This section reflects on the theoretical model of Big Social Data which was developed during the PhD project, namely the Social Interaction Model. Hence, it discusses the differences to the existing Social Data Model, the potential overfitting of the Social Interaction Model to the Social Set Visualizer, and the influence of Facebook datasets on model development.
6.4.1 Differences to Existing Social Data Model
The Social Interaction Model is compared with the existing Social Data Model. Key differences between both models are presented and discussed.
Introduction of Interactions
In applied Social Set Analysis, a notion of interactions is needed. This consists of a data structure comparable to a linked, timestamped list of one initial Action performed by a social media Actor onto a target social media Actor, and zero or manyReactions to that initial action. Therefore, the Social Interaction Model strives to provide a simple, unified data structure for Social Set Analysis, inspired by Big Social Data research in real-world scenarios.
Focus on Social Interactions
The proposed model focuses on the core of socio-technical interactions between hu-man beings over the internet. The previously established Social Data Model puts the theme of Conversations on the same level asInteractions, whereas the Social Inter-action Model argues that Conversations are resulting from Interactions. Therefore, emphasis of the theoretical model should be placed on the concept ofInteractions.
Due to the depriorization of Social Text and Conversations, it needs to be dis-cussed whether the Social Interaction Model is going too far. By elevating the notion ofInteractionsto a core principle of the model without proving much rationale for this step, apart from a healthy intuition that it simplifies the Social Set Analysis research approach and makes it easier to implement in an IT artifact. Future research might underline that the Social Interaction Model depicts a sub-model of the Social Data Model, with no means to replace the existing model as a whole.
Definition of Spatial and Temporal Dimensions
To formalize Social Set Analysis methodology, two dimensions are needed, one di-mension for time and one didi-mension for space. The formal definition of the existing Social Data Model can be extended in order to support these two dimensions of Location in Space and Location in Time. This set-enabled dimensionality has been included in the formal definition of the Social Interaction Model.
The introduction of temporal and spatial dimensions as foundational components streamlines analytical tasks of Big Social Data under the Social Set Analysis method-ology. Furthermore, it makes the model more opinionated towards the utilization of
6.4. Reflections on the Conceptual Data Model 89 a certain methodology, namely Social Set Analysis. With such a bias built into the model itself, it can be argued that it loses its applicability to other methodological approaches of Computational Social Science such as Social Network Analysis. A counterargument to this observation could be that the original Social Data Model never presented an addition or contribution to Social Network Analysis methodology.
Due to the same publication time as the Social Set Analysis approach, it was always intended to act as catalyst for the application of Social Set Analysis in Big Social Data Analytics.
Addition of Non-Textual “Artifact” Content Types
The existing Social Data Model lacks support for non-textual Artifact content types apart fromConversations, such as images and videos within social data. This is recti-fied by the proposed model through addition ofSocial ImagesandSocial Videos, from which meaningful information can be extracted by utilizing state-of-the-art machine learning approaches such as deep learning.
Therefore, it is not required to conceptually attach information on Topics, Key-words, Pronouns andSentiments to the Conversation domain as seen in the Social Data Model, but this information may be attached toArtifactsof any content type as proposed by the Social Interaction Model.
Unification of Bipartite Social Data Model
The proposed Social Interaction Model unifies the bipartite Social Data Model into one sequential concept which is based on a set-based definition of the social data and interactions. A slight refinement in the definition ofArtifacts essentially enables the proposed model to express that meaning with respect toTopics,Keywords,Pronouns, and Sentiments can be extracted from the Artifact data. With Conversations, now calledSocial Text, depicting one specific type ofArtifact data, we observe that there is no inherent conflict between both models, but rather that the Social Interaction Model depicts a logical extension and generalization of the Social Data Model.
Depreciation of Activities
Activities as defined in the Social Data Model are a vague concept that provides no clear mapping to the real-world datasets used for Social Set Analaysis purposes.
The formal definition ofActivitiesin the original publication of the Social Data Model [Mukkamalaet al.2013] concerns a mapping function fromArtifacts toActivities, and from the presented example, an Activity is a “promotion” of products by a clothing retailer on Facebook, that may span over many Actors, Actions, and Artifacts. Thus, the notion ofActivityin the Social Data Model aims tocapture the goal or intention of a social mediaActor that is behind theirActionto broadcast a certain Artifact to the social network. It is difficult to capture the underlying goals and intentions of an Actor in data for research purposes.
Alternatively, when we take into account the activity theoretic notion ofActivities, in which Activities depict unconscious motivations and fundamental needs rather
90 Chapter 6. Discussion than conscious goals [Kaptelinin & Nardi 2006], it also becomes apparent that in-formation on the unconscious intention of the Actor is very difficult to obtain for empirical studies.
Due to the empirical difficulties in acquiring data which exist for both presented interpretations of the termActivities, the concept has not been included in the Social Interaction Model.
Improved Interoperability between Data Sources
Based on the dimension of Location in Space which is proposed in the Social In-teraction Model, it is conceptually possible to interoperate between Artifacts from different social media data sources. For example, after a dimensionality reduction to the temporal dimensionLocation in Time,Artifactsfrom multiple data sources such as Twitter and Facebook can be grouped and compared for Social Set Analysis purposes.
6.4.2 Overfitting of Social Interaction Model to Social Set Visualizer
The Social Interaction Model presents an incremental contribution to Social Set Analysis, as it extends upon the preexisting Social Data Model. It builds on the learnings gathered throughout this PhD project, and itsspecifications are adapted to the special use cases of the Social Set Visualizer and the Social Set Query Language. Therefore, one could argue that the proposed Social Interaction Model is overfitted to the Social Set Visualizer IT artifact. This argument needs to be inves-tigated in future studies, which will show the utility of the Social Interaction Model and whether it can persist on its own without the Social Set Visualizer. However, the parallel, iterative development of both the theoretical model and the IT artifact during this PhD project is well-grounded in the state of the art. Additionally, all scientific contributions have been reviewed by academic peers.
6.4.3 Model Developed with Facebook Datasets
The formalization of the Social Interaction Model presented in this dissertation arises from exclusive work with Big Social Data from Facebook. Hence, it is possible that the model is too focused on this data source, and has not enough influence from other sources such as Twitter or even upcoming decentralized social networks such as Mastodon to reflect the full theoretical spectrum of Big Social Data. Even though the author is highly confident that the Social Interaction Model is applicable to other datasets, this needs to be further investigated through future research.
6.4.4 Using the Model to Express Other Forms of Online Communication and Collaboration
An application of the Social Data Model to other forms online communication and collaboration is theoretically possible, even though it stretches the definition of Big Social Data as it was utilized in this thesis.
6.5. Reflections on the Domain-specific Query Language 91