Modeling Self-Disclosure in Social Networking Sites

Auteur-es

Yi-Chia Wang, Moira Burke, Robert Kraut

Nombre Auteurs

3

Titre

Modeling Self-Disclosure in Social Networking Sites

Année de publication

2016

Référence (APA)

Wang, Y.-C., Burke, M., & Kraut, R. (2016). Modeling Self-Disclosure in Social Networking Sites. https://doi.org/10.1145/2818048.2820010

résumé

Social networking sites (SNSs) offer users a platform to build and maintain social connections. Understanding when people feel comfortable sharing information about themselves on SNSs is critical to a good user experience, because self-disclosure helps maintain friendships and increase relationship closeness. This observational research develops a machine learning model to measure self-disclosure in SNSs and uses it to understand the contexts where it is higher or lower. Features include emotional valence, social distance between the poster and people mentioned in the post, the language similarity between the post and the community and post topic. To validate the model and advance our understanding about online self- disclosure, we applied it to de-identified, aggregated status updates from Facebook users. Results show that women self-disclose more than men. People with a stronger desire to manage impressions self-disclose less. Network size is negatively associated with self-disclosure, while tie strength and network density are positively associated.

Mots-clés

Social networking sites; Facebook; computer-mediated
communication; self-disclosure; personality; audience;
context collapse; natural language analysis; applied
machine learning

URL

https://research.facebook.com/file/271555647830591/modeling_self-disclosure_in_social_networking_sites.pdf

doi

https://doi.org/10.1145/2818048.2820010

Accessibilité de l'article

Libre

Champ

Data Science, Human Computer Interaction & UX

Type contenu (théorique Applicative méthodologique)

Méthodologique

Méthode

The authors developed a machine learning model to measure self-disclosure and applied it to de-identified, aggregated status updates from Facebook users to validate the model and advance our understanding about online self-disclosure. They also used Latent Dirichlet Allocation (LDA) to identify the topics common in status updates.

Status updates were annotated by users and by trained judges.

Cas d'usage

Facebook

Objectifs de l'article

The objectives of the article are to develop a machine learning model to measure self-disclosure on social networking sites, to validate the model using deidentified, aggregated status updates from Facebook users, and to understand the contexts where self-disclosure is higher or lower.

Question(s) de recherche/Hypothèses/conclusion

The research question is how can we measure self-disclosure on social networking sites using machine learning?

The hypothesis are :
H1: Individuals with a stronger desire for impression management will self-disclose less.
H2: Women will self-disclose more than men.
H3: Network size will be negatively correlated with selfdisclosure.
H4: Average tie strength will be positively correlated with self-disclosure.
H5: Network density will be positively correlated with selfdisclosure

The authors conclude that their machine learning model was able to accurately measure self-disclosure on Facebook and that self-disclosure was higher for certain topics, such as politics and memorial, and lower for others, such as Christianity and deep thoughts.

Cadre théorique/Auteur.es

The theoretical framework of the article includes social penetration theory and communication privacy management theory. The main authors cited include Altman and Taylor, Derlega and Chaikin, and Petronio.

Concepts clés

Self-disclosure, Social networking sites, Machine learning

Données collectées (type source)

The authors collected de-identified, aggregated status updates from Facebook users (recruited on AMT). They used dataset from the myPersonality project. The dataset contains users’ status updates as well as their demographic information and self-report impression management scores

Définition des émotions

LIWC

Ampleur expérimentation (volume de comptes)

2000 Facebook status updates

Then a random sample of 412,470 English language Facebook active users for approximately one month in late 2014

Technologies associées

ML et Latent Dirichlet Allocation (LDA)

Mention de l'éthique

Non

Commentaires

Financed by the grants from National Science Foundation (IIS-0968485) and National Institute of Mental Health (R21 MH106880-01).