Ontology-Based Roles Association Networks for Visualizing Trends in Political Debate

Online resources, large data repositories and streaming social network messages embed plenitudes of interesting knowledge, often of associative nature. A specific communicative context, such as the political debate in a given country, has groupings of actors, with changing attitudes and stances towards each other and external, real or invented, threats and opportunities. A new form of associative network is introduced, that integrate flexible ontologies for complex contexts of roles and hierarchies with a labelled association structure representing observed strengths and attitudes. Twitter messages from the political landscape in Denmark up to the general election 2015 are used as a both current and relevant illustrative case. 1 Motivation and Background @bansoe And so it begins....#fv15 This tweet, like many others on May 27th 2015, initiated the Danish national elections on the social media platform Twitter. This paper outlines a graphical representational formalism that allows modeling how the election and its main actors have represented various subjects in the debate on Twitter, and with what sentiment. Twitter in Denmark is a primarily political (as opposed to a social) platform [1, 2] and as both media and politicians is very aware of this, the platform can generate a large amount of data very fast when there is an election. This is primarily due to the nature of the Twitter platform making it highly suited for breaking news [3] and thus rapidly developing events and attitudes. This is especially true when an election is in process, where tweets from political actors mostly are used to campaigning, spreading information or self promotion [4]. For many end users and indeed both media professionals and politicians, making grasps of such a large amount of data can be cumbersome bordering impossible. Therefore a modeling approach to visualize and understand some of the mechanics and contents of Twitter data in a political and electoral context is very much needed. For this paper we used the open source tool Yourtwapperkeeper 1 #fv for Danish “Folketingsvalg”, the election for the Danish national parliament. to harvest all tweets from the start of the election period in Denmark to the end of the campaign. This tool has proven a viable way of gathering Twitter data which due to API limitations can be hard to do [5]. We selected the appropriate electoral hashtags that appeared to be trending within the first 24 hours. These were #dkpol (Danish general politics) and a number of hashtags all referring to the election (#fv15, #valg15, #valg2015, #drdinstemme, #tv2valg, #ft15, #dkvalg, #ftvalg15 and #dkmedier). This gave a sample of 260,000 tweets on which our methodology is tested. Co-occurrence networks and semantic networks based on co-occurrences have been used in many areas and in different shapes. We propose an integration with an ontological concept lattice to a general model for representing complex contexts including large sets of actors with many different roles and organizational groupings – as is the case in politics. The concept lattice allows to zoom in and out, from single individual to, say, the known members of a given political party, and vice-versa, and to express asymmetric and annotated relationships, anticipating different visualizations. 2 ORAN: Ontology-Based Roles Association Networks Ontology-Based Roles Association Networks, for short ORANs, are association networks for context representation, whose nodes are concepts taken from a concept lattice serving as an ontology. Concepts may be extended with roles, so for example, concept “Smith” may refer to any message relating somehow to a single person, whereas predicating with the role “author”, forming concept “author:Smith” may refer to messages written by that person. Formally, we assume finite sets of atomic concepts A, and roles R; a concept is of one of the forms, a or r:a where a ∈ A, r ∈ R. The concept lattice is a partially ordered set of concepts 〈C, <〉 where < is transitive, and if r:a ∈ C, then a ∈ C and ra < a. The relation < is read “more specialized than”. A concept c ∈ C has an extension, written [[c]], which is a subset of some universe U (e.g., of messages) such that, for any c, c′ with c < c′, it holds that [[c]] ⊆ [[c′]]. The lattice is not fully generative, as not all roles can apply to all atomic concepts, e.g., refugees are often mentioned (concept about:refugees) but do not have voices as authors. An ORAN is a directed or undirected, labelled graph, described as a triplet of a concepts lattice, a subset of its atomic concepts and one or more association types, 〈〈C, <〉, A, T 〉. An association type is of the form P1 → P2 in case of a directed network and P1 − P2 for an undirected one, where P1, P2 are generic concepts of form c ∈ C, ∗ or r:∗, where r ∈ R. The association type generates the actual set of nodes N in the graph and the edges. Concept c indicates c ∈ N , ∗ that A ⊆ N and r:∗ that {r:a | a ∈ A}∩ C ⊆ N . All edges whose endpoints ∈ N match the association type are included, so e.g., author:∗ → ∗may create, among others, author : MembersOfDanishPeoplesParty→ refuges. When drawing these edges, they are typically shown between the given atomic concepts. The edges are referred to as associations, and each edge c1 → c2 or c1 − c2 has one or more labels, one of which is its association degree, also called strength, defined for the directed, resp. undirected case, as follows. d(c1 → c2) = { |[[c1]]∩[[c2]]| |[[c1]]| ; if |[[c1]]| > 0