A Co-word Machine?Noortje Marres and Carolin Gerlitz(Goldsmiths, University of London)Co-Word Workshop, May 23-25 2012
Co-word analysis• Network analysis of textual data:tracing relations between terms.• Co-occurrence & weighting of wordpairs in each others’ vicinity (3, 4, 5words)• Detect emergence of new significantthemes, and/or their fluctuations overtime (Callon et al, 1983; Danowski,2009)• Methodological strategy to studyhappening content.• Existing tools: Infomous; Wordij.
#1 Relevance vs. Co-word• Critique of citation and relevanceanalysis (Small, 1973; Callon et al,1983).• Tyranny of relevance and popularity.• Results in performative re-production ofpopularity ranks.• Static measure from the outside.• Aggregation and one middle: creates aEuclidean space.• Two-dimensional ordering: low or highrelevance.
#1 Relevance vs. Co-word• Co-word analysis developed in 1980sas critique of citation analysis, buildingon co-citation analysis• Focus on internal, context specificrelations: measures from the inside.• Expands focus from currentlypredominant clusters to variations overtime.• Allows to explore emergence, or‘pockets of innovation’ (Callon, 1983).• Multi-dimensional, topological orderingacross various clusters.
#2 Post-social method• Moving beyond a purely social (citation/mention/link based) ordering.• Co-word analysis as post-social method:Gives consideration to interplay of actors,devices, issues and formats (Knorr-Cetina,1997; Latour and L’Epinay, 2008).• Specific, context-based, from the inside.• Deploys analytical capacities of devices,platforms and internal relations.
#2 Concerns: Datareduction• How to reduce relations to settleupon significant ones (Kostoff, 2003)?• Distributed analytical capacities.• Indexing - the ‘indexer effect’Kostoff, 2003).• Word-pair frequency (Danowski).• Working with predefined terms:Word-grabber (Danowski).• How to balance between reductionfor significance and liveliness?• Machine-settings: weighting,distance, frequency, breadth, depth?
#2 Concerns: Platformdependency• Post-social methods as drawing onanalytical capacities of devices.• But: How can we deploy co-wordmethods to study issue relations ratherthan analysing platforms themselves?• Strategies to reduce platformdependency whilst taking technicity ofcontent and medium-specificity intoaccount.• Demarcation of source sets: notscrape platforms but issue-relatedsource sets. (deploying the ‘indexereffect’ to positive effect?)• Machine-settings: source setdefinition.
#2 Concerns: Technicity ofcontent• Co-word analysis as measurement frominside the medium.• Focus on socio-epistemic metrics.• Take medium-specificity into account: nodis-embedding of content from contextand medium (Niederer and van Dijck,2010).• Define key research questions:New emerging terms?Heating up of terms over time?Other dynamics of issuefication?Fluctuation of terms?• Possibilities for issue-profiling (detectingthe partisanship not only of actors but ofissues).
#3 Live research• Analytic and empirical focus on variationover time.• Specific response to ‘real-timeresearch’ (monitoring live media for currency).• Shifts attention from currency & liveness toliveliness and happening content (Marres andWeltevrede, 2012).• Live research: moves beyond mere in/decrease of popularity to changes of internalcomposition of co-word networks.
#3 concerns: Static vs. Dynamic• Identification of both static, stable termsand emergent, variable ones.• Static vs Dynamics as thematic of socialtheory (social order vs social change) - co-word was originally on the dynamic end ofspectrum.• Happening of content as interplay betweenstatic vs. dynamic?• Mapping static terms as definition of issuecontexts?• Intervals: When do terms become static?• Machine settings: search for emergent/stable, determine intervals for stability.
#3 concerns: Intervals• To study variations over time, a specific formof longitudinal analysis need to be defined.• Cross-sectional, interval-based, continuousaccount of time (Uprichard, 2012)?• What forms of analysis do different temporalintervals allow for?• How to visualise longitudinal data and how toaccount for non-linear dynamics?• Machine setting: interval selection, outputformats, visualisation guidelines.
Summary• Co-word analysis as tracking theformation and variation of issue-formingword-associations.• Post-social method: Re-embedding issue-term analysis in context and medium.• Live research exploring ‘happeningcontent’.• Measurement from the inside.• Challenges: data-reduction &determining significance, balancingintervals, considering static & dynamicterms and exploring modes ofvisualisation.