Knowledge Building Discourse

Knowledge Building (KB) as a pedagogy treats learning as knowledge creation not qualitatively different from what happens in knowledge-generating organizations (Bereiter & Scardamalia, 2014). It is conceptualized as an interactive system involving epistemic agents (e.g., students, teachers), knowledge objects, and sociocultural practices (Chen & Hong, 2016), with KB principles (Scardamalia, 2002) explicating the relations among them to distinguish KB from other pedagogical approaches.

Activity Theory (AT) has been previously applied to analyzing KB discourse (Hewitt, 2004; van Aalst & Hill, 2006). Originating from Vygotsgy’s work, AT attempts to bridge the space between subjects (e.g., students) and objects (e.g., tasks, problems of understanding) by recognizing various mediational means in between, i.e., tools, rules, community, and division of labor (Cole & Engeström, 1993). Compared to “quantified content analysis” widely applied to the analysis of KB discourse, the activity theory framework could afford a richer description of KB discourse, “because it accounts for both individual and communal activity, as well as multi-directional movement of individuals within the community” (van Aalst & Hill, 2016, p. 25).

Figure 1 illustrates the activity system of KB discourse. For instance, in a KB classroom using Knowledge Forum (Scardamalia, 2004), one widely used KB discourse environment, elements of the system could be defined as following (Hewitt, 2004; van Aalst & Hill, 2006):

Figure 1

Figure 1

Meta-Network

Network analysis has been widely applied in KB and, more generally, CSCL research. One direction, therefore, is to explore more advanced SNA techniques—more sophisticated indices, two-mode networks, temporal networks—in appropriate research contexts. Pertinent to the use of more advanced techniques, another direction is to seek integration among different analytical perspectives.

Dynamic network analysis (DNA) offers an opportunity to advance in both directions. As a network analysis approach, DNA “combines multi-level, multi-mode, multi-link social network analysis with cognitive science and multi-agent simulation to provide a methodology for modeling the dynamics of complex and adaptive socio-technical systems” (Schreiber & Carley, 2005, pp. 1-2). In particular, the meta-network under DNA offers an ontological framework to reference the entity classes in a socio-technical system including agent, knowledge, resource, task, organization, and location (Carley, 2002; Carley, Diesner, Reminga, & Tsvetovat, 2007/8; Carley & Hill, 2001). This meta-network approach can be combined with a theoretical framework like Activity Theory to enable a theory-driven representation of complex systems. As a matter of fact, DNA has the potential to introduce analytical power sought by (Engeström, 2009), by combining analysis of both hubs, which are traditionally the focus, and trails, which can be captured by richer trace data. DNA is now supported by a sophisticated set of tools developed by Carley and her team (Carley et al., 2007).

Knowledge Forum -> Graph Model

Guided by the operationalization above, we further constructed a “graph model” to bring various types of entities and connections in Knowledge Forum into light.

Knowledge Forum -> Graph Model

title: false

Network Representation & Queries

We then built an analytic system, using a state-of-the-art GraphDB technology named Neo4j, to represent the discourse data for dynamic queries. After the discourse data were transformed into a Neo4j database, Cypher, a graph query language provided by Neo4j, was used to dynamically extract data from the database.

A Case / Demo

The case was from a Grade 4 class, with 22 students, from a K-6 school in Toronto. The class was studying “Light” over an extended period of time, with support from Knowledge Forum (KF). In total, students wrote 380 notes across 8 KF views.

Build-on Networks

12345678910111213141516171819202122Building-on network (1 unit aggregation)t=1-2
Animation Duration
02468
About NDTV-D3
0: 1-2
510152025

Co-authoring Network

12345678910111213141516171819202122Co-authoring network (1 unit aggregation)t=1-2
Animation Duration
02468
About NDTV-D3
0: 1-2
51015202530

Reading Network

12345678910111213141516171819202122Reading network (1 unit aggregation)t=1-2
Animation Duration
02468
About NDTV-D3
0: 1-2
1020304050

Co-locating Network

12345678910111213141516171819202122Co-locating network (1 unit aggregation)t=1-2
Animation Duration
02468
About NDTV-D3
0: 1-2
510152025

Co-locating Network – Timeline

Co-locating Network – Avg. Degree

Befriending Problems

33Flash lightWhat are Rainbows made out of?colours of things colours of thingsWhat does a rainbow have to do with light?absorbing lightlight hitting lighthow things go Different colours of lightWhy are shadows sometimes small and sometimes big? how are shadows created shadowsCan a flashlight shine through many layers of cellophane? why do you see yourself in a mirror?oxyge and stars The shape of the rainbowhow are rainbows made how are rainbows made?the northern lights!RAinbowshow does light bounceHow does the eye workDifferent colours,different resultsHow do lenses work?Seeing lightLong and short shadowswhat are concave mirrors used for?The colours of shadowshow does light travel?How light travelsHow do you see colours.What do they do.Befriending networks - Author-Problem (1 unit aggregation)t=1-2
Animation Duration
02468
About NDTV-D3
0: 1-2
51015

Mediatoin Network: Painting–Word

painting 165painting 164painting 265painting 187painting 170painting 171painting 184painting 259painting 191painting 192painting 193painting 194painting 185painting 186painting 175painting 179painting 304painting 201painting 202painting 203painting 225painting 226painting 291painting 292painting 262painting 263painting 222painting 280painting 264painting 328painting 329painting 330painting 331painting 268painting 269painting 276painting 289waterlightshadowpencillookmakegreenredcolourbluecolorgetconematerilikemirrorreflect3objectuseconcavwouldprimarinewtonmixprojectscreenturnexperiknowwanthappen13yellow03waveretinafocusseefarnearsightglasslentravellinestraightshutterpicturMediation networks - Painting-Word (1 unit aggregation)t=1-2
Animation Duration
02468
About NDTV-D3
0: 1-2
5101520

Division of labor: Scaffold use

15182736172230342433282521192035261632233129My theoryI need to understandNew informationDate of ExperimentTitle of ExperimentMaterialsProcedureResultsMy problem of understandingTheory/HypothesisDiscussionOur UnderstandingWhat we still do not understandOpinionDifferent opinionReasonElaborationEvidenceExampleMy understanding of LightPutting our knowledge togetherA better theoryScaffold use networks (1 unit aggregation)t=1-2
Animation Duration
02468
About NDTV-D3
0: 1-2
102030

Author–Problem–Word

182123biggerdiffergetlightmakeshadowsmallertouchtreeWhat is a shadow?Author--Problem--Word networks (1 unit aggregation)t=1-4
Animation Duration
02468
About NDTV-D3
0: 1-4
20406080100

Future Work

The agenda in this paper is mainly twofold. First, building on earlier work in the KB community (Hewitt, 2004; van Aalst & Hill, 2006), we argue for the conceptualization of KB discourse as an activity system to better capture its complex dynamics. Second, we propose a dynamic network analysis (DNA) approach to analyzing KB discourse in light of such conceptualization. To this end, we build an analytic system using recent GraphDB technologies and demonstrate in this paper ways to inspect the discourse from different angles, without relying on qualitative analysis as activity theory researchers normally do. This work is at its nascent stage and is currently descriptive in nature, but we believe it is moving towards providing us dynamic ways to characterize different dynamic aspects of KB discourse.

Future work is needed at both the conceptual and technical aspects. First, we need to work on deep integration between KB discourse and activity theory. We also need to find ways to capture more sophisticated interactions among elements of the activity system (Hewitt, 2004) using the DNA approach. Finally, given continual innovations of network analysis, especially more advanced techniques in two-mode networks (Opsahl, 2013), we plan to derive and validate more advanced measures of networks extracted using the DNA approach. Hopefully we can surpass the current stage of descriptive measures, and towards being able to make claims about the “health” of KB discourse based on validated measures.