The term knowledge gardening, a contraction of the longer dynamic knowledge gardening (DKG), is a direct descendant of Douglas Engelbart's Dynamic Knowledge Repository (DKR). A DKR exists as a combination of humans and tools, epistemic communities and the tools they use to aggregate information resources and work products, and to collaborate. We describe TopicSpaces, an open source topic-map-based framework with which the collective hypermedia discourse of epistemic communities is federated. We define hypermedia discourse as the totality of social gestures made by such communities. That is, recorded dialogs, linked, annotated and tagged Web resources, recorded stories, virtually all addressable information resources created anywhere on the Web constitute the range of resources federated. We define federation of resources as the specific merging processes native to topic mapping. We contrast federation with traditional semantic integration processes where artifacts of knowledge are aggregated through processes of selection. Where selection processes involve “weeding” (a gardening process), federation does not perform weeding during the merge process; rather, federation involves including all resources during the merge process; social processes including reputation, trust, and dialog will help determine which resources users find most valuable in their work. TopicSpaces provides a map of the federated territory, user interface tools to facilitate some hypermedia discourse practices, and Web services to interface with other hypermedia discourse tools.
We offer a position paper that describes one approach among many to the federation
of heterogeneous information resources and world views. Our thesis is that a
subjectcentric federation is appropriate to the problem of supporting knowledge gardening
(also known as collective sensemaking—see Section 7) to find solutions to complex
and urgent problems. Our work with SRI’s Cognitive Assistant that Learns and
Organizes (CALO) project1 has taught us numerous lessons that support the need to
federate heterogeneous world views and information resources. For instance, Park and
Cheyer [
In our work, we have adopted the term knowledge gardening as a name for the federation processes. This follows Douglas Engelbart’s term Dynamic Knowledge Repository, which is his name for the combination of people, software tools, and processes as improvement communities. In some illustrations of our work, we use the term knowledge garden to name our topic map-based Web portal. Our story explores the role of topic maps in the federation of heterogeneous information resources through processes of subject identification and merging different representations of the same subject in the same map. We believe that the maintenance of well-organized information resources can contribute to improvements in knowledge gardening processes, toward improved human dialogue.
This paper is organized as follows. We first review the two elements of hypermedia discourse that we federate through topic maps. They are semantic linking (connecting), and dialogue (sometimes also known as issue) mapping. We include social bookmarking as an additional element; while hypermedia discourse2 centers on contested assertions and ideas, knowledge gardening entails the wider range of social activities on the Web. We then review our subject-centric federation, or knowledge gardening approach. We then sketch TopicSpaces, our prototype federation platform. We then introduce the knowledge gardening process and close with illustrative examples. Brief references to related work are given where appropriate.
Tags are associative reminders. In the CALO project, tags are the names of projects in
which CALO users are engaged. For instance, one typical CALO project is the CALO
“platform” itself, a project where CALO developers keep track of the design and
development progress on the product. The tag “Platform” would be used by CALO
developers as they surf the Web looking for information resources of value to the
team. They use that tag with Tagomizer, CALO’s social bookmarking application
written on top of the topic map engine TopicSpaces [
Tagging is part of the larger social gardening repertoire; tags leave trails or form
scents [
While tagging is generally thought to enable the formation of clusters of topics,
Brooks and Montanez report some interesting results [
Grouping and clustering topics with tags is not the only application for tagging.
We continue to discover new applications. For instance, Razavi and Iverson [
In some sense the entire Semantic Web enterprise is about semantic linking. In the
sense discussed here, a narrow definition is taken: semantic linking here refers to the
creation of typed connections between ideas found in documents on the Web. In that
sense, semantic linking is subject-centric by its very nature. In 2001, the Scholarly
Ontologies Project at the Knowledge Media Institute began to envision a
“complementary infrastructure that is ‘native’ to the internet, enabling more effective
dissemination, debate, and analysis of ideas”4. In 1999, three authors [
Dialogue mapping provides a common view of a growing structured representation of
streams of thoughts [
The map reads left to right, starting, essentially, with an opening question. The node “Food riots” leads to the columnist’s opening question: “How did this happen?” The columnist provided his own three answers: “Long term trends”, “Bad luck”, and “Bad policy”. From there, it is a matter of picking out questions being asked, finding answers, and identifying any arguments made in the prose. A similar dialogue map would occur if a discussion group was facilitated by a skilled dialogue mapper and similar questions and responses were recorded.
We live in a vast collection of universes of discourse, each centered on different topic domains, many of which overlap and share subjects and concerns. The issue map of the OpEd illustrated in Figure 3 could just as easily have been generated in slightly different forms, each representing a different interpretation by a different analyst. That each is somehow different contributes to heterogeneity in information resources with which we must all cope in our day-to-day and decision-making lives. A goal of our work is to federate these heterogeneous resources into a coherent representation with which we believe improved knowledge gardening is afforded.
Consider just one node in our OpEd issue map, the one shown in Figure 3, for which the label reads “700 calories of animal feed to produce 100 calories of beef”. That is a specific quote from the OpEd text; it is reasonable to expect that other analysts might pick up the same claim, even if placed in a different part of the map’s graph structure.
Claims such as that are, at once, subject to fact checking, and to entailed subjects. Fact checking can be the work of background agents, or the work of the crowd engaged in knowledge gardening. Subject entailment goes with the nature of the claim. That is, there is a relationship between animal feed and animals, and both of those two subjects exist in a web of related (entailed) subjects. Consider the simple concept map (Figure 5) of some (but not all) subjects entailed by the node illustrated in Figure 4.
By creating a topic map of dialogues, and by including all entailed subjects, we gain a broader means by which the work products of knowledge gardening can be evaluated. By linking into that map each node created by each individual, no matter how that node falls in its native dialogue map structure, we are performing subjectcentric federation: we are bringing together information resources that are about the same subject, and we are connecting those resources to all known to the map resources of the same or related subjects. We do so without editorial bias; we federate regardless of whether or not we agree with claims represented. We leave disagreements to the gardening processes in which the map’s users are engaged.
Tools that support dialogue mapping include Compendium8, bCisive9, TruthMapping10, and DebateGraph11. Compendium and bCisive are desktop tools, while TruthMapping and DebateGraph are online portals.
Mark Klein [
TopicSpaces is a servlet-based Web portal provider that includes a subject map,
which is a topic map created according to the Topic Maps Reference Model [
The platform illustrated in Figure 6 anticipates the ability to run seti@home-like
agent-based harvesting of resources found on the Web. A tuplespace platform [
TopicSpaces is a research platform, one that can support two classes of topic maps portals as illustrated in Figure 7. One class is the all-in-one portal where all the context view portals, collaboration portals, and personal workspaces are part of the same software package. TopicSpaces is built like that as a means to explore all issues related to knowledge gardening.
A second class of portal separates all the context portals, collaboration portals, and so forth from the subject map itself. Different portals can then be crafted using standard CMS platforms such as Drupal, WordPress, and other popular software products. TopicSpaces can provide Web services to those portals as needed.
What is a REST Web Service? It is simply a means to use URLs as query vehicles by way of a servlet. Web browsers make such requests routinely; type a particular URL into a browser and the server returns the entire Web page in a single HTML string. A Web service would, instead, return a small fragment of HTML, of XML, or Javascript Object Notation (JSON)13 as requested. Bookmarklets, as used by Tagomizer, del.icio.us, and other social websites, represent a kind of Web service where a short Javascript string embedded in a browser’s bookmarks is able to transport information from a Web page to the portal that accepts the Bookmarklet’s query. When we say “API”, we are specifying that there is a particular query string that goes in the URL, and that query string is interpreted by the portal to perform the requested task. Some tasks are to return a requested bit of information, the bookmarks 13 JSON: http://www.json.org/ associated with a particular tag, say. Other tasks are to update information in the topic map, to add a new bookmark, say.
The TopicSpaces REST API takes the form:
<server>/ws/<appname>/<object>/<return>/<data> For instance, asking for a Tagomizer tag in HTML where the tag is “SomeTag” is this query fragment:
/ws/tago/tag/html/SomeTag
/ws/tago/tag/json/SomeTag
We open our discussion on knowledge gardening by reviewing related work. That work is embodied in a literature under the subject of sensemaking, making sense of complex situations. Thus, knowledge gardening is our name for sensemaking.
Brenda Dervin's sensemaking methodology [
Sensemaking has been approached from the perspective of surprise, of expectation
failures [
Perform cycles of elaboration on that frame, adding and filling slots, seeking, inferring and discovering data Ask questions of the frame, detecting inconsistencies, judging plausibility, analyzing data quality Perform cycles of refactoring, where the process is to seek a new frame that better describes the situation
In the line of inquiry framework [
Jean-Claude Bradley [
Standing by itself as a new class of sensemaking portal is Science X214. The portal provides users with dashboards that consist of unread posts to groups to which the user is subscribed, lists of “signals”, “hypotheses”, and “forecasts” generated by the user. While we have only a “beginner’s” experience with the website, it appears that users post signals, an instance of which might be “Topic maps improve sensemaking”, and other users form hypotheses around such signals and later offer forecasts. We view this portal as federation of goal-oriented blogs, tightly coupled through the three classes of artifacts. In some sense, the portal, by virtue of its three specified artifacts, is naturally self-organizing in a subject-centric fashion.
As we continue to evolve our tools, and as we use them in our own research, we are beginning to understand, if even to a somewhat naïve level, what so-called best practices might look like. We now understand some best practices for tagging, and are just now beginning to practice semantic linking and dialogue mapping. Those best practices exist in the context of the larger gardening process. Gardening processes occur within some context, some goal, some working hypothesis or research question. 14 Science X2: http://sciencex2.org/ We see the process as iteration around and within this sequence: 1. Forage 2. Filter 3. Analyze 4. Synthesize Foraging and filtering are the information-seeking stages in which combinations of goal-directed search and thematic vagabonding result in discovered information resources. In this stage, one tags the resources for later harvesting. This is the stage where benefits accrue from tagging best practices. In our CALO scenario above, we described the application of a project-centric tag ontology, the use of predefined tags for specific purposes. We are learning that it is appropriate to use more than one tag for each resource discovered. While CALO prescribed project-centric tags, we further prescribe subject-centric tags. While reading a particular resource on discovery, take the time to tag the particular actors, relationships, states and other important subjects bound by the resource. This extra work pays large dividends later.
Concurrent with tagging, semantic linking serves as a transition to analysis
through partial harvesting and forging semantic connections between ideas harvested
from the pages visited. We are able to use the full suite of hypermedia discourse tools
in the foraging-filtering stages and in transitioning to analysis [
Reading this issue map from left to right, it organizes the concepts about which our literature review must speak. Toward the right, we begin to tease out of the literature each argument made, and we tie each argument to the specific citation from which it is drawn. The two key concepts were URIs from the Web community and PSIs from the topic maps community. Our analysis suggests that they behave as the same concept, and we note that through a Cohere-like coherence relation.
The analysis stage includes finding answers to research questions posed at the beginning of the process, and derives new questions to ask and finds their answers— or reports them as targets for future work. In the analysis stage, some assertions made during foraging and filtering – our Same As assertion, for instance– may come under close scrutiny by those who do not share the same world views. It is at this point where dialogue mapping services enter the arena and various actors take positions and offer arguments. That’s knowledge gardening at work.
A general outcome of this work for CALO is to provide a Web-based presence that supports knowledge gardening among communities of CALO users, as shown in Figure 10.
Fig. 10. Topic map-based gardening portal for communities of CALO users
In the work reported here, we have installed an instance of TopicSpaces and we have begun to use it in two different contexts: developing a dashboard platform for CALO, and using it to organize our thesis research, snippets of which have been illustrated here. An allied goal has been to demonstrate the ability to federate communities of CALO users where subjects important to all members of the community are shared and maintained at the Web portal.
Fig. 11. A sample portal with a bookmarks dashboard view
Figure 11 illustrates an early instance of a dashboard. This dashboard uses a REST query as follows:
/ws/tago/tag/html/GENIS:Source where the tag GENIS:Source is drawn from a tag ontology that allows us to bookmark using tags related to energy sources, uses, and issues.
Figure 11 also illustrates a context view portal as included in Figure 7, where we have created a view that facilitates navigation into the world of energy sources. Users are able to create new source links, and are also given ready access to websites tagged for the general class. A Web page for the subject Wind Energy (source) might include a bookmark dashboard that is a composite query on GENIS:Source + WindPower, which narrows the source bookmarks to those also tagged with the particular source type.
Through such tagging and annotating processes, we believe that it is possible for communities of practice to create and maintain a knowledge base that fully supports the community’s gardening activities through maintenance of dashboards of various kinds. Our project remains work in progress; we continue to explore the boundaries of dashboard construction; we have only now begun to scratch the surface of that inquiry.
Let us consider an instance of connecting dots. The semantic linking capabilities of Cohere allow us to read two different stories and lift out of them the following two ideas: • •
Antioxidant supplements reduce Free Radicals from the body
Suppose now that we happened to tag each page with the tag “freeradical molecules”. Allow that different individuals used the same tag and lifted those ideas independently.
Later, someone performs a tag-based search or discovers the tag “freeradical molecules” and notices those two ideas together. Those two ideas, to astute viewers, pose a problem: if you need freeradical molecules to fight bacterial infections, you probably don’t want to be taking high-dose antioxidants. That inference might be based on a simple, common-sense heuristic as follows:
If AndIf Then
Process X REQUIRES Substance A Process Y REDUCES Substance A
Reduce or Eliminate Process Y
With that heuristic, we now are able to suggest a claim to the knowledge garden that the exposure to AntiOxidant is contraindicated when Immune Response to a bacterial infection is active, offering the discovered ideas as evidence, as illustrated in Figure 12.
How does our federation platform facilitate this discovery? TopicSpaces provides a social bookmarking tool, Tagomizer, along with the subject map that maintains the federated knowledge artifacts. Cohere can be accessed through Web services, with which a gardener lifts ideas out of Web pages visited perhaps at the same time the tagging process is engaged. The subject map keeps track of these actions, making them immediately available to those who follow on other journeys through the garden. We note that a similar federation platform can exist with other tools such as http://del.icio.us/ providing the bookmarking capabilities. Our approach, by contrast, takes the federation itself as the starting point, building in the necessary tools and Web services.
In the broader context of knowledge gardening, one might ask “how were those
dots found in the first place?” How does our knowledge garden platform support
query? Support is found along two dimensions: full text search and navigation
through social bookmarks (tags). Consider the tagging scenario where a gardener
discovers the Web page from which the idea of Figure 12 is lifted. At the same time,
that gardener tags the site with ImmuneResponse and FreeRadical. Another
gardener discovers the Web page from which the idea of Figure 13 is lifted. That site
is tagged with AntiOxidant and FreeRadical. Tagging behaviors such as just
illustrated are suggestive of a best practices approach to tagging: freely tag with the
major terms found in an information resource. Another gardener, for reasons perhaps
related to a disease being researched, lands on the FreeRadical tag, observes two
rather surprising ideas both related to the same subject, and reasons, as suggested by
the heuristic, that, “if I am sick, perhaps I shouldn’t be taking antioxidant
supplements”. For people fighting a disease that provokes an immune response that
requires free radical molecules, the new discovery turns out to be an important one.
Such is the nature of Black Swan Events [
Acknowledgments. This material is based in part upon work supported by the Defense Advanced Research Projects Agency (DARPA) under Contract No. FA875007-D-0185/0004. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of DARPA or the Air Force Research Laboratory (AFRL). This paper has been approved by DARPA: “Approved for Public Release, Distribution Unlimited”.