keynote talk at its 2014: multilevel analysis of socially embedded learning
DESCRIPTION
An invited keynote talk given at the Intelligent Tutoring Systems (ITS) conference in Honolulu, 2014. Begins with some fun observations about being an academic in Hawaii. Motivated both by my early work studying dyadic interaction with Belvedere and a theoretical view of the multi-dimensionality of distributed learning in socio-technical networks and consequent analytic challenges, outlines a framework called "Traces" that addresses these challenges. Most of the examples are of analysis of Tapped In, a successful online network of educational professionals from 1997-2013. Probably the most comprehensive overview of my research to date.TRANSCRIPT
Multilevel Analysis of Socially Embedded Learning
Dan Suthers University of Hawaii
Supported by the National Science Foundation
“Do you go to the beach all the time?”
No. We do not always go to the beach.
We also go to the mountains.
“Health hazards on Mauna Kea:
Altitude sickness. At the summit
elevation of 13796 feet (4200 m), the
atmospheric pressure is 40
percent less than at sea level …”
Major Motivations and Ideas
! Learning (particularly in socio-technical settings) is a complex and embedded phenomenon
! Multiple theories and levels of analysis are needed
! Distributed and multimediated nature of socio-technical systems present analytic challenges
! Approaches illustrated with my work: ! Generalized concept of interaction and the contingency of acts on their setting
! Abstract transcript and analytic hierarchy
Let’s start with Learning in Socio-Technical Networks …
… and the idea that Learning is “Embedded” in multiple ways.
Learning in Socio-Technical Networks
Agency Who or what is the agent
that learns? ! Individual ! Small groups ! Networks (communities,
cultures, societies)
Epistemologies What is the process of
learning? ! Acquisition ! Intersubjective meaning-
making ! Changes in participation
and Identity The correspondence is not strict. Epistemologies can be applied at local or network levels
How do social settings foster learning?
Based on ! Suthers (ijCSCL 2006)
Levels of Agency and Epistemologies ! Acquisition Epistemologies
Learning as acquisition of information, knowledge or skills ! Local: contribution theory, given/new contract, explanation,
conceptual change, practice of skills, etc. ! Network: weak ties, diffusion theories (contagion theory,
diffusion of innovations) ! Intersubjective epistemologies
Learning as intersubjective meaning-making ! Local: co-construction, collaborative inquiry, group cognition ! Network: knowledge building, communities of scientists
! Participatory epistemologies Learning as changes in social participation and identity ! Local: identity, apprenticeship & mentoring (LPP) ! Network: expansive learning (CHAT)
A Complex Multilevel Phenomenon
Claim: individuals participate in the foregoing forms of learning simultaneously
! One might choose to focus on one form, or ! Grapple with a fundamental question:
How does learning take place through the interplay between individual and collective agency in socio-technical networks?
! Requires coordinated multi-level analysis ! Requires coordinated multi-level theorizing
Learning is Embedded
! Interactionally embedded ! Learning accomplishments are contingent on their interactional setting
! Socially embedded ! Social as source of resources ! Social entity as agent of learning
! Technologically embedded ! Affordances influence processes ! Artifacts sustain practices and activity structures
Analytic Challenges
! Embedded: need to say how activity is contingent on setting
! Multimediated: need media independent unit of interaction, while being media aware
! Distributed: need to unify diverse data streams
! Hierarchical: need multiple levels of theory and analysis
Traces Analytic Hierarchy Addressing (some of) the needs
Activity is distributed across multiple media
" Abstract transcript representation collects distributed events from multiple media into a single analytic artifact, reassembling fragmented record of activity
Local activity is hierarchically embedded in network settings, calling for coordinated multilevel analysis
" Analytic hierarchy that supports multiple levels of description (interaction, mediated associations, ties) and analysis
! Suthers (HICSS 2011) ! Suthers & Rosen (LAK 2011)
Contingencies
Mediated Associations
Uptake Ties
Interaction Affiliations
Uptake and Contingencies
Thanks to NSF, Chris Hundhausen, Laura Girardeau, Nathan Dwyer, Richard
Medina and Ravi Vatrapu
How these ideas developed, picking
up where we left off in about 2003 ….
Example 1: First Encounter of Needs
Distributed Interaction
Motivated concept of “Uptake”
! Needed a cross-media unit out of which to construct analytic accounts of interaction ! Media specific concepts (“adjacency pair”, “edit”, “reply”) are too specific
! Not a new unit, but rather a name given to all such constructs taken collectively
! Generalize beyond interaction as “reciprocal action or influence” to other forms of association
Uptake
Minimal requirement for two acts to form part of an interaction: that the existence of the first act is consequential in some way for the second act:
Uptake is present when an act takes some aspect of a prior act (or event) as relevant for ongoing activity.
Flexible and Broad: Opens up our thinking about how interaction might be accomplished
Example 2: Asynchronous Dyads
! Asynchronously interacting dyads ! Public heath problem with hidden profile materials
! Original study: representational guidance of evidence maps vs. threaded discussion
Example: Asynchronous Dyads
Closeup
Example 2: Interactional Pattern (“W”)
! Information Sharing / Round Trip in Evidence Map ! Subsequent Negotiation in Threaded Discussion
Connecting Uptake to Evidence
Motivations ! “How do you know it’s really uptake?”
! Problem of intentionality but also ! Separate evidence from claim
! Manual analysis is slow ! Sufficiently “objective” evidence would also
be computable ! Action is contingent on its setting in
many (observable) ways: let’s use computational tools to leverage this!
Contingencies
Any observed relationship between events that may evidence how one event may have enabled or influenced other events (acts)
! Include “many metaphysical shades between full causality and sheer inexistence” (Latour, 2005)
! Contingencies record how each act is embedded in a history of interaction and a social and technological setting
Some Types of Contingencies
Media Dependency ei operates on object created or modified by ej
Temporal Proximity ei took place soon after ej
Spatial Organization ei takes place in configurational context created by ej
Inscriptional Similarity ei creates inscriptions similar to those created by ej
Semantic Relatedness The meaning of inscriptions created by ei and ej overlap
Contingencies of ei on ej
(! Suthers Dwyer, Medina & Vatrapu, ijCSCL 2010)
Example 3: Early Contingency Analysis
! Analysis originally undertaken to explain convergence & divergence, but discovered emergence of representational practices
! First automated construction and visualization of contingency graph
(# Medina & Suthers, RPTEL 2009)
Example 3: Asymmetry in Roles
Example 3: Representational Practices
Example 3: Episodic View of Interaction
Abstraction to uptake between episodes of specific acts
Example 3: Multi-level Analysis Lemke: "look at at least one organizational level below
the level we are most interested in (to understand the affordances of its constituents) and also one level above (to understand the enabling environmental stabilities)"
Traces Analytic Approach
Testbed: Tapped In SRI’s Network of education professionals: PD and peer
support (Mark Schlager, Patti Schank, Judi Fusco) 1997-2013: longest running educational online
community ! 20K educators/year ! 800 user spaces ! 50 tenants ! 40-60 volunteer-run
community-wide activities/month
! Chats, discussions, wikis, resource sharing ... Good Testbed: Heterogeneous network of diverse small
groups interacting with multiple media
Automatic Discovery of Distributed “Sessions” and Influences Between Sessions
Overview of Analysis: Process Trace
Overview of Analysis: Events
Overview of Analysis: Contingencies
Overview of Analysis: Uptake
Overview of Analysis: Sessions
Process Model
XML Scripts driving Java, NLTK, iGraph <!-- ========== Content Preprocessing ========== -->!
<step bundlename="apps.analyzer" classname="apps.analyzer.script.PythonScriptStep" >!!<stepconfig scriptref="nltk/lancaster_stemmer.py" />!
</step>!
<!-- ========== Contingencies ========== --> !<step bundlename="apps.analyzer"
classname="apps.analyzer.script.ReadDiscussionMessageRule" />!
<step bundlename="apps.analyzer" classname="apps.analyzer.script.LexicalRule" />!<step bundlename="apps.analyzer" classname="apps.analyzer.script.ReplyRule" />!<step bundlename="apps.analyzer" classname="apps.analyzer.script.AddressRule" />!
<step bundlename="apps.analyzer" classname="apps.analyzer.script.SameActorRule">!!<stepconfig windowsize="300" tag="SA300" />!
</step>!<step bundlename="apps.analyzer" classname="apps.analyzer.script.TimeWindowRule">!
!<stepconfig windowsize="120" tag="TW120s" />!</step>!
<!-- ========== Activity Structure (Finding Sessions)========== -->!<step bundlename="apps.analyzer" classname="apps.analyzer.script.ActivityRule"> ! !<stepconfig graphName="activity">! ! !<weighter fileref="weights/activity_weights.xml" />! !</stepconfig>!</step>!. . . !
Case Study
Analyze 3 days of chat, centered on a session of
interest
Teaching Teachers Session 184 23:35 Mary: are all good teachers good mentors? 185 23:38 Amber: some people will take a while to get to that point 186 23:42 Amber: No..not all 187 23:51 Erica: definitely not 188 23:55 Lara: Training can help, but I think some is personality 189 24:09 Amy: some people are excellent teachers but are horrible mentors 190 24:09 Erica: some great teachers can not hold a decent conversation with an
adult 191 24:11 Amber: i had to co-ops who would be awful mentors 192 24:24 Lara: Nods 193 24:27 Dianne: That is an interesting question Maria, ... I would probably say
'yes' first off, and then wonder some more 194 24:42 Mary: it is something I have thought about often Lisa 195 24:47 Amber: I think its alot of personality 196 25:17 Dianne: one thing a mentor has to know is how to operate with a peer,
and how to be intentional about handing over, or encouraging greater independence
197 25:18 Mary: observation has made me think that it takes an extra “special ingredient” to tip the scales
198 25:34 Erica: I think if you have the passion for teaching you will want everyone else to feel the same
199 25:35 Amber: agree
Contingencies computed
! Time Window (recency): all chats within 120 seconds
! Last Contribution: last chats by same actor in 300 seconds
! Address: Actor chats ... chat addresses actor ! Reply: Chat addresses actor ... actor chats ! Lexical Overlap: weighted count of overlapping lexical items (NLTK Lancaster Stemmer)
Weighted sum of counts of above $ estimate of uptake
Uptake Graph for 3 Days of Chat
Rooms
One session across
two rooms
Two sessions in one room
Sessions (Modularity Partitions)
One session across
two rooms
Two sessions in one room
Session Partitions Collapsed
Inspect group in Data Laboratory
Teaching Teachers Session Begins
Teaching Teachers Session Ends
Sociogram
Folding contributions by actor to expose actor-actor uptake
Session 74, Rooms
Session 74, Contributions Colored by Actor, ForceAtlas2 Layout
Can we characterize “good” sessions by structural patterns?
Nodes are contributions, Colors are actors, Node size is weighted indegree
Session 74 Sociogram
Nodes are actors Node size is weighted indegree
Selecting second group
Same actors in NTraining Session
How “Communities” are Embedded in Technological
Media Mediated Associations and
Community Detection
! Suthers, Fusco, Schank, Chu & Schlager (HICSS 2013)
Contingencies
Mediated Associations
Uptake Ties
Interaction Affiliations
Characterization of Community Structure
! “I don’t know what communities are there” ! Organizational “tenants” and unsponsored ! Multiple, fluid forms of participation
! An empirical matter ! Don’t assume that the network is one community
! Don’t assume that external communities are replicated within the sociotechnical system
Communities: Technologically Embedded ! Multiple technologies for participation, each with their
own interactional and social affordances ! Choice of technologies reflect and reaffirm the
relationship between interlocutors (Licoppe and Smoreda, 2005)
! Apply this idea to collective rather than dyadic level: Communities are embedded within and make use of technological media for interaction in ways that reflect and reaffirm their nature
! Our approach identifies cohesive subgroups of actors and of actants (mediational means) simultaneously
! Suthers & Chu, LAK 2012
Intermediate level of representation
! Actor-Actor ties: useful abstraction, but hide
how enacted
! Intermediate granularity: mediated association
! Interaction traces (e.g., contingency graphs): overwhelming detail!
Portion of an Associogram
actors
discussions
files
Cohesive subclusters in associogram
Modularity Partitioning • 234 Partitions • Modularity: 0.828 Open Ord Layout in Gephi
Cohesive subgraphs of actors and artifacts via which they interact
Interpretations of Top 6 Partitions
After School Online Events
Associations via TI
Reception and other public
rooms
Chat-based CoP in a
Midwestern school district;
Discussion-based
professional development in the Southern
US
Chat-based Language
Arts in the US Midwest;
Pre-service program in Western US
Myriad of Small Clusters
Size distribution of Largest 86 partitions
Average weighted degree by actor size
(sample of every 10th partition)
Artifact/Actor ratios by actor size
Use of media in large and small partitions
! Tenant and unsponsored are similar in large partitions
! In small partitions, tenants are strongly chat based while unsponsored rely on asynchronous media
Summary & Comments ! Purely structural (graph theoretic) computations
identified cohesive subgroups that have interpretations as communities
! Diversity demonstrates vibrancy of Tapped In as “transcendent community” (# Joseph et al., CSCL 2007)
! Value to learning analytics: identify social units that are the setting or agent of learning
! Can “dive in” to examine activity of high-degree actors, structure of chat sessions in rooms, etc.
! Need algorithm for overlapping cohesive clusters ! Clique percolation fails on bipartite graphs ! Edge communities and flow compression promising
! Suthers, Fusco, Schank, Chu & Schlager (HICSS 2013)
Productive Multivocality
Bringing multiple theoretical and methodological traditions
to bear
Productive Multivocality Project
The complexity of learning requires multiple analytic “voices” (theories and methods): How to bring them into productive dialogue?
! 5 year project sharing/comparing approaches to analyzing collaborative learning
! 37+ researchers analyzed 5 corpora
! Suthers, Lund, Rosé, Teplovs & Law (Springer 2013)
Strategies for Productive Multivocality
! Dialogue about the same data, from different perspectives
! Share an analytic objective (e.g., “pivotal moments”)
! Bring analytic representations into alignment with each other and the original data
! Eliminate inconsequential differences and Iterate
! Push the boundaries of traditions without betraying
! Reflect on Practice: dialogue about methods as object-constituting, evidence-producing and argument-generating tools
Brief Comments on Design
Mediating between individual and group
Individual !" Small group
Representational affordances for intersubjective meaning-making:
! (Im)Mutable Mobiles ! Negotiation Potentials ! Referential Resources ! Reflector of subjectivity (awareness) ! Persistence (reflection)
! Suthers & Hundhausen (JLS 2003) ! Suthers (ijCSCL 2006)
Individual !" Network
! Joseph, Lid & Suthers (CSCL 2007)
act
persist
find
care
care
act
persist
find
care
act
persist
find persist
find
care
act Thanks to Viil Lid
for diagrams
Key Ideas
! Learning is interactionally embedded % Contingency and Uptake analysis of sequential structure
! Learning is socially embedded % Empirically identify the social units in a STN
! Learning is technologically embedded % Identify the mediational means (mediated associations)
! Generalized concepts, abstract transcript, and analytic hierarchy help
Summary of Concepts ! Mediation and Associations
! All interaction is mediated; actors are associated via media ! Understand how social phenomena are technologically
embedded (! Licoppe & Smoreda, SN 2005; ! Suthers & Chu, LAK 2012)
! Uptake: (! Suthers, ijCSCL 2006) ! Taking some aspect of (the trace of) a prior act or event as
relevant for ongoing activity ! A generalized unit of analysis for “interaction” broadly
understood (multi/cross-media; inter/intra-subjective) ! Contingencies: (! Suthers Dwyer, Medina & Vatrapu, ijCSCL 2010)
! Manifest relationships between acts and their setting (including other events)
! Evidence for Uptake
Traces Analytic Hierarchy
" Abstract transcript representation that collects relevant events into a single analytic artifact
" Analytic hierarchy that supports multiple levels of analysis
! Suthers, HICSS 2011 ! Suthers & Rosen, LAK 2011
Contingencies
Mediated Associations
Uptake Ties
Interaction Affiliations
Mahalo!
Dan Suthers, Dept. of ICS [email protected] Lilt.ics.hawaii.edu