Digital Enterprise Research Institute www.deri.ie
Distributed architecture for
recommendations on the Web of Data
Benjamin Heitmann
 Copyright 2009 Digital Enterprise Research Institute. All rights reserved.
Chapter

About me
Digital Enterprise Research Institute www.deri.ie
 PhD student at Digital Enterprise Research Institute,
National University of Ireland, Galway:
“Distributed architecture for knowledge discovery in
linked data”
 computer science master from University of
Karlsruhe, Germany: Transitioning web application
frameworks towards the Semantic Web (2008)
 Philosophy was a minor subject for my masters
(information ethics, political philosophy)
 28 years old, German citizen, born in Switzerland,
grew up in south Germany (Baden-Württemberg, the
part of the South which is not Bavaria :)
Benjamin Heitmann

Research interests:
Digital Enterprise Research Institute www.deri.ie
 the architecture of the Web, the Semantic Web
and the Web of Data
 the influence of these architectures on the ability
to provide recommendations
 identifying and creating best practices and
guidelines for enabling recommendations on the
Web of Data
 engineering solutions like software components
and frameworks to provide recommendations on
the Web of Data
 understanding interplay between social uptake of
standards and architecture of the Web of Data
Benjamin Heitmann

Recent work: identifying common
components of Semantic Web applications
Digital Enterprise Research Institute www.deri.ie
Authoring
User interface
Interface
(92%) Search
(32%)
Service
(81%)
Data Interface
Integration (100%)
Service
(72%) Persistent Remote
Crawler
Storage Data
(35%)
(91%) Sources
from: Heitmann, B., et al., “Towards a reference architecture for
Semantic Web applications,” Proceedings of the 1st Int. Web
Science Conference, 2009
Benjamin Heitmann

Common components of a Semantic Web
application
Digital Enterprise Research Institute www.deri.ie
 Data Interface (100%): Abstraction layer regarding
implementation, number &distribution of persistence layers.
 Persistent Storage (91%): Persistent storage of data and run
time state.
 User Interface (92%): Human accessible interface for using
application and viewing data. (“read-only”)
 Authoring Interface (32%): Edit, create, import or export
data.
 Integration Service (72%): Merge Structure, Syntax or
Semantics of data from multiple heterogeneous sources.
 Search Service (81%): Search on content + semantic features.
 Crawler (35%): Retrieval of remote data for integration
service.
Benjamin Heitmann

State of the art: recommender systems
Digital Enterprise Research Institute www.deri.ie
 Problem: to much data to
be viewed by humans.
Application logic
 Pre-selection necessary!
 current recommender
Recommendation
algorithm
systems:
 one data source with one
data model
Data source  one recommendation
algorithm
 system fine-tuned for one
closed system, e.g. fixed domain (e.g. books)
Amazon book recommendation  closed, internal system
Benjamin Heitmann

Future research: distributed architecture for
recommendations on the Web of Data
Digital Enterprise Research Institute www.deri.ie
 distributed
Application logic
recommender systems:
 multiple data sources
Recommendation
algorithm  portable across domains
Data integration
 using linked data
 Challenges:
 identify stake-holders
Data providing
Data
integration
 which algorithms are
application
provider suited for such
Data
recommendations?
Data
source 1
Data
source 2
source 3  How do architecture and
algorithm influence each
other?
Benjamin Heitmann