Affective Signals in a Social Media Recommender System

Auteur-es

Jane Dwivedi-Yu, Yi-Chia Wang, Lijing Qin, Cristian Canton-Ferrer, Alon Y. Halevy

Nombre Auteurs

5

Titre

Affective Signals in a Social Media Recommender System

Année de publication

2022

Référence (APA)

Dwivedi-Yu, J., Wang, Y.-C., Qin, L., Canton-Ferrer, C., & Halevy, A. Y. (2022). Affective Signals in a Social Media Recommender System. https://doi.org/10.1145/3534678.3539054

résumé

People come to social media to satisfy a variety of needs, such as being informed, entertained and inspired, or connected to their friends and community. Hence, to design a ranking function that gives useful and personalized post recommendations, it would be helpful to be able to predict the affective response a user may have to a post (e.g., entertained, informed, angered). This paper describes the challenges and solutions we developed to apply Affective Computing to social media recommendation systems.

We address several types of challenges. First, we devise a taxonomy of affects that was small (for practical purposes) yet covers the important nuances needed for the application. Second, to collect training data for our models, we balance between signals that are already available to us (namely, different types of user engagement) and data we collected through a carefully crafted human annotation effort on 800k posts. We demonstrate that affective response information learned from this dataset improves a module in the recommendation system by more than 8%. Online experimentation also demonstrates statistically significant decreases in surfaced violating content and increases in surfaced content that users find valuable.

Mots-clés

affective computing, recommendation systems, social media

URL

https://research.facebook.com/file/1116060675655243/Affective-Signals-in-a-Social-Media-Recommender-System.pdf

doi

https://doi.org/10.1145/3534678.3539054

Accessibilité de l'article

Libre

Champ

Machine Learning

Type contenu (théorique Applicative méthodologique)

Applicatif

Méthode

The authors collected training data for their models through a combination of engagement data on the platform and human labeling of posts. They used a two-tower architecture multi-class classifier to train their models.

Cas d'usage

N/A

Objectifs de l'article

The objectives of the article are to explore the challenges and solutions of applying Affective Computing to social media recommendation systems, and to demonstrate the effectiveness of incorporating affective response information in improving recommendation systems.

Question(s) de recherche/Hypothèses/conclusion

The research question is how can affective response information be leveraged to improve social media recommendation systems?

The hypothesis is that incorporating affective response information in recommendation systems will lead to improvements in user engagement and content quality.

The conclusions are that incorporating affective response information in recommendation systems can lead to significant improvements in user engagement and content quality, and that this approach can be applied to a variety of social media platforms.

Cadre théorique/Auteur.es

The theoretical framework of the article is Affective Computing, and the main authors cited include Jan Mizgajski, Mikołaj Morzy, Saif M Mohammad, Svetlana Kiritchenko, and Jessica Gall Myrick.

Concepts clés

Affective computing, Recommendation systems, Social media, Engagement data, human labeling, and Multi-class classification

Données collectées (type source)

de-identified Facebook posts, using a combination of engagement data and human labeling.

Définition des émotions

The authors do not provide a specific definition of emotions, but they do provide a taxonomy of affects that covers the important nuances needed for the application.

Ampleur expérimentation (volume de comptes)

800k posts

Technologies associées

Affective Computing, Recommendation systems

Mention de l'éthique

The authors do not explicitly mention ethics, but they do note that explicitly asking users their affective states regarding content has the drawbacks of being intrusive and potentially unreliable.