2.3 Theoretical Models of Big Social Data
2.3.4 Social Interaction Model (2018)
In 2018, I presented an addition to the existing theoretical model of Big Social Data, developing the Social Interaction Model. The Social Interaction Model extends the existing Social Data Model through proposing of a radically simplified concept, which is formally grounded in set theory and relational algebra. The Social Interaction Model is included asPublication V[Flesch2018] of this dissertation and contributes to the state-of-the-art theory of Big Social Data.
The conceptual foundation of the Social Interaction Model is illustrated in Fig-ure 2.5. The Social Interaction Model distinguishes between three major components of social data: Actors, Interactions and Artifacts. These core components of the Social Interaction Model show strong commonalities with the Social Data Model.
The Social Interaction Model incorporates several foundational concepts from the Social Data Model, namely Social Data, Interactions, Actors, Actions,Artifacts, Con-versations,Topics,Keywords,Pronouns, andSentiments. At the same time the Social Interaction Model extends it with novel concepts of Reactions, Social Video, Social Images, and Social Text, which are displayed with thick borders in Figure 2.5. It reorganizes the two pillars ofInteractions andConversations proposed by the previ-ous Social Data Model into a streamlined top-down flow, where Social Text, which is renamed from the former concept of Conversations, is extended by Social Video and Social Images. The Social Interaction Model expresses that all three types of Artifacts, namely Social Video, Social Images, and Social Text, can be analyzed for
2.3. Theoretical Models of Big Social Data 25
Big Social Data
Interactions
Actors
Actions Reactions
Artifacts
Social Videos Social Images Social Text
"Conversations" in Social Data Model
Topics, Sentiments, Kewords, Pronouns
Figure 2.5: Social Interaction Model
aspects of Topics, Sentiments, Keywords, and Pronouns, which were previously in-troduced in the Social Data Model but limited to the concept ofConversations.
Furthermore, in Figure 2.5the use of boxes vs. ellipses displays the hierarchy of concepts used within the old two-tier Social Data Model. Boxes depict a higher rank in the hierarchy, whereas ellipses depict a lower rank. The top-down flow of the Social Interaction Model emphasizes that a more fine-grained hierarchy is needed, e.g. more than two levels as in the old model. Hence, former hierarchies are only illustrated for the reader to better understand and compare the transformation of the Social Data Model.
In the Social Interaction Model, Actorsdepictany kind of user or entitythat can be interacted with in the realm of social media. Each Actor consists of a unique Location in Space and a set ofArtifactswhich depict the actor’s attributes, i.e. name, date of birth, profile picture or self-descriptive bio text. Actors depict both origin and destination of social-technical interactions.
An Interaction consists of one initial Action and zero to many Reactions that respond to the initial Action or one of its Reactions. In this model, Actions always occur between two Actors, one originating Actor and one receiving Actor. As Ac-tions are always directed at other Actors, they follow the concept of other-directed or "transactive" actions [Berkowitz & Gibbs 1979]. Consequently, Reactions always originate from oneActor and are targeted at anotherActionor Reaction.
Moreover, every singleActionorReactionspawns a newly created set of Arti-facts. Artifactscan be any kind ofuser-generated content, such as text posts or
com-26 Chapter 2. Research Methodology ments, media uploads, profile pictures, et cetera. When Actors want to interact with an Artifact, they are limited to referencing the Actions or Reactions which spawned the Artifact in question from a newly-created Reaction. By definition, Artifacts de-pict only the result from an Action or Reaction. Each individualArtifact consists of a certain content type and a user-generated payload. The conceptual model pro-posed in this thesis supports three content types, them being Social Videos, Social Images andConversations. Those sets can be aggregated on the specificInteraction, resulting in a set of all Artifacts created during a certain Interaction. Interactions within this proposed conceptual model depict interactions in line with the theory of socio-technical interaction [Vatrapu 2010].
Through computational analysis of individual Artifacts, further information re-garding Topics, Keywords, Pronouns, and Sentiments can be extracted. This can be achieved by means of machine learning, e.g. for sentiment analysis [Boiy &
Moens 2009,Neethu & Rajasree 2013], or more specialized techniques for text analysis [Abbasi & Chen 2007,Abbasi & Chen 2008,Abbasiet al. 2013].
Theformal definition of the Social Interaction Modelis illustrated inFigure 2.6.
Temporal and spatial dimensions are included as a core component of the proposed model, therefore streamlining data analytics tasks in the realm of Social Set Analysis.
These temporal and spatial dimensions have also been utilized in the prior theoretical model of interaction in socio-technical networks by [Suthers et al. 2010,Suthers &
Rosen 2011], but not in the two versions of the Social Data Model. Furthermore, the Actions and Reactions in the Social Interaction Model are a specialized case of the
“uptake” concept introduced by [Sutherset al.2010] for the realm of Big Social Data.
Within the Social Interaction Model, the context of eachInteraction refers to its
Actors
Artifacts
Artifacts
Actions Reactions
Time Location in Space
t1 t2
t3 t4
β
1: β
2: β
3: β
4:
α
1α
2α
3α
4Aα1 Aα2 Aα3 Aα4
lα1 lα2 lα3 lα4
α
1α
4α
2α
3α
4β1(α1, α4)⇒ Aβ1
β2(α4, β1)⇒ Aβ2
β3(α3, β2)⇒ Aβ3
β4(α2, β3)⇒ Aβ4
Figure 2.6: Formalization of the Social Interaction Model
2.3. Theoretical Models of Big Social Data 27
Location in Space and Time, thus allowing analytics along spatial and temporal dimensions. Location in Space is provided by Actors who are part of the initial Action. As by definition everyActor exhibits a certain Location in Space. Therefore, theLocation in Spacefor the whole interaction is set by the location attributes of the initial Actionof eachInteraction. Location in Time is attached to every singleAction andReaction as a conventional time stamp.
For comparison across the dimension of time, a set-based intersection of various time periods along the time axis is calculated. Similarly, for comparison across loca-tion in space,set inclusions and exclusionsare calculated based on the dimension of space, for example a list of relevant Facebook walls. Based on this established pro-cedure, we can design a theoretical model that fits to analytical Social Set Analysis methodologies as they are employed in various case studies.
The reasons for a conceptualization of the Social Interaction Model as a gener-alized theoretical model of interactional Big Social Data are manifold. First, it is to extend upon the core ideas put forward through the Social Data Model. Secondly, it is to apply learnings from a series of set-based Visual Analytics studies performed during my PhD project using the Social Set Visualizer software tool. And lastly, it is to improve the existing theoretical data model in which the Social Set Analysis approach to Big Social Data Analytics is grounded.
Accordingly, the Social Interaction Model introduces a set-based definition of Interactions and the resulting Artifacts within Big Social Data. It proposes a two-dimensional theoretical framework based on the two concepts of location in space and location in time. Such a framework provides additional theoretical coherence with the empirical application of set-based analytics as presented throughout recent publications by our research group [Vatrapuet al.2014] and the multiple publications on this topic included in this dissertation. Therefore, the concepts of the previously utilized Social Data Model are refined, formalized and simplified.
Learnings from various computational implementations during the course of this thesis are incorporated in the Social Interaction Model. Key differencesbetween the existing Social Data Model and the Social Interaction Model present the inclusion of non-textual artifact content types, the unification of a previously bipartite Social Data Model, the depreciation of an empirically vague notion of Activities, and the improved interoperability between different sources of Big Social Data.
In order to simplify the theoretical model for practical utilization in Big Social Data Analytics,ActionsandReactions arelimited to one-on-one relationships. This is grounded in the nature of the most widely used social media platforms, which also limit the structured flow of discussions to one Action or Reaction at any single point in time, therefore not natively supporting one-to-manyActions or Reactions.
Due to this restriction, a tree-like data structureemerges from the formalization of the Social Interaction Model as one initialAction exists for eachInteraction, with one-on-oneReactions developing a tree structure from this initialAction. This struc-ture of the model reflects the strucstruc-ture of the Big Social Data that was analyzed in the course of the different projects und publications that were part of this thesis.
Due to the filtering along the spatial and temporal dimensions of the data at hand,
28 Chapter 2. Research Methodology initial reference points for the analyzed interactions are established on which the tree structure crystallizes.