from random sampling)

Therefore the conclusion drawn from this data set needs a careful examination and further validation.

Method

1. Time-to-peak Distribution

Through the empirical analysis, the authors claim that the time-to-peak distribution approximately follows an exponential distribution and they found that a large fraction of videos peak within the first six weeks.

Observation I: The videos with higher socialness has the highest growth leading up to the peak as well as the highest declination rate.

Observation II: The videos with higher socialness enjoys the peaks that are systematically higher than less social videos.

2. Socialness & categories

Different measurement for the level of socialness would yields different categories that are most social, see Fig.5.

Observation III: In terms of the fraction of videos within the category that are highly social, the most social category is "Pets & Animal"; in terms of the fraction of views within the category that are social the category becomes "Education"; Regarding the absolute number of social views within category the answer is "Music".

If Facebook and Twitter are regarded as two categories representing social network sites and micro-blogs they found:

Observation IV: The Twitter views (by a factor of 4.5) are more highly concentrated near the day of peak viewing than the Facebook views (by a factor of 2.4) which may be result of Twitter's real-time sharing paradigm. In addition the Twitter views are more likely to be associated with highly shared videos than Facebook views are.

3. Viral vs. Popular videos

The authors track two videos a viral video and a popular music video which received most of its views from searches. They found that

Observation V: The viral video tends to peak more sharply and wane more rapidly whereas the popular music video exhibits a steady and regular growth pattern after the peak view.

Observation VI: YouTube related video and search are the other two major reasons account for large number of views besides the social sharing. Most of the popular videos (top 1% videos in terms of views) result from the related videos and search.

If defining popular ratio as:

$PR(video)={\frac {\text{Views in the Second Month}}{\text{Views in the First Ten Days}}}$

A video is called short-term popular if its PR is low and the one with large PR is called long-term popular, see Fig 12.

Observation VII: the density of viral and non-viral videos in short-term popular videos set are similar. However, almost no viral videos belongs to the long-term popular videos (because viral video fade quickly after its peek view).

4. Ranking social sites

By the viral videos, the authors propose a method to rank the social blogs in terms of their propensity of spreading the viral videos. Let $V_{p}$ be the set of viral videos (with at least 60% of social views in the first month) and $V_{u}$ as the set of unpopular videos (with less than 100 views in the first 30 days) and $W_{100}(u)$ be the set of videos with at least of 100 views from url u.

$r(u)={\frac {|V_{p}\cap W_{100}(u)|}{|V_{p}\cup V_{u}|}}$

r(u) is applied to filter out ($r(u)<r_{low}$) some outliers such as Facebook. For each video $v\in V_{p}$, $views(u,v)$ denotes its number of view coming from url u. Based on the notions, the rank function for each url becomes:

$R(u)=\sum _{v\in V_{p}}views(u,v)$