Twitch使用者行為與分級審核制度之分析

Abstract

Twitch是擁有13 億用戶數的知名遊戲直播平台，每天有著超過9.9萬個頻道在進行直播。本論文對 Twitch 的研究可分為兩個部分進行探討。

在第一部分中，我們關注Twitch於歐、亞、美三洲的伺服器每日所需面對的流量，並使用直播數和觀眾數作為估計上傳與下載流量的指標。我們說明此二項指標擁有易收集的特質，並且對於像韓語或日語此類主要使用區域較為集中的語言，它們的全球直播數與觀眾數非常接近其區域的直播與觀眾數。然而，對於較廣為使用的語言例如英語或西班牙語，我們便無法直接將其全球直播與觀眾數作為其區域的直播與觀眾數。我們發現，利用串流時各直播的主播放清單(Master Playlist)中的ORIGIN屬性，我們便可以成功地分離多區域語言的洲際直播數與觀眾數，進而對各語言的區域流量進行數據分析。

在第二部分，我們研究了Twitch平台的內容分級制度。我們列出了Twitch分級制度的沿革，並說明其舊制度並未達到預期的目的，這是由於直播主不願意將其直播的內容正確地進行分級所導致的。我們的研究更發現，隨著新分級系統的引入，Twitch開始主動積極的對直播內容進行分類，這項改動使其分類的準確性獲得了巨大的提升。

最後，我們透過研究Twitch於西元2024年2月退出韓國市場的案例，了解此事件如何影響Twitch在韓國直播平台的地位，以及此次退出如何意外的揭示了Twitch新分級制度仍舊不完善。

Twitch being the leading live game streaming platform, has over 1.3 billion users and more than 99,000 channels streaming daily. In this thesis, our study regarding Twitch consists two parts. In the first part, we present our study of the daily traffic on continental servers, using streamer and viewer counts as representative metrics for the traffic of ingest and outgest servers, respectively. We show that obtaining the global counts for both streamers and viewers is simple and offers several benefits for certain languages. For uni-region languages such as Korean and Japanese, their global streamer and viewer counts closely match their regional counts. However, even at the continental scale, obtaining both counts remains challenging for multi-region languages like English and Spanish. By utilizing the language and ORIGIN attribute in the master playlist of a stream, we were able to successfully separate the continental results for multi-region languages, allowing us to display regional traffic statistics for each language. In the second part, we investigate how Twitch moderates its mature content. We present the changes of Twitch's mature label system, and show that the old system failed to serve its purpose, due to streamers being reluctant to label their stream content as mature-rated. However, since the introduction of the new system, Twitch started to automatically apply the unremovable mature label by the category currently streaming, leading to a huge increase in the rating accuracy. Lastly, we present a case study on Twitch's exit from the Korean market in February 2024, highlighting its impact on the Korean traffic load and how the exit unexpectedly exposed weaknesses in Twitch's new mature label system.