結合意見探勘的YouTuber推薦系統之研究

A Study of YouTuber Recommendation System Combined with Opinion Mining

陳竑嘉(Hung-Chia Chen)

指導教授 ： 蕭瑞祥

淡江大學/商管學院/資訊管理學系碩士班/碩士(2018年)

https://doi.org/10.6846/TKU.2018.00694

摘要

自2004年Web2.0使用者提供網站內容的技術被提出後，網站經營模式朝向吸引使用者來提供內容的模式發展，例如：由使用者提供影片並且透過影片的觀看次數創造網站流量的YouTube，而Facebook、Twitter這些社群平台也是其中之一。YouTube的影片創作者又稱YouTuber，YouTuber所創作的影片也會影響使用者對品牌的觀感，由於有些廠商會找YouTuber進行業務配合，將該廠牌產品放入YouTuber影片當中，對於觀賞影片的人不只可能去嘗試該產品，更可能透過YouTuber的影片對於該廠牌產生認同，所以YouTuber對於社群影響力是不可小覷的。但由於YouTuber的討論僅止於口耳相傳，使用者須經過不斷的瀏覽及搜尋才能找到所喜歡的YouTuber。因此本研究提出基於意見探勘的YouTuber推薦系統，不同於傳統YouTube透過使用者點擊大量影片後進行頻道的推薦，頻道當中包含YouTuber及其他不屬於YouTuber的頻道。本研究提出之系統透過意見探勘的方式取得觀賞影片後使用者們對於影片及製作該影片之YouTuber的相關評論並且使用辭庫比對法進行分析，進一步進行推薦。在透過實驗測試並且比較傳統型YouTube所推薦的YouTuber，由F-Measure評估方法得知，本研究所提出之基於意見探勘的YouTuber推薦系統比傳統YouTube所推薦的YouTuber之準確率來得高18.39%。另外由Mean Average Precision(MAP)評估系統精確度得知，本研究提出之系統之MAP值較傳統YouTube之MAP值高32.44%、MRR值高37.19%。而期望透過本研究能夠更進一步改善傳統YouTube的YouTuber推薦準確率問題。

關鍵字

YouTube ； YouTuber ； 推薦系統 ； 意見探勘

並列摘要

Since 2004 Web 2.0 had been raised, users providing website content was proposed, the business model of the website is developing towards attracting users to provide content. For example, YouTube that provides users with videos and creates website traffic through the number of views of the videos, The social networks like Facebook and Twitter are among them.
YouTube's video creator, also known as YouTuber, will also influence the user's perception of the brand, as some vendors will find YouTuber for business cooperation, and put the branded product into YouTuber videos for those who watch the video. It’s not just possible to try the product, it’s more likely that YouTube’s video will identify the brand, so YouTube’s influence on the community is not to be underestimated. However, since YouTuber's discussion is only word of mouth, users must constantly browse and search to find their favorite YouTuber.
Therefore, this study proposes a YouTuber recommendation system combined with opinion mining, which is different from traditional YouTube in which channels are recommended after a user clicks a large number of videos. The channels include YouTuber and other channels that do not classify as YouTuber. The prototype system proposed in this study obtains viewers' comments on the videos and the YouTuber making the videos through opinion-finding methods and analyzes them using the dictionary comparison method to make further recommendations.
Through the experimental test and comparing the YouTuber recommended by traditional YouTube, the F-Measure evaluation method shows that the YouTuber recommendation system combined with opinion mining is 18.39% higher than the YouTube-recommended YouTuber. In addition, the accuracy of the Mean Average Precision (MAP) evaluation system shows that the MAP value of the system proposed by this study is 32.44% higher than the traditional YouTube MAP value, and the MRR value is 37.19% higher. It is hoped that through this research, the YouTube recommendation accuracy rate of traditional YouTube can be further improved.

並列關鍵字

YouTube ； YouTuber ； Recommendation System ； Opinion Mining

參考文獻

一.中文文獻

[1] 丁昭尤，《體驗行銷、體驗價值、觀光意象、 遊客滿意度與忠誠度關係之研究— 以台東青山休閒農場為例》，碩士論文，國立台東大學環境經濟資訊管理所，2009

[2] 李文誦，《InternetTV平台上情緒感知多媒體代理人之設計與實作》，碩士論文，中正大學通訊工程研究所，2009

[3] 林祐任，《未來性資訊檢索系統基於網路論壇之研究》，碩士論文，淡江大學資訊管理學系碩士班，2015

[4] 徐舜基，《整合準則權重於多準則協同過濾推薦之研究》，碩士論文，中國文化大學資訊管理研究所，2010