- The use of knowledge management systems is often hampered by the heavy overload for publishing information. In particular, uploading a document and then profiling it with a set of meta-data and keywords is a tedious and time-consuming activity. Therefore, one of the main goals for such systems should be to make publishing of explicit knowledge as natural as possible. In the project described in this paper, we exploit a semantic wiki editor to support document publishing by means of textual descriptions augmented by ontology-defined annotations. Such annotations are then managed as entries in metadata profiles. Moreover, we can publish semantic-wiki-based documents that do not require any further activity to be profiled and included in a knowledge base as they are self-describing. The semantic wiki project is part of a collaborative knowledge management system that has been developed to support project teams and communities of interest.
Workers in a modern organization need to organize, share,
and retrieve knowledge that is collaboratively generated in
heterogeneous workgroups. Various knowledge management
(KM) systems have been proposed to support such
fundamental tasks. However, experiences showed the limited efficacy
of group tools and KM systems that prevented from a real
adoption by their users. In particular, we are interested in two
main socio-technical problems that can heavily invalidate the
process of knowledge creation and circulation within
organizations [
In the MILK project (Multimedia Interaction for Learning
and Knowing, [
Collaborative processes are enabled by effective profiling,
which is in turn enabled by agreed and shared ontologies
and by natural interaction means to exploit them. The issue
addressed in this paper is how to promote the involvement of
people in creating and exchanging knowledge by providing
them with an easy way of publishing group knowledge. The
chosen approach is to combine the definition and sharing of
personal ontologies with the creation of profiles and annotation
of documents using a wiki interface [
The rest of the paper is organized as follows. Section II outlines the MILK project, which is the environment that embed the proposed interaction techniques. Section III presents how ontologies are used in MILK. Section IV describes the exploitation of ontologies in creating MILK profiles, and semantic-wiki annotations. Section V presents related work. Finally, Section VI summarizes the paper contribution and outlines future work.
MILK is a KM system that allows a community of persons
to store, organize and retrieve documents in a multimodal and
multichannel way. The multi-tiered architecture of the system
(Fig. 1, [
At KM level, MILK defines a profiling mechanism that
unifies knowledge descriptions associated with documents,
people, projects and communities (collectively named
elements). Any kind of element can be included in the same
indexing and retrieval process. So, for example, given the
profile of a project, it is possible to collect information about
any other project that has similarities with the current one,
know who is interested in the same subject, and identify
documents regarding the same topics. The profiling process
is driven by ontologies that describe specific domains [
At interaction level, the aim of the system is to provide
users with various contextualized interfaces and interaction
mechanisms fitting different working situations, supplying
users with different views over the same contents, to promote
awareness and learning according to the activity the user is
undertaking [
The need of supplying multimodal and multichannel
presentation of documents — meaning, to be able to present the same
document in different formats according to the most suitable
device currently available to the user — led to an important
choice in MILK’s information architecture. MILK documents
are collection of files, each one proposing a representation
of the actual content of a document. For example, for each
document might, or rather should, exist a representation that
is well-suited for a regular PC (e.g., the full pdf version of
a paper), one for large displays (e.g., a slide-show presenting
the content), and one for handheld devices (e.g., the textual
abstract of the paper). It is important to remark that all these
files should be considered as being the same document or, at
most, different affordances of the same document [
The use of profiles to support indexing and document retrieval has the additional advantage of being more accurate and less computationally expensive than dealing with textual, or multimedia documents.
The MILK system has been designed — and developed — to be accessible through three main channels: desktop PCs, handheld devices (e.g., mobile phones and PDAs), and large screens for synchronous collaborative interaction. The last two devices are mostly devoted to information access. So, for example, a large interactive screen can be used to support a faceto-face meeting by giving access to appropriate documents and people.
Desktop PCs have the crucial role of supporting most knowledge and information creation. In this paper we concentrate on this aspect: how to improve keyword-based profiling and annotation of documents using a PC GUI.
MILK on the desktop helps people to manage information
and easily access the right sources for creating new
contents [
Through the desktop client, MILK users can access
documents in a shared repository and use a centralized ontology to
profile the documents they want to share. Moreover, (i) thanks
to the features described in the next sections, they can adopt
personal (multiple [
In MILK, the task of publishing documents is basically supported in two ways: automatic, via FTP upload, and userdriven, through the MILK GUI (Fig. 2). In the former case, the MILK archiving server can be mounted as an external drive for the file system (as it can be easily done with the most common operating systems) and documents can be dropped in a project or community folder; in the latter case, each document is uploaded via a browse-select-upload interaction protocol as it typically happens in interactive Web applications. In both cases, the MILK system reacts by creating a profile and by filling it with available information (e.g., user identity, taken from LDAP authentication, and current date and time) possibly augmented with automatically extracted keywords and keyphrases1 that are checked with respect to MILK’s ontologies2. Moreover, users can then check, correct, and complete the profile through the MILK GUI. In case the document is uploaded via FTP, the user will be prompted to review and complete the document profile at the next login.
The MILK GUI provides a standard way to use ontology terms, as keywords, in document profiles: a panel displays such terms in a classic tree fashion from which a term can be added to an element profile by simply dragging and dropping it. Otherwise, terms can be entered by typing them in the appropriate form fields. Typed terms can be part of the ontology or new keywords. In the latter case, they are collected and proposed to the authorized administrators of the system as candidate terms for a reference ontology that is shared organization-wide.
In its original version, MILK was limited to rely only on a centralized ontology. Such ontology was designed in different versions to accommodate the needs of the test settings in which the MILK system was deployed. Generic APIs and graphical user interfaces were implemented to directly manipulate ontologies. Ontology maintenance required administrators that were familiar with basic principles of knowledge engineering and with MILK technology.
Terms and relations in the reference ontology are used to compute correlations of elements. Moreover, the use of terms is monitored to collect additional information. This information is presented to administrators, and possibly endusers, with the graphical tool depicted in Fig. 3.
Taking into account both approved and candidate terms, the
tool computes occurrences and co-occurrences of the keywords
in document profiles and highlights the most relevant terms
and correlations. The identity of term users is collected and
used to evaluate the authoritativeness and reliability of the
proposal: user expertise is adopted as a parameter in the
presentation, so to weight co-occurrences according to the
experience gained by the person profiling a document [
In a graphical panel, terms are drawn as circles, and cooccurrences are depicted as lines among them. The relative relevance of a single term is shown by the size of the corresponding circle — the bigger the circle, the more used/important the term. Circles are blue when terms are already part of the shared ontology, and red otherwise (Fig. 3 presents a particular experiment). The strength of the correlation is presented by
2MILK supports a reference ontology and personal ontologies: the automatically extracted keywords are searched in the central ontology and, if not present, added as candidate terms for the reference ontology. adopting an expressive color palette for the line connecting two terms: blue, semi-transparent lines are used to show weak correlation (few co-occurences, or used by non-expert users), red, opaque lines are used to show stronger correlations (many co-occurrences, or used by expert users). Shades in between — of color and transparency — are used for intermediate values.
The tool can also be exploited to evaluate how the interests of the community or organization evolve over time: by monitoring terms that are gaining in importance (alone or clustered with others). Such information can be useful to administrators when maintaining the ontology: new terms that are already popular in the community should be quickly accepted.
Moreover, end-users’ awareness about the most relevant topics is a valuable added-value information, therefore, the same information regarding popularity and correlations can be presented to end-users to promote community awareness. For example, organization executives can easily see what is being produced, if there are new topics spontaneously becoming popular, and how the topics co-occur in published documents.
As a future work, we are studying a mechanism to make the administrators’ interventions on the shared ontology evident to users, as a feedback, that could be notified of changes on status of new terms.
A goal of our work is to design features that allow individuals to easily define personal ontologies that can be shared with colleagues, instead of only allowing users to type new keywords. Our approach is to integrate an ontology creation process with the most common desktop activity: Web browsing.
In the everyday activity, we use the Web to extend our knowledge by learning from documents and from terms exploited in the searching. The aim of the experiments we are conducting is to extend MILK beyond the basic activities on 2 1 shared ontologies and repositories to include the terms used during browsing as personal ontology terms. We are testing an extension for Mozilla Firefox that monitors the queries sent to Google (Fig. 4).
The user interface of the browser extension consists of a sidebar with an upper “permanent” panel, and a lower “temporary” panel. Whenever a Google page is loaded, the keywords used in the search are extracted by parsing the page URL and added to the bottom panel of the sidebar (step 1 in Fig. 4). The list of keywords in the panel is temporary, and it is reset every time Firefox is started. By right-clicking on any of the keywords currently displayed, the user can search the term again with Google (like with a bookmark) or make the keyword part of a permanent set (step 2 in Fig. 4).
When the choice to make a keyword part of the permanent personal ontology is made, a query can be sent to Swoogle3, via its Web Service APIs, to find out if the term is present in other existing ontologies. To see if matching terms exist, a new browser tab with the results of the query is presented to the user to let her select the most appropriate term and solve possible ambiguities. In practice, the user is requested to associate the term with an existing RDF namespace — picking among those returned by Swoogle — or to use a default one.
At the moment, direct manipulation of the set of terms stored in the permanent set — beyond the re-arrangement of the terms according to the is-A relation, which can be done with the extension — exploits external editors, such as Prote´ge´4.
Besides using terms as bookmarks for Google searches,
the general utility of creating a personal ontology is to use
it in KM systems such as MILK (i.e., by MILK GUI) to
better organize personal knowledge or retrieve information
from a shared repository: needless to say, it is the vision of
the Semantic Web [
For this reason we decided to be compliant with Semantic Web technologies: the upper panel relies on a RDF file that permanently stores a user’s personal ontology, that can be loaded and managed by the MILK client GUI (as described next) as well as by any other RDF-based application.
The ontology panel in the MILK desktop GUI (Fig. 2) is
composed of multiple tabs: besides the one for the shared
reference ontology — that is retrieved from the MILK server, —
others can be added to present the candidate terms and
personal ontologies (as the ones described by Boselli and De
Paoli [
When a document is profiled with terms that are only part of a personal ontology such terms are managed as new candidate terms with the advantage of carrying along additional information. That is, a portion of the personal ontology, rather than a simple keyword, is sent to MILK’s ontology management component: ancestors (according to standard is-a or part-of relations), descriptions and relations are used to qualify the terms whenever the users correctly specify the namespace of the ontology from which they took the term as a reference, or when users are able to maintain their own ontologies, e.g., using Prote´ge´.
To take advantage of this feature, MILK administrators
could exploit ontology merging techniques [
Further usage of ontologies, granting an ever higher expressive power, is achievable through semantic annotation, as described in the next section.
The MILK GUI is a standalone application, built on Eclipse Rich Client Platform5. The advantage of this choice is to deliver a (cross-platform) tool that is easily extensible, via plug-ins, to allow for rapid proofing of new ideas.
The interface is structured in perspectives that collect related views. For example, the View with Context perspective is composed of a main view, which fills most of the available space, surrounded by smaller panels displaying ontologies, related elements, project structure, and available people (Fig 2). The main view (i.e., the edit metadata view of Section II-A) is essentially form-based, meaning that metadata is entered in profiles by filling in the fields of a form. This is widely recognized in Computer Supported Cooperative Work (CSCW) literature as a tedious task. In particular, this kind of interaction
is not perceived as a collaborative activity. Forms suggest a rigid conception of profiles, whereas the MILK Metadata Management System was designed for more flexibility (e.g., to deal with heterogeneous elements).
The popularity of tools such as Wikipedia [
For these reasons we are testing a “Semantic Wiki
perspective” in the MILK client (Fig. 5). It is a wiki in the sense that
users can annotate elements’ profiles as free-text, and we claim
it to be a semantic wiki because of the enhancement it carries
in knowledge processing, being tightly integrated with MILK’s
ontologies [
In this sense, the semantic wiki perspective plays the important role of providing direct access to the creation of machine-processable documents, at the same time facilitating the human-computer interaction. In the current prototype, the only actions to be performed on a form are the assignment of a title, and the choice of the project or community of which the document will be part (otherwise the file cannot be correctly created on the archiving server). Then, an editor like the one outlined in Fig. 5 opens and the user can start typing the content.
As already described, MILK ontologies are presented in the panel on the left, and annotations can be made on the document by simply selecting words or sentences in the text area and then dragging one or more ontology terms on them. By this action, a RDFa6 annotation is created in the document, and the user perceives it through the change of the annotated portion of the text that turns underlined. Code-level editing features are also available, so that annotations can be added and removed directly in the source code. If the term is taken from the personal ontology, but it is not part of the shared MILK ontology, then the term is added to the list of candidates (with all its properties and relations) when the document is saved.
An example clarifies the use of the semantic wiki to support
collaborative ontology evolution; it refers to annotations in the
field of architecture [
Summarizing, users can edit profiles by adding annotations through the MILK Graphical User Interface (GUI) in three ways, in increasing order of carried information: 1) directly type a new term in the proper form field.
In this way no additional semantic information (i.e., definitions, properties, and relations with other terms) can be provided. In this case, the keyword is just a candidate term; 2) drag and drop a term from a personal ontology. In this case, an ontology fragment, rather than just a term, is proposed as a candidate; 3) use terms from the personal ontology in the Semantic
Wiki perspective, to create document annotations.
MILK’s wiki was designed according to a simple, yet
consolidated architecture [
MILK wiki
MILK client wiki interface article view editor view metadata view source view MILK Metadata Management
System Document
Repository Personal Ontology
PersoPnael rOsnotonloagyl
Ontology MILK Ontology
Manager Organization reference ontology
• the article view is a read-only display of the wiki representation of the document. With this view users can have a general view over the annotations of the document: annotations are outlined by the underlined text. Using mouse tooltips it is possible to view annotation details. Using XSLT and CSS technologies, the wiki description can be rendered on a range of devices or exported in various formats, with various appearances; • the editor view presents a compact editable view of the document: the text and the annotation can be modified by picking terms from the available ontologies. Also in this view, annotations are only shown by text underlinings; • the edit metadata view presents the list of the annotations in a form: metadata can thus be quickly edited and the changes are automatically inserted in the wiki document; • the source view allows the user to directly edit the source code of the wiki document (i.e., RDFa annotations): tags are shown and can be directly edited (Fig. 5) in the body of the document.
To make changes permanent, the document has to be saved: upon this action, the MMS is in charge of parsing the wiki document (the wiki perspective relies on Jena APIs to generate documents) to extract information and update the document profile. The RDFa document is stored in the repository along with other representations and documents.
Current implementation does not allow users to write or perform queries on the document repository directly from the wiki, for example using SparQL7 or XQuery8: the reasons are due to MILK’s knowledge discovery approach rather than to technical issues — the design or integration of a query engine able to translate the queries for the MMS could be very useful and is planned as future work.
parser (Jena) LDAP
MILK supports the definition of permissions and access rights and policies so to separate the creation and editing phases of a document from the browsing of the knowledge base. For example, only certain views are displayed when browsing documents, and the switch to “edit mode” requires an explicit action by the user (i.e., to click a button or open a different perspective). Thanks to such separation it is also possible to deploy different versions of the MILK PC client — for example, a Java Web Start browse-only version for users working from casual locations or shared workstations.
The work presented in this paper took inspiration from other
efforts made in developing semantic wikis. Among others,
SEMPERWIKI, developed by Oren et al. [
Moreover, thanks to the reuse of semantic terminologies — enabled by MILK client ability to load ontologies from various locations and by the Firefox extension to edit personal ontologies — users can be supported all across their main working and learning activities without making them quit the tools to which they are used.
The browser-extension approach is already widely in use to enrich the browsing experience with additional information. For example, the Sage extension9 makes an RSS/Atom feed reader available inside the browser as a sidebar, reducing the need to use an external application (i.e., feed aggregator) to automatically collect information about topics interesting to the user.
From a cognitive and interactive perspective, the process of
creating and editing a personal ontology with the extension
presented in Section III-B is similar to the organization of
regular Web bookmarks [
Other interesting research issues may arise from
investigating MILK’s interoperability with other tools for the Semantic
Web. For example, the knowledge base generated by MILK
users could be accessed by tools such as DBin [
In this paper we have presented the ongoing work to design
a KM tool integrated with the Semantic Web philosophy [
We believe that users will be more involved in the cooperative activities since they can add personal annotations and use their personal ontologies. Moreover, the possibility of discovering new terms and relations among documents is a rewarding feature to convince persons to collaborate.
Semantic interoperability (between users and applications)
is facilitated by (i) the capability of the Firefox Extension to
preserve references to standard namespaces (Section III-B),
(ii) MILK features for co-occurrence analysis [
Apart from the desired features mentioned along this paper, future work is the finalization and evaluation of the implementation, which is currently in a preliminary version that cannot be distributed to our test users. Other future works include the integration of email and RSS feeds (by means of a Mozilla Thunderbird extension or of a Chandler parcel) with MILK so that users can edit their ontologies and possibly also publish documents directly from the mail client.
This work was partially supported by the MILK project (IST-2001-33165): http://www.cootech.disco. unimib.it/milk/ Some of the features presented here as part of MILK have been introduced as results of the previous research in the ATELIER project (IST-2001-33064): http://atelier.k3.mah.se/