YouTube

Paul Covington, Jay Adams, Emre Sargin

YouTube represents one of the largest scale and most sophisticated industrial recommendation systems in existence. In this paper, we describe the system at a high level and focus on the dramatic performance improvements brought by deep learning. The paper is split according to the classic two-stage information retrieval dichotomy: first, we detail a deep candidate generation model and then describe a separate deep ranking model. We also provide practical lessons and insights derived from designing, iterating and maintaining a massive recommendation system with enormous user-facing impact.

K. Sokolova, Hajer Kefi

Abstract Nowadays social network influencers play an important role in marketing by introducing products to their audience. In this article, we investigate the persuasion cues related to beauty and fashion influencers present on YouTube and Instagram. More precisely, we investigate how the para-social interaction (PSI) the audience creates with the online influencer, along with their perceived credibility, are related to the purchase intention and how they are, in turn, related to the social and physical attractiveness and attitude homophily. We base our research on four beauty influencers popular in France and control our results by the age of the participants and by the influencer. We find that attitude homophily is positively related to PSI but, surprisingly, the physical attractiveness shows negative relationship or no evidence of relationship. Both credibility of the influencers and PSI exhibit significant and positive relationships to purchase intention.

Sami Abu-El-Haija, Nisarg Kothari, Joonseok Lee, Apostol Natsev, G. Toderici, Balakrishnan Varadarajan, Sudheendra Vijayanarasimhan

Many recent advancements in Computer Vision are attributed to large datasets. Open-source software packages for Machine Learning and inexpensive commodity hardware have reduced the barrier of entry for exploring novel approaches at scale. It is possible to train models over millions of examples within a few days. Although large-scale datasets exist for image understanding, such as ImageNet, there are no comparable size video classification datasets. In this paper, we introduce YouTube-8M, the largest multi-label video classification dataset, composed of ~8 million videos (500K hours of video), annotated with a vocabulary of 4800 visual entities. To get the videos and their labels, we used a YouTube video annotation system, which labels videos with their main topics. While the labels are machine-generated, they have high-precision and are derived from a variety of human-based signals including metadata and query click signals. We filtered the video labels (Knowledge Graph entities) using both automated and manual curation strategies, including asking human raters if the labels are visually recognizable. Then, we decoded each video at one-frame-per-second, and used a Deep CNN pre-trained on ImageNet to extract the hidden representation immediately prior to the classification layer. Finally, we compressed the frame features and make both the features and video-level labels available for download. We trained various (modest) classification models on the dataset, evaluated them using popular evaluation metrics, and report them as baselines. Despite the size of the dataset, some of our models train to convergence in less than a day on a single machine using TensorFlow. We plan to release code for training a TensorFlow model and for computing metrics.

Jawed Karim

The study examined student exorbitations on youtube's content of domestic violence made by baim wong and Paula verhoeven. The domestic violence prank video narrated the domestic abuse report to the old kebayoran police. The theories used are the theories of reception analysis of Stuart hall (encoding-decoding) that have three types, the dominant positions of hegemonic, negotiation and opposition. The purpose of this research is to describe the growing noble university student's conception of domestic violence prank content in the baim Paula youtube channel. The data-gathering technique USES in-depth interviews with ten informers who are active students of the creative communications and design faculty, the budi university. The results of burial with 50 answers of the informant, there were 13 dominant exorcists, 15 negotiation empties and 22 elimination of opposition. It's the research. The results of outusing the use of opposition to domestic violence prank content that went viral in social media was much richer than the informants. With the violent prank video in the home, the informant thinks it could change people's view of the police image. Nevertheless, the informant says there are many other ways to create educational content on domestic violence that is more informative.

Zeyang Liu, Yiwen Chen, Ying Lin, Maomao Ai, Dongling Lian, Yuanhui Zhang, Yanxiong Shen, Feng Yu

YouTube, a widely recognized global video platform, is inaccessible in China, whereas Bilibili and TikTok are popular platforms for long and short videos, respectively. There are many videos related to laryngeal carcinoma on these platforms. This study aims to identify upload sources, contents, and feature information of these videos on YouTube, Bilibili, and TikTok, and further evaluate the video quality. On January 1, 2024, we searched the top 100 videos by default sort order (300 videos in total) with the terms “laryngeal carcinoma” and “throat cancer” on YouTube, “喉癌” on Bilibili and TikTok. Videos were screened for relevance and similarity. Video characteristics were documented, and quality was assessed by using the Patient Education Materials Assessment Tool (PEMAT), Video Information and Quality Index (VIQI), Global Quality Score (GQS), and modified DISCERN (mDISCERN). The analysis included 99 YouTube videos, 76 from Bilibili, and 73 from TikTok. Median video lengths were 193 s (YouTube), 136 s (Bilibili), and 42 s (TikTok). TikTok videos demonstrated higher audience interaction. Bilibili had the lowest ratio of original contents (69.7%). Treatment was the most popular topic on YouTube and Bilibili, while that was the prognosis on TikTok. Solo narration was the most common video style across all platforms. Video uploaders were predominantly non-profit organizations (YouTube), self-media (Bilibili), and doctors (TikTok), with TikTok authors having the highest certification rate (83.3%). Video quality, assessed using PEMAT, VIQI, GQS, and mDISCERN, varied across platforms, with YouTube generally showing the highest scores. Videos from professional authors performed better than videos from non-professionals based on the GQS and mDISCERN scores. Spearman correlation analysis showed no strong relationships between the video quality and the audience interaction. Videos on social media platforms can help the public learn about the knowledge of laryngeal cancer to some extent. TikTok achieves the best flow, but videos on YouTube are of the best quality. However, the video quality across all platforms still needs enhancement. We need more professional uploaders to ameliorate the video quality related to laryngeal carcinoma. Content creators also should be aware of the certification, the originality, and the style of video shooting. As for the platforms, refining the algorithm will allow users to receive more high-quality videos.