Knowledge building is nevertheless a social process. Knowledge building has been described in terms of 12 underlying principles (Scardamalia, 2002), each of which is characterized by a series of social and technological determinants (loc: Page 17)

Most recently, the relationship between knowledge building, knowledge building environments, and 21st century skills has been examined (Scardamalia, Bransford, Kozma, & Quellmalz, 2009) (loc: Page 17)

This echoes the take–home message of Collins and Halverson (2009): that schooling must be reformed to meet the needs of society. The relationship between knowledge building and 21st century skills can be elucidated by contrasting a “top-down” approach that starts with identifying goals and desired outcomes and “working backwards” to develop implementation strategies to a “bottom–up” or “emergent” one in which goals are not fixed in advance but rather emerge as thinking progresses. Scardamalia et al. (2009) contrast both approaches and ultimately suggest integrating the two models through a series of investigations including “demonstrating how a broader systems perspective might inform large-scale, on-demand, summative assessment” (p.4, emphasis theirs) (loc: Page 18)

The study of social networks that underlie a knowledge building community is a worthwhile and important endeavour, and at least two recent doctoral dissertations (Palonen, 2003; Philip, 2009) demonstrate the utility of doing (loc: Page 18)

To find effective measures of knowledge building we need to contemplate another type of network analysis - one that identifies relationship between ideas, as well as between people contributing those ideas. And more specifically, we must be able to identify the types of interactions between people and ideas that yield knowledge growth. (loc: Page 19)

Because ideas are the central objects in a knowledge building community, successful communities tend to generate many of them. This perfusion of ideas can be simultaneously generative and paralysing. It is generative in that the expression of ideas often generates further expression of ideas, akin to “brainstorming”. It can be paralysing by virtue of the sheer number of postings that must be read, processed, understood and assimilated. Eliciting blue-sky scenarios from teachers and other practitioners struggling with this sort of information overload include the desire to have computers help them make sense of the diversity of ideas they are confronted with. (loc: Page 19)

n The Cambridge Handbook of the Learning Sciences (Sawyer, 2006), knowledge building is identified as one of five foundations of the learning sciences. It is characterized by six themes that set it apart from other educational models: 1. Knowledge advancement as a community rather than individual achievement; 2. Knowledge advancement as idea improvement rather than as progress toward true or warranted belief; 3. Knowledge of in contrast to knowledge about; 4. Discourse as collaborative problem solving rather than as argumentation; 5. Constructive use of authoritative information; 6. Understanding as an emergent. (loc: Page 22)

Bereiter and Scardamalia (2003) differentiate knowledge building from other constructivist approaches such as Learning by Design (Holbrook & Kolodner, 2000; Kolodner, 2002), Project-Based Science (Marx, Blumenfeld, Krajcik, & Soloway, 1997), and Problem-Based Learning by its focus on creative work with ideas. Scardamalia and Bereiter (2007) also differentiate it from A. L. Brown & Campione's (1994) Fostering Communities of Learners model. Knowledge building is a process of sustained idea improvement fostered by communities in which participants take responsibility for the advancement of community knowledge (Scardamalia, 2004). Perhaps most important is the notion that knowledge building concerns itself with the process of innovation and the creation of new knowledge. Scardamalia and Bereiter (2006) describe the benefits to students in terms of a cybernetic system, highlighting the fact that “the main value is this epistemic one - a feedforward effect, in which new knowledge gives rise to and speeds the development of yet newer knowledge” (p. 99). They hint that assessment - in terms of judging theories and models generated by students - should focus on the generative capacity of such contributions rather than conformity to the status quo. (loc: Page 23)

Assessment is part of the effort to advance knowledge-it is used to identify problems as the work proceeds and is embedded in the day–to–day workings of the organization. The community engages in its own internal assessment, which is both more fine–tuned and rigorous than external assessment, and serves to ensure that the community's work will exceed the expectations of external assessors. (p.12) This principle has been extended to encompass the notion of concurrency: that is, providing feedback to the users in real time (Chan & Lee, 2007; Teplovs, Donoahue, Scardamalia, & Philip, 2007). (loc: Page 24)

A knowledge building environment (KBE), as differentiated from a knowledge building tool, must support a culture of creative work with ideas. A KBE, as Scardamalia (2003) defines it is: “any environment (virtual or otherwise) that enhances collaborative efforts to create and continually improve ideas.” (loc: Page 25)

Improvable Ideas: Knowledge Forum supports recursion in all aspects of its design-there is always a higher level, there is always opportunity to revise. Background operations reflect change: continual improvement, revision, theory refinement. There are at least two types of “background operations”: those undertaken by participants and those conducted in a more or less automated fashion by computing infrastructure. Ideally such automated background operations support creative work with ideas without quelling the willingness of participants to engage in risk-taking with ideas (Scardamalia, 2003). Recent advances in information visualization suggest that we now have available the necessary computing power to create analytic representations that can be used by non–experts. Whereas the computing power may be available, what constitutes a useful and usable visualization is not clear. (loc: Page 25-26)

One way is to highlight those ideas that seem related or linked to other ideas (and are therefore potentially more useful or powerful). Another is to consider the visualizations themselves as improvable ideas. (loc: Page 26)

Coherence-producing mechanisms for dealing with information overload are similarly concerned with the relationships between entries. Ideas can be “multiply tagged, linked, referenced, subordinated, superordinated” and so on (Scardamalia & Bereiter, 1993, p. 39). (loc: Page 27)

One of Shneiderman's (1996) basic tenets of information visualization is the need to provide the ability to zoom and filter. This maps closely onto the notion of “rise-above”. Rather than having notes subsumed in other notes, it might be more powerful to provide a sense of zooming into and out of details provided by risen-above notes. A view–of–views approach, which is a powerful step toward rise–above views, is a good first approximation at the ability to zoom in and out at the view level (Scardamalia & Bereiter, 2006) (loc: Page 27-28)

Storage and retrieval for situating ideas in a communal context (Scardamalia & Bereiter, 1993) stresses the importance of placing ideas “in the context of entries by others” (p.39). Put another way, this feature highlights the design challenge of allowing ideas to find other ideas. (loc: Page 29)

Information visualization can be employed to understand the relationships between views, either within the same database or across apparently disparate databases. It can be used to present areas of overlap between collections of notes. Information visualization can be used as a way in to a database from another community without overwhelming the visitor. (loc: Page 30)

Information visualization has the potential to facilitate the embedding of assessment directly in the knowledge building environment. Benchmarks can be included in the discourse environment and information visualization techniques can be used to indicate links between the benchmarks and student contributions. Information visualization has the potential to allow creative assessment of groups and individual processes, contributions rates and so on. (loc: Page 32)

Scardamalia (2003) enumerates the “knowledge base for the design of knowledge building environments” by drawing on dynamics of externalized, organizational memory (Norman, 1993), collective cognitive responsibility for knowledge creation (Scardamalia, 2002), a developmental knowledge building trajectory that enables the link between learning and knowledge creation, intertextuality (Bakhtin, 1986), cultures of innovation (Drucker, 1985), the objectification of knowledge (Bereiter, 2002), and the improvability of ideas to explain anomalous facts (Lakatos, 1970) (loc: Page 33)

well–designed information visualizations can amplify cognition through perception (Card, Mackinlay, & Shneiderman, 1999). (loc: Page 33)

issue of scalability. In a world where networks or communities of practice are becoming commonplace, how can one judge the promisingness of within– and between–community linkages, particularly given the near–infinite number of such links? In this case, using information visualization techniques based on content can lead to opportunistic linking among and between communities. However, Scardamalia (2003) offers an admonishment against overuse of automated content analysis: that such an approach may curtail risk–taking with ideas (loc: Page 34)

Among the most important lessons learned from several decades of research on KBEs are the following: a KBE can increase opportunities and possibilities for knowledge creation, but it loses that capacity when it becomes overly prescriptive. It must provide a flexibility consistent with the emergent goals of knowledge creation. It should not use prompts, intelligent agents, prescribed project, fixed task sequences, templates or other means to guide users to known endpoints. It must, instead, capture the human capacity for inventiveness and convert that inventiveness into something of social value. (Scardamalia, 2003, p. 271) (loc: Page 34-35)

As van Aalst, Sha, and Teplovs (2010) note, interest in formative assessment was piqued after Black and Wiliam (1998) published a major review of the field. The distinction between summative and formative assessment can be traced to an early seminal paper by Scriven (1967) (loc: Page 35)

She states that whereas summative assessment can exist in the absence of formative assessment, the converse is not true. That is, “formative assessment requires summative judgement to have preceded it” (p.468), even if the summative assessment is implicit. This represents one notion of formative assessment. A more expansive concept would include any return of information that agents would find helpful in improving their practices. (loc: Page 36)

In terms of knowledge building, formative assessment is conceptualized as being concurrent, embedded and transformative (Scardamalia, 2002) (loc: Page 36)

Very thorough and impressive inquiry thread analyses have been conducted by, for example Zhang, Scardamalia, Lamon, Messina and Reeve (2007) (loc: Page 37)

Wasserman and Faust (1997) describe social network analysis as a methodology that focuses on relationships and patterns of relationships. As such it “requires a set of methods and analytic concepts that are distinct from the methods of traditional statistics and data analysis” (p. 3) (loc: Page 37)

list of topics that have been studied using network analytic methods, including community (Wellman, 1979), group problem solving (Bavelas, 1950; Bavelas & Barrett, 1951; Leavitt, 1951), diffusion and adoption of innovations (Coleman, Katz, & Menzel, 1957, 1966; Rogers, 1979), and cognition (Freeman, Romney, & Freeman, 1987; Krackhardt, 1987). (loc: Page 38)

A problem with the de Laat study is that the authors equate participation with learning. That is not entirely unreasonable, as the questions they were interested in answering were less about the nature of the content than about the relationships among the participants. In their study the content analysis was used to categorize messages as either “learning processes” or “tutoring processes”, but the relationship among ideas within those messages was not explored. (loc: Page 39)

Brown and Duguid's (2000) Networks of Practice (NoP) model represents a more generalized case of Lave and Wenger's (1991) Communities of Practice model. Individuals in a Network of Practice may never meet each other but are connected essentially by their ideas even if only as expressed in virtual documents such as electronic mail. In a NoP individuals share knowledge and the emergent social network structure is determined by such sharing. (loc: Page 40)

As with Lave and Wenger, Brown and Duguid focus on knowledge sharing and not on knowledge construction. Nevertheless the notion that information can take on a social life of its own is an important one (loc: Page 40)

Social network analysis can, and has been conducted on the social dynamics of a knowledge building community (Philip, 2009). But there is another equally important type of network analysis to be considered in the case of knowledge building: the network of ideas. (loc: Page 40)

Latent semantic analysis (LSA) represents both a statistical technique and a model of human knowledge acquisition. Landauer and Dumais (1997) propose LSA as a model that could provide a solution to the question of, how do individuals know so much given as little information as they get? (loc: Page 41)

LSA provides a high–dimensional (yet still reduced in dimensions as compared with “reality”) representation of the associations between words and the documents containing those words. The final output from LSA is a series of measures that describe the relationships between words, documents, or words–and–documents (loc: Page 41)

The Hyperspace Analogue of Language (HAL) (Burgess, Livesay, & Lund, 1998; Burgess & Lund, 1997a, 1997b; Lund & Burgess, 1997) provides an alternative vector representation of memory. It is similar to LSA in that it seeks to create vector representations of patterns of co–occurrences of words. It differs chiefly in its use of syntactic elements in addition to purely semantic ones. HAL is also designed to derive “sentential meaning” from texts, whereas LSA is less capable of doing so (Burgess, et al., 1998). (loc: Page 41)

Two transformations are typically applied to the original term–document matrix: a local “log” transformation and a global “inverse entropy” transformation. These transformations have analogies in psychological theories of learning and knowledge acquisition. The contents of each cell in the term–document matrix are replaced with the logarithm of the original value plus one (to facilitate those cases where the original contents are zero). The local (in the sense of being applied to each cell) log transformation serves to amplify the relation between words that co–occur in multiple contexts. The global (in the sense of being applied across the contents of an entire row in the term–document matrix) inverse entropy transformation mimics conditioning effects (Rescorla & Wagner, 1972). It puts more emphasis on words that are not evenly distributed across contexts (loc: Page 42)

The subsequent steps of LSA are not represented in any existing theory of knowledge acquisition. The singular value decomposition (SVD) and dimension reduction serve to condense the original matrix into something that represents a higher–order (or “latent”) semantic structure. It is this latent semantic space that provides a representation of the associations between stimuli (i.e. words or documents, which may represent specific learning episodes). The multi–dimensional representation of stimuli in this space represents a quasi–stable self–configured system analogous to physical objects, weather systems, ecosystems, and Hopfield nets (Hopfield, 1982). (loc: Page 42)

Landauer and Dumais (1997) go on to show that LSA can be used to accurately model word–knowledge acquisition by school–children, and how LSA can provide useful insights into conditioning, perceptual learning, chunking, classical association theory, analogs to episodic, semantic, explicit and implicit memories, the origin of discrete concepts (or words), expertise, and contextual disambiguation. The authors also explicate the relationship between LSA and Kintsch's (1988) Construction–Integration (CI) model of text comprehension. (loc: Page 42-43)

73). Kintsch proposes a model that features a “knowledge net”, which consists of nodes that are propositions, schemas, frames, scripts, and production rules. The links between the nodes form an associative net, and the “meaning” of a node is determined by its position in the net. This definition of “meaning” differs from the psychological definition of meaning, which in this model can be conceptualized as the nodes in the net that are activated at a given point in time. The number of nodes that are active is limited by the capacity of short–term working memory. Hence, the “meaning” of a node will be relatively consistent over time but may vary depending on which nodes have been activated. Thus, this “knowledge net” representation combines features of episodic and semantic memory and procedural and declarative knowledge. It represents a radical constructionist (Bettencourt, 1993) representation of knowledge, since the meaning of any particular term is derived solely by the activation and proximity of other nodes (loc: Page 43)

The representation of knowledge as a knowledge net supports the notion of “emergent structures” since they are associative nets that change based on perception and experience. This represents a fundamental difference between this representation and more fixed representations such as scripts and frames (Schank, 1982) (loc: Page 44)

Note that Schank (1982) modified the script notation by introducing the concept of Memory Organization Packets (MOPs), which can be used to create scripts that are more appropriate to specific contexts. Kintsch and Mannes (1987) have shown how scripts can emerge from an associating knowledge net. Another approach to determining meaning is to examine the semantic relations between elements. Collins and Quillian (1969) used this approach. This approach, along with others, proposes a model of the organization of human memory (Minsky, 1975; Quillian, 1966; Rumelhart & Ortony, 1977; Schank & Abelson, 1977; Smith, Shoben, & Rips, 1974). (loc: Page 44)

Whereas the propositional representation of knowledge shows tremendous potential in educational research, its use is hampered by the lack of means to automatically code passages. It is impossible, for example, to propositionalize an entire textbook or a protracted (e.g. year–long) online discussion. Latent Semantic Analysis (LSA) may provide the requisite model for moving propositional analysis beyond the brief experiment. LSA provides a vector–based equivalent of the associative net discussed above. A proposition can be represented in high-dimensional space as a vector of numbers. This vector contains information about the proposition's relationship with other propositions in the knowledge net. The dimension reduction that is performed after the singular value decomposition serves to eliminate the random “noise” - the accidental misuses of words or details of the discourse space that are largely irrelevant to the knowledge it contains. This is not to say that LSA provides a complete model of human cognition. Indeed, LSA's “knowledge” has been likened to a “well–read nun's knowledge of sex” (Landauer et al., 1998, p. 5). LSA's only method of deriving meaning from the world is to “read” (or mechanically scan) text passages. Nevertheless, Kintsch (1998) closes his discussion of LSA by stating that “the preliminary results that are available at this point suggest a great potential for LSA for psychological knowledge representation” (p. 92) (loc: Page 44-45)

In a study that investigated matching students to text, Wolfe et al. (1998) assessed study participants by giving them tests that determined their knowledge of the human heart and the circulatory system. Participants were then assigned one of four readings that ranged in difficulty from elementary school level to medical school level. Their findings demonstrated that participants learned best when the material they read was neither too easy nor too difficult, given their current level of understanding. This they term the “Goldilocks” principle (similar to Vygotsky's (1978) Zone of Proximal Development), and link it with Kintsch's (1994) argument that learning from text is the ability to link it with previous knowledge. The authors cite a variety of works that demonstrate that learning from text requires appropriate levels of prior knowledge (McKeown, Beck, Sinatra, & Loxterman, 1992; Means & Voss, 1985; Moravcsik & Kintsch, 1993; Schneider, K”orkel, & Weinert, 1990; Spilich, Vesonder, Chiesi, & Voss, 1979) (loc: Page 46)

Knowledge Forum consists of a database that stores participant–generated ideas. The software helps users organize these artifacts into collections called “views”. Views consist of two–dimensional Cartesian planes onto which iconic representations of ideas can be placed. The view also provides a background onto which organizational information can be placed using graphics or text. (loc: Page 53)

The process of scaffolding is made explicit in Knowledge Forum through the use of “thinking types” or scaffold supports that are computer–mediated and customizable. They allow teachers and students to use scaffolds and rubrics flexibly and for students to tag their notes by thinking type (Andrade, 2000; Chuy, Scardamalia, & Bereiter, 2009; Lai & Law, 2006; Law & Wong, 2003) (loc: Page 53)

A limitation of the current system of scaffold supports that is available in Knowledge Forum is that the scaffold supports are only available for use at the note authoring level. That is, they are designed to be inserted directly into notes. There is no facility for scaffolding the creation of views. (loc: Page 54)

The Student ToolKit (STK) consists of two components: a Student ToolKit for queries (STKq) and a Student ToolKit for visualizations (STKv). The STKq facilitates the retrieval of notes that match criteria such as “show me my notes that overlap with curriculum guidelines”, “show me my growth in vocabulary”, and “show me key terms and phrases in the curriculum guidelines that are not represented in my notes to date”. The results of such queries can be fashioned to students in the form of “proto–views” that the students can then work up into views that they author. Alternatively, the temporary proto–views can be discarded, as may be appropriate for an interim assessment of, say, vocabulary growth (loc: Page 62)

The tools and techniques developed in the thesis have direct impact on future work that may employ methods from design–based research. Confrey (2006) traces the development of design–based research from early work by Piaget (1976), Vygotsky (1978), and Dewey (1981). Design researchers, she says “make, test, and refine conjectures about the learning trajectory based on evidence as they go, often collaborating with or acting as the teachers, and assembling extensive records on what students, teachers, and researchers learn from the process. They then conduct further analysis after the fact to produce research reports and/or iterations of the tasks, materials, and instrumentation” (p. 136). She highlights the early, seminal contributions of Collins (1992) and Brown (1992). A key aspect of design experiments or design studies is the provision of feedback that informs design iterations. One problem with the provision of such feedback is the time–sensitive nature of this information. Feedback is necessary but the analyses required to provide meaningful, intelligible feedback to the system are often tedious and time-consuming (Zhang, et al., 2007). Software that is designed to assist in the detection of patterns and trends in the developmental trajectories may serve to help shorten the delay between iterations and therefore may facilitate more powerful interventions. Perhaps more importantly, the software can serve to help design researchers detect trends and anomalies in ways that were previously difficult or impossible to detect. Thus the software can act as a lens onto the system being studied. By doing so, it has the potential to mitigate some of the problems associated with design experimentation. Specifically, as Collins, Joseph and Bielaczyc (2004) explain, design experimentation: involves putting a first version of a design into the world to see how it works. Then, the design is constantly revised based on experience…. Because design experiments are set in learning environments, there are many variables that cannot be controlled. Instead, design researchers try to optimize as much of the design as possible and to observe carefully how the different elements are working out. (p.18) (loc: Page 68-70)

The design of the Knowledge Space Visualizer (KSV) follows Shneiderman's (1996) threefold mantra for information visualization: (1) provide an overview, (2) provide the ability for the user to zoom and filter, and (3) provide details on demand (loc: Page 71)

One of the shortcomings of visual analyses is that they tend to be difficult to reproduce. Screen captures are an effective way to convey the resultant images, but the state of the system is not typically saved. The KSV works in concert with Knowledge Forum by allowing the state of a visualization to be saved to the database. This process allows researchers to share their visual analyses with each other, which in turn should promote collaborative analysis of data and more powerful interpretations of the data. (loc: Page 80)

Whereas terms are typically scaled according to their frequency of occurrence, there is no need to think that is the optimal representation for questions about semantic fields. Scaling by alternative criteria such as total number of people using the terms, may be advantageous. Hassan–Montero and Herrero–Solana (2006) have suggested a number of different ways to improve tag clouds including alternatives to the usual alphabetical ordering of the terms in the cloud. Term clouds are capable of representing higher dimensional data. In the example shown in Figure 42 the intensity of the font colour is scaled along with font size. It may be advantageous to use colour to indicate another dimension. (loc: Page 135)

tag cloud (loc: Page 135)

A problem with term clouds, in the traditional application of the technique, is that they do not do a particularly good job of showing semantic field growth. There are at least two reasons for this: no chronological aspect, and no sense of change vs. the previous term cloud (i.e. a sense of “delta”). The use of font sizing to scale the terms in the cloud is also questionable, as it merely highlights frequently used terms rather than highlighting changes over time. To address these shortcomings, a new graphical form of tag clouds was used. Small multiples (Tufte, 1990) were used to present a chronology of semantic field changes. In this technique, the term clouds reflect the changes between two other term clouds, and the term counts are indicative of the changes in term frequency rather than the absolute numbers. Only increases in term counts are represented (loc: Page 136)

Earl (2004) differentiates among “assessment of learning”, “assessment for learning” and “assessment as learning” primarily in terms of who the recipient of the assessment result is. Assessment of learning is used to communicate findings of student performance to other parties (e.g. parents) and to justify the grades that are assigned. Assessment for learning is characterized by having teachers as the primary recipients of the assessment results. In contrast to “assessment of learning” students are seldom compared to one another. Assessment for learning represents a shift from summative to formative assessment. Assessment as learning targets the students as recipients of the feedback. Students are regarded as “active, engaged and critical assessors [who] make sense of information, relate it to prior knowledge, and use it for new learning” (p. 97). (loc: Page 146-147)

The systems architecture model calls for the creation of user–generated views that can be based on content–based visualizations such as those demonstrating overlap with curriculum guidelines. Such demonstrations of coverage (or lack thereof) can be treated as artifacts within the discourse space and should have the effect of creating additional discourse around them, which in turn could be the source material for further visualizations. (loc: Page 160)

Other researchers have devised advanced, powerful, but time–consuming techniques for the analysis of threads in Knowledge Forum. For example, Zhang et al. (2007) propose a form of analysis called “inquiry threads” as a means to understand the socio–cognitive dynamics of knowledge building. (loc: Page 164)

It would be particularly helpful if a user could, while looking at a view, be able to ask for assistance in clustering notes and receive assistance in describing the resulting clusters. One could imagine a system in which a user could ask for a view to be organized into circular clusters, with a concise tag cloud rendered within each resultant cluster. (loc: Page 165-166)

The determination of coverage can be done either through explicit semantic links (i.e. referencing the curriculum documents), implicit semantic links (i.e. having a sufficient number of LSA–based note–to–note cosines that exceed some threshold), or possibly by extending the knowledge building environment to allow participants to mark a note as providing evidence of curricular coverage (loc: Page 169-170)

Some preliminary results from early pilot studies suggests that users experience a sense of wonder: they seem engaged by the fact that their ideas can be cast in a more global context. In both cases people appear to be driven by a sense of wonder. In the case of researchers it is typically curiosity about the system that they are studying. In the case of participants it is often a sense of wonder about oneself. It may be particularly powerful to make it easy for students and teachers to display the semantic space around a particular note. Wonder, then, contributes to transformative assessment. It is not externally imposed against some foreign criterion, but rather internally generated. Well-designed and well–implemented assessment tools should be capable of generating positive feedback, in the cybernetic sense, for wonder and in turn both depends on and encourages engagement on the part of students (loc: Page 172)

A significant improvement to Knowledge Forum would be the inclusion of a facility to allow flexible post hoc assignment of relationships between notes. For example, one could indicate that one note “explains” the ideas in another note. Scaffolding the creation of links that indicate the relationships between notes could become as important as scaffolding the content of notes. Once the relationships between notes are thus identified, visualization techniques can be applied to make sense of the resulting patterns. (loc: Page 179)

Collins and Halverson (2009) envision a future in which technology enables people of all ages to pursue learning on their own terms. In his review of their book, Grover (2009) notes the authors' self–admitted ambiguity around next steps and implementation (loc: Page 181)