以使用者喜好分群為基礎之實況串流頻道推薦系統

Abstract

隨著科技不斷進步，影音串流的技術也正在快速發展，透過實況串流平台觀看節目的使用者越來越多。但是，實況串流平台上同時會有上千甚至上萬個頻道在線上，使用者無法一一尋找自己喜歡的頻道，因此一個好的推薦系統對使用者來說是必要的。而現今實況串流平台的推薦方法是推薦人氣最高的影片給使用者，但這卻忽略了使用者的個人喜好。因此我們將使用者依照個人喜好分成許多群，被分到同一群的使用者，他們觀看頻道的喜好是相似的。接著將使用者喜歡的內容推薦給在同一群的其他使用者。除此之外，我們觀察到有些使用者對於頻道的忠誠度極高，他們總是只會觀看相同的兩、三個頻道。因此我們將MRV(Most Recent View)與我們的方法合併作為我們的推薦系統。我們的資料來源為Twitch，其為目前當紅的遊戲實況串流平台。我們會在實驗的部分驗證我們的推薦系統比現有的實況串流平台推薦系統更加準確。

With the advance of technology, live streaming services have been growing rapidly and the number of viewers who watch programs or shows on live streaming channels have highly increased. However, there are thousands of channels online simultaneously, it is difficult for a viewer to look for his or her favorite channels by searching channels one by one. Therefore, a precise recommendation system is necessary for a viewer. The existing recommendation system on live streaming channels recommends popular channels, however, this approach does not take viewer’s preference into account. For this reason, we separate the viewer into multiple group based on their personal preference. The viewers in the same group have similar preference on watching channel i.e., watching behavior. Next, we recommend a viewer’s favorite to the other viewers who are in the same group. Besides, according to our observation, there are some viewers showing high loyalty to some of the channels, that is, they always watch the same channels. Hence, we combine our proposed method with MRV(Most Recent View) as our recommendation system. Our data are extracted from Twitch, which is a well-known live streaming service, and we will verify that our recommendation system is more accurate than the existing recommendation system in our experiment.