[论文解读] Trending Videos: Measurement and Analysis

Iman Barjasteh, Ying Liu+1 more|arXiv (Cornell University)|Sep 26, 2014

Complex Network Analysis Techniques参考文献 12被引用 8

一句话总结

本文利用九个月内收集的超过8,000个视频的高分辨率时间序列数据，对YouTube热门视频进行了测量与分析。研究发现，热门视频表现出独特的统计特性及通过格兰杰因果关系揭示的非对称方向性关系，显示观众流量强烈流向热门内容类别，而部分类别则保持孤立。

ABSTRACT

Unlike popular videos, which would have already achieved high viewership numbers by the time they are declared popular, YouTube trending videos represent content that targets viewers attention over a relatively short time, and has the potential of becoming popular. Despite their importance and visibility, YouTube trending videos have not been studied or analyzed thoroughly. In this paper, we present our findings for measuring, analyzing, and comparing key aspects of YouTube trending videos. Our study is based on collecting and monitoring high-resolution time-series of the viewership and related statistics of more than 8,000 YouTube videos over an aggregate period of nine months. Since trending videos are declared as such just several hours after they are uploaded, we are able to analyze trending videos time-series across critical and sufficiently-long durations of their lifecycle. In addition, we analyze the profile of users who upload trending videos, to potentially identify the role that these users profile plays in getting their uploaded videos trending. Furthermore, we conduct a directional-relationship analysis among all pairs of trending videos time-series that we have monitored. We employ Granger Causality (GC) with significance testing to conduct this analysis. Unlike traditional correlation measures, our directional-relationship analysis provides a deeper insight onto the viewership pattern of different categories of trending videos. Trending videos and their channels have clear distinct statistical attributes when compared to other YouTube content that has not been labeled as trending. Our results also reveal a highly asymmetric directional-relationship among different categories of trending videos. Our directionality analysis also shows a clear pattern of viewership toward popular categories, whereas some categories tend to be isolated.

研究动机与目标

• 测量并分析YouTube热门视频的生命周期动态，其与普通热门视频不同，具有短暂的病毒式传播特性。
• 识别热门视频与非热门内容在观众流量模式和用户行为方面的统计差异。
• 探究上传者个人资料在视频成为热门的可能性中所起的作用。
• 利用格兰杰因果关系对热门视频之间的时间序列关系进行建模，以揭示观众流量流动模式。
• 比较不同内容类别之间的观众流量动态，识别主导性或孤立的趋势。

提出的方法

• 收集了超过8,000个YouTube视频在九个月内的高分辨率观众观看时间序列数据及相关统计信息。
• 从视频被宣布为热门的时刻起即开始监控，从而实现对关键生命周期阶段的分析。
• 分析上传者个人资料，评估特定用户特征是否与热门传播成功相关。
• 应用格兰杰因果关系并结合显著性检验，检测成对热门视频时间序列之间的方向性关系。
• 对热门视频进行分类，以比较不同类型内容的观众流量动态。
• 使用统计显著性检验验证观众流量模式中的因果关系，避免虚假相关性。

实验结果

研究问题

• RQ1 与非热门内容相比，YouTube热门视频具有哪些独特的统计属性？
• RQ2 上传者个人资料在多大程度上影响视频成为热门的可能性？
• RQ3 热门视频时间序列之间存在哪些方向性关系？这些关系在不同内容类别中如何变化？
• RQ4 是否存在不对称的观众流量流动，即某些类别持续驱动其他类别？
• RQ5 某些热门视频类别是否保持孤立，对其他类别影响甚微或几乎无影响？

主要发现

• 热门视频表现出独特的统计属性，如观众流量迅速飙升和生命周期较短，与普通热门视频形成鲜明对比。
• 格兰杰因果关系分析揭示了高度非对称的方向性关系，观众流量主要流向热门内容类别。
• 某些热门视频类别充当了枢纽，推动其他类别的观众流量，而其他类别则保持孤立，影响微弱。
• 研究发现，热门视频的上传者通常具有独特的个人资料特征，暗示其在病毒式传播中可能扮演关键角色。
• 方向性关系分析揭示了清晰的内容传播模式，表明观众流量趋势并非随机，而是遵循有结构的流动路径。
• 结果表明，热门视频不仅仅是观看量高的内容，而是一类具有短暂性、由注意力驱动的独特现象。