Using Network Visualizations as Knowledge Communicators

I thought I might share my final paper for Prof. Tversky’s class, Using Network Visualizations as Knowledge Communicators. The paper asks the question, in what circumstances are network visualizations an effective tool for transferring knowledge?

Using Network Visualizations as Knowledge Communicators

Overview

The rise of the Internet as a means of communicating information in the late twentieth century and early twenty-first century has brought with it a series of challenges related to the organization and transferability of knowledge. One of the greatest challenges is the sheer abundance of information, leading researchers and technology developers to implement a variety of solutions to improve the manageability of knowledge. A method often employed is providing meta-level analysis capabilities, such as summarization and visualizing of a field at a high-level, and then augmenting with tools for drilling-down into specifics, such as the Grokker visual search engine. Others have responded by creating large maps and visualizations which illustrate how knowledge is related (e.g., Börner, Chen, & Boyack, 2003). And others have posited the need for better solutions. For example, in the field of education, researchers have proposed SPINE (Scholar Practitioner Information Networks for Education):

It has come to a point where the education community needs to have at its disposal better tools for visualizing and measuring the connections between artifacts, their contents, and the constituencies that generate, evaluate, and consume them. Such a tool could potentially reorient members’ explorations and conceptualizations of ideas, and provide a deeper understanding of the field in which such ideas are embedded. A new researcher to the field of education should be able to understand central pieces of research and practice, the communities based around them, and their current level of activity. Professionals should be able to both contribute to the research community through their experiences on the ground and understand enough of the current research to affect their own practices. Sub-disciplines within the field of education should be able to see how they interrelate with each other, as well as how their work informs educational practice. (McFarland, J. & Klopfer, E., 2006)

Each of these examples illustrate a compulsion to represent information in a way that can be easily understood by newcomers and provide a resource for experienced information seekers to augment their own knowledge. Among the examples of knowledge organization and management solutions, a common attribute is the use of the node-link map, which we will also refer to as network visualizations. Network visualizations can be useful for illustrating how some entities are related to other entities, and can vary depending on visualization needs, including differences in size, complexity, and representation details (e.g., color, link labeling, and spatial layout). Although not all knowledge management solutions employ the network visualization (e.g., Grokker uses a hierarchical combination of packed circles), it is certainly one of the most popular diagrams for illustrating knowledge relationships (Wang, 2006). For example, Google Inc. held a student competition in 2006 for the best gadget. The winning project was called MapMyWord, which is an interactive “graphic dictionary and thesaurus… [that] displays meanings, synonyms, and their relationships (16 types) in color-coded radial clickable graphs? (interface shown in figure 1). The centerpiece of the project is a node-link map which shows the relationships between words and synonyms.


Figure 1. 2006 Google Gadget Awards winner, MapMyWord

MapMyWord illustrates the centrality and perceived utility of the node-link map as a way of communicating information. Considering the wide-use of this representational tool, especially in web-based environments, this paper will concern itself with the following question: in what circumstances are network visualizations an effective tool for transferring knowledge? To address this question, this paper will be divided into four major sections: a) contemporary uses of network visualizations, b) how humans process these visualizations, c) learning outcomes related to the use of network visualizations, and d) problems, solutions, and conclusions.

Contemporary Uses of Network Visualizations

Network visualizations are used in a variety of applications to accomplish a wide-range of goals. One popular use is visualizing communication across entities or individuals (e.g., social network analysis, analysis of computer networks). Other popular uses are illustrating human activity on the World Wide Web (e.g., via page linking). Still other uses are for personal productivity or educational reasons. An interesting example of this type of use is called Pathway, which is a tool designed for the Macintosh which displays a graphical network representation of one’s traversals in Wikipedia, illustrated in figure 2 (Lorson, 2007).


Figure 2. Pathway, a tool for visualizing interaction in Wikipedia (Lorson, 2007)

The main activity of the system is to track page visits in Wikipedia, then plotting the activity on a network visualization. According to the authors, the visualizations allows you to “keep track of everything: what you’ve looked at, how you got there and just how it all fits together.? In sum, the author implies that the visualization acts as some sort of memory aid, as well as a culminating utility by illustrating how it “all fits together?. A further interesting aspect of Pathway is the product tag-line, which is “Making Wikipedia a no-brainer.? This tagline implies that employing a network visualization will somehow reduce the cognitive resources needed to use Wikipedia. However, how can the developers be certain that this is the case? Isn’t it possible that a network visualization adds another level of complexity onto an already complicated arrangement? How can we be sure that a network visualization is an effective means of illustrating how things “fit together??

An example of using network visualizations for educational purposes is the concept mapping technique, pioneered by Novak (1998). Concept mapping is quite different from other forms of mapping in that it takes subjective knowledge into consideration, unlike most maps which are often designed as universal communication devices (e.g., street map or subway map). The reason for this dependency is concept mapping’s foundations in constructivist learning and Ausubel’s assimilation learning theory, which stipulates that meaningful learning is a process in which new information is related to an existing and relevant aspect of an individual’s knowledge structure (Novak & Gowin, 1984; Novak, 1998; Ausubel, 1968). Paper-based concept mapping techniques have been transferred to computer-based environments with CMapTools, which allows users to create concept maps and share them with others on the Internet (see figure 3).


Figure 3. CMapTools (Novak & Cañas, 2006).

Other contemporary uses of network visualizations go beyond static map displays and use node-link maps in conjunction with animation. For example, Digg Labs uses a network visualization to illustrate which stories on the Digg website users are “digging? in real-time. Links are drawn between nodes to show which stories are related to which users, and then disappear to make room for more stories (see figure 4).


Figure 4. Digg Labs visualization using animation

This example, among the others presented, illustrates the network visualizations are currently being used for a wide-variety of purposes to achieve diverse goals. However, are these appropriate uses of node-link maps and how can we be certain they provide benefit over other representational mechanisms, such as texts, outlines, lists, or other diagrams such as hierarchies, charts or graphs? In the next section, we will discuss the research on how network visualizations are processed in order to shed light on this question.

Processing Network Visualizations

One way of explaining how humans process network visualizations is by equating this type of diagram with other types of diagrams, most notably charts and graphs. Kosslyn (1989) provides some explanation how graphs are processed and tips on designing better charts and graphs. One major factor to consider is the limits to short term memory: “The capacity of short-term memory is notoriously limited: information can be held in short-term memory for only a few seconds once one has shifted one’s gaze from the stimulus, and only a small amount of information (about four groups of items) can be held in this store at the same time? (p. 191, illustrated in figure 5). Hence, graphs that are too complex will reduce overall comprehensibility.


Figure 5. Visual Information processing

He also advises that networks should be designed with gestalt principles in mind, such as the following:

  • Good continuity- marks that suggest a continuous line will tend to be grouped together.
  • Proximity- marks near each other will tend to be grouped together.
  • Similarity- similar marks will tend to be grouped together.
  • Good form- Regular enclosed shapes will be seen as single units.

Interestingly, many of the contemporary uses of network visualizations violate many of Kosslyn’s recommendations (e.g., Digg labs visualization), especially with respect to attending to overall diagram complexity. In sum, unlike graphs that aim to be easily readable, it is not uncommon to be presented with a network visualization which requires a good deal of cognitive resources to comprehend.

Other researches shed light on how network visualizations are processed. Novick and Hurley (2001) posit that some types of information are fundamentally better suited being represented on one form of diagram versus another. Hence, some things are more psychologically predisposed to be represented on a network (as opposed to matrix or hierarchy):

We proposed that a network is most appropriate when the represented world contains only a single set of objects. A number of subjects (mostly from the computer science population) indicated that this condition holds when the number of objects to be represented is relatively small. When the number of objects is large, however, the network is likely to be unwieldy and therefore a matrix is preferred. […] Bertin (1981, p. 129) is explicit on this issue, stating that the network ‘ceases to be a means of discovery when the elements are numerous. The figure rapidly becomes complex, illegible and untransformable.’ With 20/20 hindsight, we concur. (p. 199-200)

In addition to Novick and Hurley’s conclusions, other conditions affect human ability to process network visualizations. Wallace, West, Ware and Dansereau (1998) conducted a study which looked at various visual features of network visualizations and how these features affect student ability to recall information on the map. They concluded that use of color, shape and proximity in network visualizations facilitated learning by improving the organization of information compared to the control condition where such attributes were excluded.

In addition to color, space, and proximity, Chmielewski, Danereau and Moreland (1998) found that common region is also important for learning from node-link maps. Common region is:
…based on the phenomenon that objects that share a common region of space (i.e., contained within a perceptually defined area) are seen as being grouped together and separate from other groupings. This principle differs from closure and proximity in that it does not rely explicitly on figure completion or distance. As long as objects occupy a location in space that is delineated by an explicit border (e.g., bounded, homogeneously colored, or textured), they will tend to be grouped together (e.g., cities in maps showing state boundaries)

In this study, the researchers presented students with node-link maps where in the experimental condition nodes related were grouped together with a rectangle, thus manipulating common region, and in the control condition no such grouping was provided. The study invokes the notion of field dependence and independence, which has to do with one’s ability to disambiguate a figure from its surroundings (Coren, Ward, & Enns, 1992). Those who are field-dependent have trouble disambiguating a figure from its background; however, field-independent persons can easily perform this task (Witkin, Moore, Goodenough, & Cox, 1977). The results of the study show that people who are field-dependent perform better on a recall test where common-region is used (e.g., nodes are enclosed in a rectangle), and field-independent persons perform better on a recall test when common-region is not used (e.g., no rectangle is employed). The researchers reach the conclusion that field-dependent “individuals require more structure, whereas field-independent individuals tend to thrive in tasks where information is relatively unstructured.? The study is significant in that it highlights that individual differences are key when it comes to comprehending a network visualization.

Learning from Network Visualizations

In addition to studies that examine the intricacies of how people process network visualizations (e.g., effects of gestalt principles, psychological predispositions), other studies have analyzed learning outcomes from students creating or comprehending network visualizations. This section will discuss some of those studies as a way of shedding light on in what contexts network visualizations are best used.

Nesbit and Adesope (2006) conducted a meta-analysis of 55 studies using knowledge maps and concept maps for learning, totaling 5,818 participants from grade 4 to postsecondary in such domains as science, psychology, statistics and nursing. Although there was significant heterogeneity in the studies, all employed posttests to measure recall. The meta-analysis found that “in comparison with activities such as reading text passages, attending lectures, and participating in class discussions, concept mapping activities are more effective for attaining knowledge retention and transfer? (p. 434). The study shows that concept maps may be slightly more effective than other creation activities such as writing summaries and outlines, however, the effect is quite small. It also finds that concept maps are somewhat more effective than studying other summary formats, such as lists and outlines. Further, studying concepts maps for central ideas and detail ideas were more effective than reading text passages; however, the effect may be stronger for central ideas. However, there is “insufficient evidence to determine whether studying concept maps is particularly efficacious for knowledge transfer and development of learning skills?. The researchers note that the positive effects in favor of concept maps may be due to learner engagement rather than any specific property of concept maps. In sum, this study highlights that there are no obvious reasons not to use network visualizations (either for study or communication of knowledge), especially as replacements for text passages, outlines, or lists.

In addition to the learning outcomes research related to node-link maps, other research has investigated the motivational and affective outcomes involved in studying and creating concept maps. Hall and O’Donnell (1996) conducted a study to understand the emotional differences a student experiences when studying a knowledge map versus a traditional text. The 43 students who participated would read the text (both map and traditional text) and 2 days later perform a free recall. The study reveals that the map group exhibited significantly higher levels of motivation and concentration during studying and testing compared to the traditional text group (p. 99). Additionally, the map group significantly outperformed the text group in the recall task. However, the subjective and objective outcomes were independent (e.g., those more motivated did not necessarily recall more). In sum, knowledge maps may be more engaging for people to interact with compared to traditional texts.

Other studies have found concept maps are more motivational as ways of communicating one’s knowledge as compared to text writing. Czuchry and Dansereau (1996) conducted a study where they had a concept mapping activity replace paper writing, and asked students to evaluate their impressions of concept mapping compared to traditional paper writing. The results indicated that students, “regardless of course level, rated the mapping assignment as more interesting and noted that they learned more from it than a traditional paper assignment? (p. 94). The study also found that women thought the mapping assignment were easier than traditional writing assignments, where men found the difficulty the same. This study illustrates that node-link maps can have motivational consequences; however, individual differences should be taken into account.

Problems, Solutions and Conclusions

Network visualizations, although useful for learning contexts, present a number of problems. The first of which is the structure of the visualization makes nuance difficult to achieve because all nodes and links are of equal weight. This is especially problematic where the relationships are variable. For example, many network representations of related knowledge or concepts must be statistically generated using semantic analysis, hence leading to relationship values with varying weight. For example, figure 6 shows related concepts on a network visualization from a library catalog:


Figure 6. Library catalog showing relationship between concepts (retrieved from http://aqua.queenslibrary.org/?q=spatial%20cognition)

The problem is readily visible in the diagram, where topics such as “psycholinguistics? are on the same level as “toddler? in their relationship to “spatial cognition?. Although there is clearly a relationship between “toddler? and “spatial cognition?, it is certainly a different kind of linkage than “psycholinguistics?. Hence, a problem with plotting concepts on a network visualization is that it is difficult to illustrate the variances in relationships, since all nodes and links are equal in weight (although differences can be made more apparent using gestalt principles, as the research earlier presented suggests).

A second issue with current uses of network visualizations relates to a discrepancy between complexity and tools for navigating such complexity. For example, many displays of networks include thousands, if not million of nodes and links, presented on a static display that will fit onto a normal computer screen. These types of network visualizations, although visually stunning, may not be useful to future research studies unless the network can be manipulated and zoomed-in on. For example, figure 7 shows a network visualization of aging research (Boyack & Börner, 2003).


Figure 7. Visualization of aging research (Boyak & Börner, 2003)

Although the diagram is quite artistically inspired, its static presentation prevents it from being further studied by others (e.g., there is no ability to manipulate, highlight elements, remove elements, zoom in, etc.). In observing such complex networks, Kosslyn’s (1989) research on how individuals process graphs or charts should be recalled. Kosslyn found that increased complexity will coincide with decreased comprehension, and that the number of perceptual groups should be limited to between four and seven (p. 196). Although figure 7 would certainly violate some of Kosslyn’s notion of what makes an understandable diagram, it certainly could be alleviated with tools for cutting away at complexity and zooming-in on relevant elements.

Other problems with network visualizations include the questionable nature of what gets represented on a node and what gets represented as a link (B. Tversky, personal communication, November 1, 2006). In some instances, such as plotting network data, the nodes and links seem obvious (e.g., the physical objects are represented as nodes, where the communications are represented as links). However, it is not always evident why some map design choices are made, and if the choices are made simply because there are too few options (e.g., one must choose between a link or a node, or choose a different type of diagramming tool).

In conclusion, there are several “take-aways? with regard to the use of network visualizations. These “take-aways? will come in the form of questions which can be used when considering using a network visualization to communicate knowledge. 1) Is the data you are trying to represent psychologically predisposed to being represented on a network? Novick and Hurley (2001) research suggest that people may be more psychologically predisposed to think of networks as good at representing single set of objects, and those more technically inclined to think that set should be relatively small in size. 2) Does your user population thrive in structured or ill-structured environments? If they prefer structure, the networks should invoke common region, or nodes should be clearly boxed together. If they prefer ill-structured environments, explicit grouping of nodes should not be used (Chmielewski, Dansereau, & Moreland, 1998). 3) Are there concerns about motivation? Node-link maps can be more engaging to create and to study than traditional texts and should be considered if motivation is an issue (Hall & O’Donnell, 1989; Czuchry & Dansereau, 1996). 4) Do the relationships between nodes and links need to be nuanced? If so, the node-link map may not be an appropriate tool and text may be a better alternative. 5) How complex is the data you are trying to represent? If the data is complex, one should be aware that comprehension may decline if too much information is placed on the visualization (Kosslyn, 1989). This might be counteracted by providing tools for navigating the map, such that information can be removed, highlighted and zoomed-in on. 6) Do you need your user population to remember something? Network visualizations lead to better free recall than texts, lists and outlines (Nesbit & Adescope, 2006). However, the research which correlates node-link maps to recall ability is more substantial where people create their own node-link maps versus simply studying a pre-completed node-link map.

References

Ausubel, D. P. (1968). Educational Psychology: A Cognitive View. New York: Holt.

Bertin, J. (1980). The basic test of the graph: A matrix theory of graph construction and cartography. In P. A. Kolers, M. E. Wrolstad, & H. Bouma (Eds.), Processing of visible language, 2, (585–604). New York: Plenum.

Börner, K., Chen, C., & Boyack, K. W. (2003) Visualizing Knowledge Domains. Annual Review of Information Science & Technology, 37, 179-255.

Boyack, K. W. & Börner, K. (2003). Indicator-Assisted Evaluation and Funding of Research: Visualizing the Influence of Grants on the Number and Citation Counts of Research Papers. Journal of the American Society of Information Science and Technology, Special Topic Issue on Visualizing Scientific Paradigms, 54(5), 447-461.

Chmielewski, T. L., Dansereau, D. F., Moreland, J. L., (1998). Using Common Region in Node-Link Displays: The Role of Field Dependence/Independence. Journal of Experimental Education, 66(3).

Coren, S, Ward, L. M., & Enns, J. (1992). Perception. Fort Worth, TX: Harcourt Brace.

Czuchry, M., & Dansereau, D.F. (1996). Node-Link Mapping as an Alternative to Tradtional Writing Assignment in Undergraduate Psychology Courses. Teaching of Psychology, 23(2), 91-96.

Digg Swarm (2007). Retrieved January 7, 2007 from http://labs.digg.com/swarm/

Grokker. (2007). Retrieved January 7, 2007 from http://www.grokker.com

Hall, R. H. & O’Donnell, A. O. (1996). Cognitive and Affective Outcomes of Learning from Knowledge Maps. Contemporary Educational Psychology, 21, 94-101.

Kosslyn, S. M. (1989). Understanding Charts and Graphs. Applied Cognitive Psychology, 3, 185-226.

Lorson, D. (2007). Pathway: Making Wikipedia a no-brainer. Retrieved January 7, 2007 from http://pathway.screenager.be/about-pathway/

McFarland, J. & Klopfer, E. (2006). Network Searching: A New Way of Seeing the Education Knowledge Domain. Teachers College Record, Retrieved Janauary 7, 2007 from http://www.tcrecord.org/content.asp?contentid=12844

Nesbit, J. C., & Adescope, O. O. (2006). Learning With Concept and Knowledge Maps: A Meta-Analysis. Review of Educational Research, 76(3), 413-448.

Novak, J. D. & Gowin, D. B. (1984). Learning How to Learn. Cambridge: Cambridge UP.

Novak, J. D. (1998) Learning, Creating, and Using Knowledge: Concept Maps as Facilitative Tools in Schools and Corporations. Mahway, NJ: Lawrence Erlbaum.

Novak, J. D. & Cañas, A. J. (2006). The Theory Underlying Concept Maps and How to Construct Them, Technical Report IHM, Florida Institute for Human and Machine Cognition, Retrieved January 7, 2007 from http://cmap.ihmc.us/Publications/ResearchPapers/TheoryUnderlyingConceptMaps.pdf

Novick, L & Hurley, S. M. (2001). To Matrix, Network, or Hierarchy: That Is the Question. Cognitive Psychology, 42, 158-216.

O’Donnell, A.M., Dansereau, D.F., & Hall, R.H. (2002). Knowledge Maps as Scaffolds for Cognitive Processing. Educational Psychology Review, 14, 71-86.

Tversky, B. Visuospatial Reasoning. In Holyoak, K. J. & Morrison, R. G. (Eds.), The Cambridge Handbook of Thinking and Reasoning (209-240). Cambridge: Cambridge UP.

Wallace, D. S., West, S. W. C., Ware, A., & Dansereau, D. F. The Effect of Knowledge Maps That Incorporate Gestalt Principles on Learning. Journal of Experimental Education, 67(1).

Wang, W. Wang, H., Dai, G., & Wang, H. (2006). Visualization of large hierarchical data by circle packing. In Grinter, R., Rodden, T., Aoki, P., Cutrell, E., Jeffries, R., & Olson, G. (Eds.), Proceedings of SIGCHI Conference on Human Factors in Computing Systems, (517-520), New York: Association for Computing Machinery.

Witkin, H. A., Moore, C. A., Goodenough, D. R., & Cox, P. W. (1977). Field-dependent and field-independent cognitive styles and their educational implications. Review of Educational Research, 47, 1-64.

Leave a Reply

Your email address will not be published. Required fields are marked *