This paper presents Pundit, a novel semantic web annotation tool, and demonstrates its use in producing structured data out of users annotations. Pundit allows communities of scholars to produce machinereadable annotations that can be made public and thus consumable as web data via SPARQL and ad-hoc REST APIs. Pundit is highly congurable and can deployed in custom instances to include well-de ned and agreed annotation vocabularies. Such instances can be distributed as bookmaklets to community users so they can create uniformly structured data in a certain application scenario. Basing on the provided APIs, some demonstrative applications have been developed, exploring di erent use scenarios, ranging from philosophy to journalism and cultural heritage. The main aim of this paper is to demonstrate how such uniformly structured annotations can be quickly re-used on the web to make information discoverable or to visualize it in interesting ways.
efrancesca.didonato@sns.it - http://www.sns.it/
Annotation is a primary activity for scholars and professionals. It consists in
enriching a content with some new information, which possibly helps in
understanding or searching the content itself. While until few decades ago annotations
were sketched by hand on the side of a book, today web technologies have the
potential to make them in nitely replicable, remotely accessible and easy to
share. Web annotations systems and bookmarking/clipping tools are popular
nowadays both among generic users (e.g. social tagging) and among scholarly
communities (e.g. Zotero4, Mendeley5 users). However, existing annotation
systems are generally limited to textual comments, tags or prede ned metadata
templates (e.g. bibliographic records). Furthermore, annotations are often
isolated into closed systems and very rarely are connected to the Web of Data.
The simple idea behind our work is that of making of annotations a vehicle to
create new semantic web data, actually adding links and, ultimately, knowledge
to the so called Global Data Space [
An exhaustive state of the art in semantic annotation goes beyond the purpose
of this paper and can be found in the literature [
Annotations in Pundit are essentially triples that connects di erent kinds of items together. A triple has the form [ subject - predicate - object ], where the subject and object can be segments of text and images (e.g. [text - describes image]>) or entities from the web of data (e.g. [text - has author - Dante(from Freebase.com)] or [image - depicts - Florence(from DBPedia.org)]). The most expressive annotation interface provided by the Pundit client is the "triple composer". It allows users to drag and drop items into triples, or select them from the web page (e.g. by selecting a text or an image), as well as searching into available vocabularies and data sources. However, other annotation wizards support speci c kind of annotations, as putting two segments of text in relation, or attaching tags and comments to a text segment. Image annotation of a segments of images is supported by a dedicated module as shown in 1. The Pundit client is a JavaScript application that can be deployed as a library, to then be easily included in existing web sites to make the content \annotable"9, as well as delivered as a bookmarklet. A bookmarklet is a simple link (bookmark) that, once added to a web browser allows loading Pundit on every web page and annotating its content.
In Pundit, an annotation contains information at a twofold level. The rst
one is the \annotation metadata" and deals with the act of annotating, including
information on the author, the time of creation and the involved web resources.
Pundit bases on the Open Annotation data model (OA)10 [
the annotated web resources, being them web pages segments or other kind of entities (places, persons, etc.) While no restrictions are applied and no assumptions are made by Pundit regarding the ontologies used in annotation graphs, a certain knowledge of the structure of the single annotations into a notebook has to be owned by a developer to implement a meaningful visualization based on such \free shaped" data.
So far, one of the most successful approaches to foster the reuse of data on the web is to create a consensus around vocabularies and ontologies within a certain community. In Pundit we try to follow this pattern by make it possible to deploy customized annotation clients, in the form of JavaScript libraries or bookmarklets, which can be distributed to users by \community leaders". A custom client possibly includes a precise set of a well-de ned set of \relations" to be used in annotations to create typed links among items or taxonomies where relevant web entities are collected and ready to be annotated. Both taxonomies and relations are represented in JSON and can be easily extracted from existing vocabularies (e.g. SKOS) or ontologies, as we did in the Wittgenstein's brown book pilot11. Aggregating items in collection, for sharing and publishing, is a common pattern in social clipping and bookmarking tools. In Pundit, annotations are collected in notebooks that users can optionally make publicly accessible. When a notebook is public, the annotations contained in it are not only shown in the Pundit client (e.g. when a user loads the Pundit bookmarklet on one of the annotated web pages) but, more interestingly, a notebook can be consumed by means of open REST APIs and accessed by a variety of web applications. Each notebook provides a SPARQL endpoint to query its content. In other words, a notebook is an independent RDF graph created by a given user in time and connecting a variety of web resources. 4
Regarding how to use annotations, and the semantic data they enclose, to drive end-user applications, there are mainly two \dimensions" that can be explored: { Annotation centric approach. This is commonly used in clipping systems where each clip is the result of a single annotation and is shown as a \box" containing some information (e.g. pictures, links, tags) about the annotated items. We mainly based on this approach in designing Ask, a prototype web application to search over public notebooks and manage personal ones. { Item centric approach. As annotations graphs in a notebook can be in fact consumed as a unique and bigger RDF graph, a possible way of looking at the data is that of focusing the visualization on the annotated items and their relations with other items. This approach clearly bene ts from an apriori knowledge on ontologies and custom vocabularies used in annotations, as it needs to take into account the nature of the information and deals with the \meaning" of annotations. 11 DM2E blog, Wittgenstein Brown Book experiment, http://dm2e.eu/dm2e-to-startwork-on-wittgensteins-brown-book/
Ask12 is a web application where public notebooks stored in Pundit can be searched and explored. At the time of writing a new version of the tool is being released. Personal notebooks are accessible to their owner and can be made public or kept private. By default, Ask provides a general purpose visualization of notebooks where single annotations are shown as \metadata boxes". However, alternative visualizations (such as the one described in the following sections) can be easily plugged by providing a compliant REST API. Ask is currently subject to intense development, and one of the most interesting recent features is the prototypal faceted browser available in alpha version13. 4.1
Edgemaps [
The same pattern can be applied to several contexts and to address very diverse use cases, as economics or journalism. The data journalism demonstrative application shows the use of annotations, in this case quotations from politicians and public persons taken from online news papers, to produce dynamic graphics. A Pundit bookmarklet has been deployed containing a small set of relations (or properties) to tag, describe and date in time politicians' declarations. The associated visual tool has been developed in JavaScript and provided as a web API, which gets a notebook (id) as argument and builds a timeline where annotated declarations are shown along with the trend of a nancial indicator. The idea is that of creating a tool for journalists to demonstrate and reveal possible existing connections among what important persons says and how the market behaves. (Fig. 4). A live demo can be found online18. 4.3
Timeline visualization has become a common practice for showing data containing time-related information and several tools already exist that allows creating such type of visualization. Instead of developing another one, Pundit reuses TimelineJS19. This is an example of the advantages of decoupling annotation 17 CiTO Ontology: http://purl.org/spar/cito/ 18 Journalism demo, http://ask.thepund.it/?#/timeline/31951d93/20120927 19 TimelineJS: www.timeline.verite.co creation and consumption. A pundit API has been created that allows to extract time-related information from a notebook given its id and to convert them in a TimelineJS compliant JSON to feed the timeline, as shown in Fig. 5. Annotations in the notebook simply need to have date information, i.e. they have to contain triples having as subject a text fragment or an image, as predicate \dates to" (for a date) or \start date" and \end date" (for a period in time) and as object a date.
A similar approach has been used in another experiment conducted in the
context of the Burckhardtsource.org platform that aims at mapping and
producing a critical edition of the extensive correspondence of 400 European
intellectuals with Jacob Burckhardt over a period of more than half a century from
1842 to 1897. 20 [
In this paper we presented preliminary results in leveraging on structured semantic annotations to create interactions and visualization of collaboratively 20 www.burckhardtsource.org 21 http://timeliner.okfnlabs.org/ created data. In our examples we used Pundit: a customizable and exible semantic web annotation tool. In deploying the tool for di erent use scenarios, we highlighted a simple pattern consisting of developing custom vocabularies, perhaps aggregating existing data, distributing a simple tool to annotate web resources of interest and, nally building on third party applications to consume the generated information and address speci c data visualization needs. 6
The research activity underlaying this work is being partially funded by the European Union's Seventh Framework Programme managed by REA-Research Executive Agency22 ([FP7/2007-2013][FP7/2007-2011]) under grant agreement n. 262301, and by the GramsciSource project of the MIUR, FIRB 2012, p. RBFR12MZ8R 003. Pundit was originally developed in the Semlib project23.