One of the powerful and popular tools that are used to support Collaborative Knowledge Engineering are Semantic Wikis. They are easily accessible and provide ACL mechanisms, but they lack a good versioning mechanism. In this paper an extended version of such a mechanism is proposed. Besides elements that appear in every Wiki system, like simple changelog and place for discussion, it incorporates changes ontologies, rich metadata, semantic metrics and reasoning unit tests. All of them are gathered into the form of the provenance graph that can be serialized into Turtle syntax and (automatically) analyzed. First prototype of such a mechanism for Loki wiki called BiFroST was developed.
Knowledge bases are created not by a single knowledge engineers, but by a team of people who either sit in the same room (in a company) or people who do not even know each other (as in the Wikipedia project1). It can therefore be easily seen that there is a need to move from traditional Knowledge Engineering (KE) to Collaborative Knowledge Engineering (CKE) and to consider issues related to the collaborativeness.
Primarily, following issues should be taken into account, cf. [
Among tools that support CKE teams and tries to address these issues are: (a) WebProtege2, a collaborative version of open source ontology editing tool ? The author pursues his doctorate under the supervision of Grzegorz J. Nalepa. 1 See: https://www.wikipedia.org/. 2 See: http://webprotege.stanford.edu/.
Protege that was used, inter alia, in the process of the International Classi cation
of Diseases (ICD-11) preparation, where 270 doctors have worked together on the
base that consists of 45k classes [
In this paper, one speci c group of tools that supports CKE teams will be considered. They are called Semantic Wikis and they are quite interesting because of their popularity (see section 2). Like traditional wikis, they are conceptually simple but powerful for massive cooperation, and also provide a semantic layer on which one can de ne concepts and relations that are easily machine-understand.
What is important from the CKE point of view, is the fact that every Wiki system has integrated collaboration mechanisms such as ACL (Access Control Lists), which allows for gaining di erent access levels for di erent users and version control for tracking all changes that was done within wiki pages. Among Semantic Wikis' drawbacksis the fact that they provide only simple \straight" changelogs in which all versions are saved with small comments, as well as some place to discuss editions, but nothing more. This is not enough to really support CKE within wikis, because \classical" changelog does not provides any way to specify used resources nor to automatically evaluate con icts.
In this paper extended versioning mechanism for Semantic Wikis is proposed with taking care of CKE collaborative issues presented above. It is based on a semantic graph in which changes are related to each other as well as to some additional information, what gives a possibility to automatically analyse the changelog and provide some feedback about collaboration quality.
The rest of the text is structured as follows. In part 2 Semantic Wiki systems and their capabilities are characterized. Section 3 outlines the approach. The paper is concluded in section 4. 2
Wiki systems are a kind of containers for text pages (\wiki pages") and media les (images, audio les, and others). Wiki pages are human readable plain text les that uses special structures to denote di erent types of content (e.g. headers, images). Pages within a Wiki are usually grouped within the so-called namespaces and identi ed by unique names. They are also linked to each other creating a hyperwikitext structure4.
Semantic wikis enrich the wiki technology with semantic information. They
are quite popular, as evidenced by the large number of semantic wiki systems.
Examples are: KnowWE [
Loki [
Semantic Wikis were used in many projects, ranging from historical ones, like
Catalogus Professorum Lipsiensis, knowledge base that describes life and work
of 1.400 professors of Universitat Leipzig in 1409-2009 [
Extended versioning mechanism for Semantic Wikis proposed in this paper is based on the PROV graph6, a W3C standard prepared for describing data provenance. PROV graph represents relations between entities (wiki pages and their revisions), activities and agents (users). E.g. the graph states that the first revision of a page welcome was created on 31/08/2016 by Chris (what consists of three triples that connect selected wiki page{subject, with main page, creation time and author)7. New facts (triples in the RDF form) are added to the changelog after one of three possible user activities: page creation, edition and deletion. The graph incorporates also pieces of information provided by modules, which were selected in order to most comprehensively catch the properties of the changes (see Figure 1): Changes ontologies: main activities within Wikis are page creation, edition and deletion, but these ones do not exhaust the issue. We should take into account that there are di erent types of changes (see Section 1) and each edition is stated by the factual change (What was done? E.g. Typos or other small bugs xed or New content added ) and the goal (Why it was done? E.g. Errors xing or Knowledge database expansion) that form some kind of ontologies (one for every task's \type", e.g. one for preparing conference papers and the other for developing system speci cation within wiki). Rich metadata: besides \standard" timestamp and user, the framework also provide more information about each page edition: internal (other wiki pages) and external (e.g. books) sources of knowledge in the form of an URI, as well as about user: name, e-mail address, etc.
6 See W3C documentation on: https://www.w3.org/TR/prov-primer/. 7 Detailed speci cation of created triples is provided on the page: http://loki.ia.
agh.edu.pl/wiki/docs:prov#how_the_prov_plugin_works.
Semantic statistics: how many concepts / instances / relations were added / removed / corrected in this revision? Reasoning (unit) tests: as an analogy to the unit tests in software engineering, we added a possibility to de ne test cases that will be executed during page saving to ensure the minimum level of knowledge quality.
Issue tracker: it is not possible to cover all aspects of changes by automatically computed characteristics (semantic statistics and reasoning tests mentioned above). There should be also a place for discussion between experts.
The advantage of the proposed solution over \classical" changelog within CKE setting is based on the power of the PROV graph based on the RDF triples that can be (automatically) queried with SPARQL. With this in mind exemplary use case scenarios are indicated, as a response for issues described in Section 1: Get rid of bad changes. With tests' statistics, one can quickly identify which changes are bad and should be examined. In a more rigorous case, a wiki's administrator can block the ability to save changes if new revision is worse than previous one (less tests were passed).
Sources analysis. By combining tests' statistics and sources lists, to determine which sources have low quality and shouldn't be used in the future. User types identi cation. Thanks to placing the solution on the top of changes ontology, we can identify di erent kinds of users, e.g. good users and bad users (who introduces bad changes) or creators (they add a lot of text), annotators (they provide many new relations for existing text) and proofreaders (users that mainly x spelling).
Underdeveloped pages indication. Semantic statistics provide information about how many concepts and relations are stated on the page. If there are not too much, maybe it is a good time to pay attention to the page? Motivation by gami cation. Accurate metrics allow for awarding points to users, giving them badges for di erent tasks (e.g. badge for adding 5 new relations in the wiki pages) and creating leaderboards. These ones can motivate them to create better knowledge base what is the main goal.
The BiFroST (BiFroST Framework for Semantical Tracking), a prototypical implementation of proposed versioning mechanism for Loki Semantic Wiki is being developed and tested in collaboration with students at AGH-UST8. At the time of submitting the paper, PROV graph is created within framework (see Figure 2). This rst prototype was used in small experiment to collect triples about 940 user activities. Practical implementation of the SPARQL queries scenarios presented above is a future work. In this paper the enhanced semantic versioning mechanism is proposed as well as prototypical implementation called BiFroST. Besides elements that appear in every Wiki system, like simple changelog and place for discussion, the proposal incorporates changes ontologies, rich metadata, semantic statistics and reasoning unit tests. All of them are gathered into the form of the provenance graph that can be analyzed using SPARQL queries. First prototype implementation was developed and used in small experiment at AGH UST.
Acknowledgements. The paper is supported by the AGH University of Science and Technology Grant 15.11.120.855.