請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/68327
標題: | Instagram 平台 YouTubers 推薦 YouTubers Recommendation on Instagram |
作者: | Chih-Ling Hsu 許芷菱 |
指導教授: | 李瑞庭(Anthony J. T. Lee) |
關鍵字: | YouTuber 推薦,網紅,深度學習,注意力機制,冷啟動問題, YouTuber recommendation,internet celebrity,deep learning,attention mechanism,cold start problem, |
出版年 : | 2021 |
學位: | 碩士 |
摘要: | YouTube已成為最大製造網紅的社群平台之一,網紅可透過YouTube頻道與他(她)的追隨者或頻道訂閱者保持緊密聯繫。許多網紅不僅在YouTube 上發布影片,更在其他的社群平台上分享相關的內容,尤其在熱門社群平台如:Instagram,推廣他們的頻道和吸收更多的追隨者,以增加其能見度與影響力。然而,有些YouTubers在YouTube上很紅,但在Instagram上卻不紅,反之亦然,因此,本研究提出了一個深度學習的架構,推薦YouTubers 給 Instagram上的使用者。所提出的研究架構包括三個階段,第一階段,我們從YouTube蒐集資料,並提取出文字、影片與社群相關的特徵;第二階段,我們從Instagram蒐集資料,並提取出圖片、社群與關注相關的特徵,接著將兩個平台的特徵融合,以特徵化每位YouTuber;最後,我們提出一個推薦模型,結合上述的融合特徵和注意力機制來學習使用者的喜好,並將得分最高的前k位YouTubers推薦給使用者。實驗結果顯示,我們提出的方法優於所有的比較方法,並可減輕冷啟動問題的效應。我們的研究可幫助YouTubers、網紅或企業擬訂有效的行銷策略,亦可幫助使用者發現更多感興趣的YouTubers。 YouTube has become one of the largest platforms for creating internet celebrities, where the internet celebrities build and maintain a channel to establish tight connections with their followers or subscribers. Moreover, many internet celebrities tend to share their posts on multiple platforms, especially on popular one like Instagram, to promote their YouTube channels and acquire more followers to increase their reputations and influences. However, some YouTubers popular on YouTube are not popular on Instagram, and vice versa. Therefore, in our study, we propose a deep learning framework for recommending YouTubers to users on Instagram. The proposed framework contains three phases. First, we extract the features from the data collected from YouTube, including textual features, video features and social features. Next, we extract the features from the data collected from Instagram, including photo features, social features and following features. Then, we combine these features together to characterize each YouTuber. Finally, based on the combined features, we design an attentive recommendation model for computing the recommendation score of each YouTuber, and recommend top-k YouTubers with highest scores to users. The experiment results show that our proposed model outperforms the state-of-the-art methods in terms of precision, recall, F1-score and Normalized Discounted Cumulative Gain (NDCG), and mitigates the effect of cold start problems. Our study can help YouTubers, internet celebrities and businesses formulate effective marketing strategies, and assist users in discovering the YouTubers of interest. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/68327 |
DOI: | 10.6342/NTU202100693 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊管理學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-1002202109394800.pdf 目前未授權公開取用 | 2.46 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。