Video Watch Time and Comment Sentiment: Experiences from YouTube · EMOTICONES

Auteur-es

Yang, Rong; Singh, Sarvjeet; Cao, Pei; Chi, Ed; Fu, Bo

Nombre Auteurs

5

Titre

Video Watch Time and Comment Sentiment: Experiences from YouTube

Année de publication

2016

Référence (APA)

Yang, R., Singh, S., Cao, P., Chi, E., & Fu, B. (2016). Video Watch Time and Comment Sentiment : Experiences from YouTube. 2016 Fourth IEEE Workshop on Hot Topics in Web Systems and Technologies (HotWeb), 26‑28. https://doi.org/10.1109/HotWeb.2016.13

Mots-clés

Sentiment, video watch time

URL

https://ieeexplore.ieee.org/document/7785813

doi

https://doi.org/10.1109/HotWeb.2016.13

Accessibilité de l'article

Open access

Champ

Machine Intelligence

Type contenu (théorique Applicative méthodologique)

Applicative

Méthode

We used an analysis tool called Scarlett to perform the sentiment analysis of the comments. Scarlett summarizes text sentiment by detecting and analyzing positive and negative opinions expressed towards a given list of entities, and by finding snippets that represent the opinions in the text. For each comment, Scarlett generates two values to represent its sentiment: polarity value and magnitude value.
We looked at the daily record of video watch time as well as the sentiment of the comments.
Sentiment analysis of comment

Cas d'usage

YouTube

Objectifs de l'article

In this paper, we analyze how video watch time correlates with the comment sentiment. In particularly, we want to verify the findings in a recently published paper by Park et al, which claims that the video watch time positively correlates with the percentage of negative comments on the video

Question(s) de recherche/Hypothèses/conclusion

Research question(s) : While video watching is now an indispensable part of the general public media consumption, yet very little is known about the relationship between how users interact with each other and how that affects video consumption patterns.

Hypothesis(es) : In this paper, we explore the relationship between user commenting behavior and how that might or might not be predictive of video consumption patterns such as watch time.

Conclusion(s) : Contrary to recent findings, we found that video watch time is correlated with the positive sentiment expressed in the comments of YouTube videos. More precisely, videos with more positive sentiment on average in the comments are more likely to be watched longer; while videos with negative comment sentiment on average are more likely to have shorter watch durations. These results suggest that users prefer videos that evoke positive emotional responses.
Our analysis shows the opposite results on the correlation between video watch time and negative comment percentage. That is the video watch time is negatively correlates with the percentage of negative comments. Given that the data from the previous paper was collected using a third party web crawler, we suspect that there might be bias with the crawler itself. Moreover, the data was collected much earlier than the data we used for our analysis. It is possible that user behavior might have changed over time.

Cadre théorique/Auteur.es

View duration and comments on YouTube (Park et al., 2016)

Concepts clés

Sentiment analysis

Données collectées (type source)

We looked at the daily record of video watch time as well as the sentiment of the comments. From the watch time log, we extract the following information for each viewed video :
Daily total views / Daily watch time in hours / Average watch time per view in minutes / Average watch time duration (i.e. ratio of average watch time per view to total video length)

For each of the viewed video on a given day, we extract the following sentiment related signals :
Average sentiment polarity of the comments created on the day of view / Percentage of positive comments created on the day of view / Percentage of negative comments created on the day of view

Définition des émotions

Methodological explanation of their classification (quantitative)

No definition

Positive and negative labeling

Ampleur expérimentation (volume de comptes)

Using one day of watch time data from over 10k videos

Overall, we included 10252 videos in the dataset. All of this videos have at least 10 comments that satisfy the sentiment magnitude condition (i.e. larger>1.0).

Technologies associées

Analysis tool called Scarlett

Mention de l'éthique

ND

Finalité communicationnelle

Understanding patterns of user engagement in YouTube, therefore, takes on paramount importance for researchers, marketers, and engineers alike.

Résumé

While video watching is now an indispensable part of the general public media consumption, yet very little is known about the relationship between how users interact with each other and how that affects video consumption patterns. In this paper, we explore the relationship between user commenting behavior and how that might or might not be predictive of video consumption patterns such as watch time. Contrary to recent findings, we found that video watch time is correlated with the positive sentiment expressed in the comments of YouTube videos. More precisely, videos with more positive sentiment on average in the comments are more likely to be watched longer, while videos with negative comment sentiment on average are more likely to have shorter watch durations. These results suggest that users prefer videos that evoke positive emotional responses. If the findings here generalizes to other social media, this suggests a motivational design finding that is useful for other system designers.