Preserving the Web of Linked Data

There are many sides
to preservation.

Web of Linked Data?

We are loosing thousands of Alexandria libraries each day

We have lost so much of the early Web history, just as we have lost so much of early Human history.

Kalev H. Leetaru - University of Illinois

The forces of decay

• Link Rot
• Content Drift

Link Rot

Content Drift

Significant change in content within a 3-Month Period

Strategies

• Observational: perceived as discrete
  • Snapshot
  • Web archive
• Historical: perceived as continuous
  • Versioning systems
  • Transactional
  • Notification-based

Memento: travelling to the Web of the Past

https://tools.ietf.org/html/rfc7089

Archive or Archiving?

Linked Data archiving as the product

• RDF indexes for versioning
  • Dydra, Virtuoso, XRDF3X, ...
• Representations of versions, provenance & time:
  • PROV, LDPatch, LODE, ...

Technical
(Increasingly) Popular research tracks.

Linked Data archiving as the process

• Some technological building blocks
  • Linked Data interfaces, change detection, publishing, crawling & querying

Technical, as well as Infrastructural & Societal.
Rather unknown territory (but there are technologies).

What assumptions are there about data evolution?

• Historical Data
  • Provenance is a timeline.
  • Only truth can exist at the same time.
  • Timeseries databases, Wikipedia
• Versioned Data
  • Provenance is a directed acyclic graph.
  • Multiple truths can exist at the same time.

Decay becomes more complex

• Link Rot
• Content Drift
• Concept Drift
  • "Please don't change your vocabulary"
    (Check out DRIFT-A-LOD workshop)
  • Problem in other domains as well (Machine Learning)

Study these issues within Linked Data

• Link Rot
  • Subject or Object cannot be dereferenced
  • Dataset/Interface is gone
• Content Drift
  • Context graph of Subject or Object has changed
• Concept Drift
  • Predicate or Object change meaning

Archiving for the
Reproducibility of Query results

• Sustain the validity of claims
• Backwards compatibility of applications

Federated querying is highly affected

How to shape a decentralized Quality of Service?

The Hyperlink is the simplest form of decentralization,
which we are already failing to preserve.

Persistent Identification

Persistent Identification

• Dependency on publisher registering the PIDs
• Possible loss of connection between PIDs and the original
• Dependency on the PID provider

Possibly replacing one potential Link rot problem by another

Who are you to tell me my URI is not persistent?

ISWC Resources track:

Consensus on and trust in persistence in a decentralized Web:
community-driven? standardization? blockchain,...?

Robust links

	<a href="B"
	data-versionurl="URL of snapshot of B"
	data-versiondate="datetime of snapshot of B">

http://robustlinks.mementoweb.org/spec/

Robust Links

Open Annotation
& Memento vocab

Can be linked
to PROV

Who will be responsible for archiving?

• Publisher
  • Snapshot
  • Versioning systems
• 3rd party
  • Traditional
• Hybrid: Publisher and/or 3rd party
  • Transactional
  • Notification-based

Data archiving intrests more than curators & activists

For instance, Data driven journalism.

• Product: transparency of the editorial process
• Process: interaction with users, public

Scolary communication, cultural heritage, legal publications, community databases (Wikipedia & Wikidata)

Archivability of Linked Data

Linked Data is in essence easier to archive.

• Raw, self-contained data
• Already machine processable/understandable
• No obfuscation by client-side scripting

Accessibility of content to stimulate archiving.

The content in HTML+RDFa that dokieli produces is accessible (readable) without requiring any CSS or JavaScript, ie. text-browser safe. Breaking this "rule" in future development should be considered an anti-pattern (or a bug) in dokieli.

dokieli documentation, Sarven Capadisli

Choices in Linked Data interface
increase or decrease archiving.

Intelligent Server

High resource granularity

Intelligent Client

Data not as accessible

Need to participate in archiving process

Prevent mistakes from the past in standardization

• Query interfaces: what can be archived?
• Protocols: is it accessible?
• Domain Modeling: can the semantics be preserved? How to select the subgraph?

There are many sides
to preservation.

We don't start from scratch,
many technologies are there.

Start covering the uncovered sides.

Add archiving to the discussion.