Facebook « networked timelines » to study life events
Raphaël Charbey, Christophe PrieurThere has been a lot of research on personal networks taking into account the passage of time, and on the effect of life events on the structure of these networks (see, for instance, Bidart et al., 2011; Lubbers et al, 2010; Widmer, 2016, Fischer & Offer, 2020). While a social network service such as Facebook gives, of course, a partial vision of an individual’s personal network, it still provides valuable data to investigate the changes in their relationships.
Our work relies on a survey conducted from 2013 to 2015 (Bastard et al, 2017 ; Charbey & Prieur, 2019) among more than 16 000 respondents, from which we gathered, through their Facebook accounts, both their networks of friends, and all their posts along with the reactions they got (comments and like) from their friends. This (anonymized) dataset gives, for each respondent (called ego), a timeline of events (posts and comments, of whom only keywords have been extracted from the initial text) involving ego and their friends (alters) over a period of up to seven years (starting when they created their Facebook account). Besides this data collected through the Facebook API, the survey also included a short questionnaire asking for sociodemographic information such as age, gender, profession, place of residence.
The aim of this study is to show how this large amount of « real-time » interaction data goes along with the significant changes that leave their mark on the personal network of the respondents. As a proxy for ego’s social circles, we use a network clustering method (namely, the Louvain method) on the network as it was at the moment of the data collection. For successive time slots of several months, we then use a ranking of these clusters relying on their participation in ego’s posts in the given time slot. The changes in these rankings give information about the dynamics of their importance among ego’s network (appearance of new groups of friends, turn-over in ego’s active relationships, increase or decrease of the interaction with ego, etc.)
Once significant changes have been detected in the network of one specific ego, a thorough investigation of the timeline with keywords appearing in posts and comments gives qualitative elements on ego’s important life events (new partner, job change, birth of a child, serious health issue, etc.). Then these events can be related to the qualitative changes not only at the cluster level, but on the detailed structure of the network of alters.
Now processing thousands of such networked timelines (only at the cluster level) enables a categorization of structural changes, also combined with simple sociodemographic variables. This double approach (both egocentered and on many individuals) uses a mixed method: quantitative when computing structural changes in networked timelines and categorizing these changes among thousands of individuals; qualitative when investigating some specific individuals’ timelines to understand the events that may explain the structural changes.