We have recently developed a prototype using a social argumentation framework to allow virtual communities to check alternative facts. We use a graph-theoretic representation of an argument and also utilize semantic web and linked data principles in creating an argumentation graph. The creation of the argumentation graph is crowdsourced and mediated by expert moderators in a virtual community. We also discuss some research challenges and future applications of our framework.
So-called "alternative facts" are often shared on social networks and spread like wild re across all sorts of social media. Some people claim certain facts are true while others call them "alternative facts" to indicate they are false. Who is right when considering these potential "alternative facts"?
Fake news, imbibed with alternative facts, is an age old
issue not only in politics but communication in general. In
fact, many social scientists [
Just as with propaganda and misinformation, these
socalled "alternative facts" have to be examined critically
using evidence-based reasoning in the age of the World Wide
Web. Collaborative interaction is seen as one of the keys for
developing critical thinking and evidence-based reasoning in
online forums [
We thus use structured discussions to guide users of our
virtual community in critically analyzing proposed
alternative facts. We utilize fundamental argumentation principles
in a graph-theoretic framework that also incorporates
semantic web and linked open data principles [
Argumentation has been shown to be a natural,
substantiated approach for analyzing the veracity and reliability of
assertions and claims [
Argumentation is a formal way of working through ideas or hypotheses. An argument, in this formal sense, is a structure composed of Stances, Claims, and Evidence. In our argumentation framework, a Stance is the nal conclusion composed of Claims and Evidence, and their associated Sources. Stances are fundamental stands on a topic and can be mutually exclusive, should have cohesive substructures, and are composed of atomic argumentation components (Claims, Evidence, and Sources). A Claim can be directly supported by a Source or have multiple Evidence components, each supported by its own Source. Multiple Sources can support multiple Evidence nodes. Our methodology also incorporates Ratings for each Source and user in the system. Di erent trust, authority, and other attribute dimensions are amalgamated and weighted in a Summary Rating.
We create an Argumentation Graph, GA = (V; E; f ), composed of a set of vertices, V , edges, E, and a function, f , which maps each element of E to an unordered pair of vertices in V . Each fundamental Claim, Evidence, or Source in an argument thus constitutes an atomic argumentation component, va, and is embedded as a vertex in the graph such that va 2 V . The vertices contain not just the component's semantic content, but also the ratings, authority, trust, and other attribute dimensions of each atomic argumentation component. The edges e 2 E contain weights along the various dimensions of trust and authority as well as pro/con positions, while the function f maps how they're connected. Depending on the context of the argument, this graph can be undirected or directed, where the temporal component gives the direction to the directed graph.
In terms of a graph, we therefore see the set of vertices V as the set of Claims, Evidence, and Sources; the set of edges E as a set of links that may connect any two vertices. Each subgraph or path traversal that can be obtained from a graph results in a Stance. There are two ways to represent the stances: one way is by making the Stance another node in GA that is added by the moderators in a top-down manner. The other is to designate each sub-graph as a different Stance. Once he GA is formed, we can form subgraphs which represent the di erent stances we can infer from the argumentation graph where each sub-graph would be a separate Stance. Our approach supports both ways of determining the various stances (what we call top-down vs bottom-up). 2.1
This Argumentation Graph is one component of our threepronged system which also includes the User Interface and the Social Community Network that actually creates and supports the overall argument. We organize the community and system to work together in helping critically analyze alternative facts. Members of this virtual community can take three major roles: Users: the information seekers who ask questions and examine the resulting Stances; Responders: Users who have some degree of expertise or background to add Claim, Evidence, and Source nodes; Moderators: contributors that guide the question and answer ow, matching Responders to new questions, evaluating answers for quality assurance, etc. These roles are dynamic as they may evolve over time, and may be multi-faceted with di erent functions and capabilities.
This general framework can be adopted for speci c applications or communities, whose organization can vary widely. For instance, di erent sites may certify Responders in di erent ways: some might require academic quali cations whereas others would allow the Users to also be the expert Responders. In fact, di erent systems may have very di erent organizations, with some sites not having any Moderators at all while others employing the Users to also help moderate the site.
These roles can thus be generalized for discussions amongst
domain experts as well as for formal classroom settings using
threaded discussions between students and teachers. Our
framework can be applied to everything from analyzing
alternative facts to structuring discussions in online courses.
We have also developed an initial set of metrics to quantify
the structure of these threaded discussions by measuring the
redundancy of posts, the compactness of topics, and the
degree of hierarchy in sub-threads [
Existing social networking sites like Facebook could leverage their current userbase and use our framework to create a Social Community Network for checking alternative facts within Facebook itself. Even new platforms like the recently announced news platform WikiTribune could be represented in our framework.
More work needs to be done, though, as questions remain: is crowdsourcing fact-checking the best way to verify alternative facts? Is our approach an optimal way to do so? How can we best incentivise people to contribute, especially the Responders who actually construct the argument and provide Evidence and Sources. Other questions surround algorithms to optimize expert- nding, Top-down vs bottom-up construction of these structured discussions, and community question-answering techniques which we can incorporate to deal with newly submitted questions. Finally, we will need to nalize approaches for certifying Responders and Moderators and ensure that he semantic web components which constitute the elements of these arguments are searchable on the Web. 4.