Semantic Technologies for Decision Support

- A Pattern-based Approach

STeDS is a research project at IDA-HCS, partly financed by CENIIT. The project started on January 1st 2012, and the project leader is Eva Blomqvist.

Abstract. This 6-year project aims to assist the adoption of Semantic Web technologies in Decision Support Systems (DSS), based on the use of Ontology Design Patterns (ODPs), with specific focus on industrial actors. ODPs have in other contexts been successfully applied to reduce the complexity and cognitive load of using semantic technologies for both developers and users. Moreover, this project will create a forum for research on semantic technologies at IDA.

Project Overview

Project staff includes the project leader (Eva Blomqvist) and currently - september 2012 - one Master student (Robin Keskisärkkä). Industry partners include Saab, VSL Systems, FOI, and the SPAWAR Pacific (US).

The research is conducted based on the following general questions:

(Q1) What tasks (functionalities) in a DSS can be improved through semantic technologies, and what is the nature of the improvement?
(Q2) How can technologies and methods (those relevant according to Q1), be adapted and specialized to fit DSS, and in particular industrial DSS development?
(Q3) In what way do ODPs facilitate the practical creation, maintenance and usage of the formal models (ontologies) required by the semantic technologies (resulting from Q1 and Q2), and what ODPs are needed?

Vision and Goals in Short

The vision of the project is to make semantic technologies available to, and widespread in, industrial DSS. Short term goals include the identification of tasks within DSS that can be supported by Semantic Web standards, technologies, and methods, as well as adapting the technologies and methods for the needs of DSS, i.e. supporting DSS developers and users. The use of ODPs will ease the uptake of the technologies, as well as provide ready-made configurable ontological components.

On a more concrete level, this project will produce, apart from new knowledge in the field, the following tangible results:

A software framework for semantically enhancing DSS, consisting of components representing empirically well founded solutions based on semantic technologies.
ODPs to facilitate easier building of the formal models that the semantic technologies rely on.
Specialized methodologies for development of the semantic parts of a DSS, including evaluation and validation of the semantic components and their contribution to the system.

Project Plan Summary (from the original application - for 1st year results see below)

The project will explore the three research questions in parallel, starting from a list of hypotheses of where semantic technologies can contribute. Initially five hypotheses, based on our experience of semantic technologies and DSS, respectively, were given:

Situation detection - Given a stream of data, e.g. sensor data, semantically representing it utilizing ODPs, and detecting complex events (c.f. CEP) by performing reasoning on the stream.
Information filtering and integration - Given a large set of data from heterogeneous sources, filtering the data based on ODPs able to express a unified view of the relevant parts, i.e. matching the original models to the ODP model, and using it to query the data (c.f. [16] sect. 6).
Information enrichment - Given an ODP as a view on data, extracting additional data for that ODP from text or structured sources, i.e. using the ODP model for Information Extraction.
Model extension - Given a set of input data, semi-automatically develop or extend a formal model of that data, tailoring it to a specific DSS task by applying ODPs (c.f. method in [19]).
Tracking and sharing decisions - Given a decision-making process, tracking, sharing and comparing decisions, and applying metrics on the process, e.g. efficiency of information flow, c.f. the preliminary work of the Decision and Decision-making W3C incubator.

Starting from the first hypothesis, the following tangible results will be delivered after the first year (January 2013):

ODPs, similar to the ones in [10,19], to support detection of situations, i.e. complex events, and a specialization of an existing methodology, e.g. [14], to create the ontologies needed.
A software prototype realizing CEP for DSS, e.g. based on [5,18] in combination with ODPs.
A component-based architecture design for the general semantic software framework.
A revision of the hypotheses (including priorities), and a revised project plan for the following years.

Results during the first 8 months of the project (until August 2012)

Activities and results

During spring 2012 the main focus of the project has been on confirming, prioritizing and revising the hypotheses that are expressed above. Through an extensive literature study (about 60 papers have been reviewed) as well as interviews with 7 representatives of industry and non-university research organizations (i.e., all the industry partners listed in the original application, plus two additional organizations), a new list of hypotheses has been established and appropriately prioritized. A full report of the study has been accepted for publication in the Semantic Web Journal (IOS Press), a preprint can be found here. The revised list of contribution opportunities, to be used for this project, consists of the following hypotheses (in order of priority):

Information integration - in particular the integration between information provided by human actors and existing structured data.
Flexible information filtering - in particular the opportunity to have context- and user-specific filtering that evolves along with user needs.
Information aggregation, event and situation detection - beyond the plotting of raw information on maps or in graphs, i.e., truly "intelligent" data analysis is needed. Timeliness and flagging of potentially important situations, or events is essential, as well as information summarization together with the possibility of drill-down into more detailed information.
Model evolution - to be able to handle changes and concept evolution in the real world, without extensive human intervention.
Decision sharing - conveying the meaning of decisions to external parties.

Of less importance was the tracking of decisions and post-event analysis, which has therefor been taken off the list for the time being. Most likely it is not irrelevant, but because the above list consist of prerequisites to such analysis, which have to be solved first, our industry partners want to see them solved first. Also, the information extension issue has been taken off the list, since proper information integration and filtering are prerequisites to it, and none of the interviewees from industry viewed extension as important at this point in time due to the fact that there are more pressing issues to solve.

In addition to this survey, a Master student has been engaged in the project (PhD student is still to be employed, potentially it is the same person). The student is currently exploring av evaluating existing semantic CEP solutions, in order to build a software prototype realizing CEP for DSS. The current focus relies mainly on existing technologies, but will be enhanced by ODPs during the coming months. This work is mainly related to the third contribution opportunity in the list above, however, in our previous list of hypotheses this was ranked as number one, whereby this was the natural place to start. In order to start covering the issues that received a considerably higher priority after the interviews with industry representatives, we will in parallel study how linked data approaches can contribute to information integration and filtering, and in particular explore the possibility of using ODPs as information filters.

In terms of activities supporting the goal of bringing together semantic technologies research at IDA, a seminar series and a mailing-list have been started. During spring 2012 five seminar sessions were held, where four were focusing on semantic technology research at IDA from at least three distinct divisions (ADIT, AIICS and HCS). One of these seminars consisted of a presentation by Fredrik Heintz, project leader of the related CENIIT-funded project on stream reasoning. Further collaboration between these two projects is going to be explored in the future. In addition it should be noted that the seminars have also been visited by researchers outside IDA, e.g. from ISY, as well as some of the industry partners of this project, which indicates that these topics are also relevant outside the department and that the seminars fill an important community-building function.

The fifth seminar was given by an invited speaker: Prof. Fabio Ciravegna, from Sheffield University, speaking about the intersection between NLP and semantic technologies, and its use for collective intelligence and situation awareness applications. For the autumn we are planning three more seminars, to be held in october and november, where one will be held by an invited speaker; Dr. Valentina Presutti, ISTC-CNR, Rome, Italy.

Currently an ongoing effort resulting from the visit of Fabio Ciravegna is the writing of an FP7 project application for the latest Security call, which concerns the use of social media analysis for making sense of what is going on in the real world during security-critical events such as demonstrations and riots. By applying CEP on top of existing data analysis technologies, we will be able to provide better DSS to emergency responders and police, in the context of social unrest situations. If funded, the project will be a complement to this project and will finance the remaining part fo the PhD student (to be employed).

Another project application that has been submitted, is for a research project for young researchers at Vetenskapsrådet. That application has a similar focus as this project, and would complement it by allowing the PhD student to work full time on these topics. A third project application has recently been submitted to Energimyndigheten. This is a joint effort with the IEI division of Energy Systems (LiU) where we aim to test and apply semantic technologies for information integration and filtering to a case within the area of mapping energy efficiency of enterprises (energikartläggningar). This in order to start covering the most highly prioritized issues above in some real-world case.

In summary, the project has so far produced (or contributed to - in case of project applications) the following tangible results:

A journal article accepted for the Semantic Web Journal, summarizing the survey (literature survey and interviews) results concerning industry needs and requirements of semantic technologies in DSS.
A seminar series and a mailing-list as the start of a community around semantic technologies at IDA and beyond.
A project application sent to VR (response expected in November).
A project application sent to Energimyndigheten (response expected in October).

In addition to this, ongoing activities include:

A study performed by a Master's student (financed by the project) on the implementation and evaluation of current semantic CEP technologies.
An EU project application being developed jointly with the University of Sheffield, as well as one of the industry partners of the project (VSL Systems).

Project publications:

Blomqvist, E.: The Use of Semantic Web Technologies for Decision Support - A Survey. To appear in: Semantic Web Journal, IOS Press, 2012.

Planned activities (end of year 1 + year 2)

During the coming months, until the end of the first project year, we aim to finalize the study currently conducted by the Master's student, and in addition have a PhD student employed (potentially the same person). In addition, now that we have grounded our hypotheses in real world needs, we will look further into what kind of ODPs are needed for these particular tasks (starting from the top of the prioritized list) and how the necessary models can be constructed based on them.

As we have now grounded our hypotheses in industry experience, and turned them into a set of prioritized focus areas for the project, we will continue to develop some preliminary solutions for the problems with highest priority during year 2. The PhD student will focus on the third subject area (semantically enhanced CEP methods), where the Master's student has been working during 2012, while the project leader will focus on the first two priorities (in the updated list - information integration and filtering), as well as on developing appropriate ODPs. During year 2 of the project the first attempt towards a comprehensive software architecture will also be set up, and the focus of research will turn more towards evaluation of the prototypes already set up during year 1, as well as other current technologies.

The following tangible results will be delivered after the second year (January 2014):

ODPs, similar to the ones in [10,19], to support information integration and filtering, including information drill-down and summarization, and a further adaptation and improvement of the specialized methodology, e.g. an adapted version of [14], to create the ontologies needed.
A thorough evaluation (with industry involvement) of the prototype and existing methods for semantic CEP resulting from the Master's student's work in year 1, analysing its contribution to DSS in industry use cases.
A second version of the software prototype realizing CEP for DSS, including improvements based on our current evaluation of existing technologies (to be reported by the end of 2012) and the industry evaluation to be performed at the beginning of year 2 (see above).
A first software prototype (demo system) realizing the information integration and filtering tasks for a specific DSS case, e.g. in the energy domain or for social media monitoring (depending on the success of our project applications), based on the use of ODPs.
An improved component-based architecture design for the general semantic software framework.
A revision of the hypotheses (including priorities) if any new evidence from industry arises, and a revised project plan for the following years.

Detailed Project Description (updated version of initial application)

Background

The Semantic Web [1] has been researched for more than ten years; nevertheless, the techniques have only to a certain extent been applied to Decision Support Systems (DSS). At the same time organizations have to cope with an increasing information overload [2-3], thus increasing the need for DSS. Relying on experiences from Content Management Systems (CMS), we note that until recently few CMS used semantic technologies. However, in recent years this has changed, since: (i) Semantic Web standards have emerged, leading to the availability of stable and scalable software frameworks, and (ii) projects such as IKS have developed specialized methods and tools applying specifically to CMS. To some extent, similar methods are relevant for DSS and other frameworks are under development, e.g. distributed Web-scale reasoning [4], and stream reasoning and Complex Event Processing (CEP) [5] in OWL, c.f. the LarKC project. Still, there is a need for methods allowing industry actors to more easily adopt and adapt current semantic technologies to DSS.

We focus on DSS specifically for delivering the right information to the right person, at the right time and location, with an appropriate quality, to meet the information demand of a decision task. This may imply information filtering, enrichment, and context awareness. Information quality can include time aspects, or tracking provenance and assessing reliability. Such a DSS should exploit the recent success of Semantic Web technologies, but doing that it needs to address the following challenges:

There is a reluctance to adopt semantic technologies in industry, since they are (i) perceived as "difficult" and complex, e.g. requiring knowledge of logical formalisms, and (ii) because there are in many cases a lack of empirical evidence showing the benefits of such solutions [6-8].
Few research projects have studied (i) what semantic technologies are suitable for a DSS setting, (ii) what tasks within a DSS they can solve, and how they need to be tailored to DSS, and (iii) in what way those technologies increase the quality and/or performance of existing solutions.

These challenges are confirmed by industry interested in applying semantic technologies. The project will thus be conducted in collaboration with several industrial actors, e.g. Saab and VSL Systems, as well as research organizations, e.g. FOI and the Space and Naval Warfare Centre Pacific of the US Navy. These organizations will, through their long experience of DSS, provide use cases and requirements of semantic DSS, and act as validators and potential adopters of project results.

Design Patterns (DPs) have proven effective in other fields, e.g. software engineering. DPs are well-tested and consensually agreed solutions to recurrent problems. DPs for semantic technologies [9-11] are still in their infancy, but play an important role in the adoption of semantic technologies, and are at the very forefront of semantic technology research. The term Ontology Design Pattern (ODP) was coined simultaneously by the project leader [12], and Aldo Gangemi [13], in 2005. Since then, we have also proposed a pattern-based ontology design methodology [14] and empirically evaluated it [15].

Applying ODPs in DSS is a completely novel approach, which facilitates the adoption of Semantic Web technologies (c.f. 1(i)), and is a means to tailor technologies for use within DSS (c.f. 2(ii)). The type of ODPs to be used are Content ODPs (CPs), which can be manifested as small models, and described in a simplified manner as a tuple CP=<R,V,O>. R is a set of ontological requirements expressing the tasks the CP solves, e.g. inferences or queries, V a set of terms expressing its vocabulary, and O a set of logical axioms using V as lexical grounding. In the rest of this text, we let ODP refer to CPs as described here, more specifically we focus on CPs where O is expressed using OWL (a W3C recommendation); additionally we will use the term ODP model to refer to O.

Project Plan

The project will last for the coming 6 years (2012-01-01 - 2017-12-31), and the research will be conducted based on the following questions:

(Q1) What tasks (functionalities) in a DSS can be improved through semantic technologies, and what is the nature of the improvement?
(Q2) How can technologies and methods (those relevant according to Q1), be adapted and specialized to fit DSS, and in particular industrial DSS development?
(Q3) In what way do ODPs facilitate the practical creation, maintenance and usage of the formal models (ontologies) required by the semantic technologies (resulting from Q1 and Q2), and what ODPs are needed?

Research will be conducted so that these questions are addressed partly in parallel, i.e. Q2-3 are studied based on a hypothesis of Q1, then verifying it through empirical studies (returning to Q1). Q1 implies studying current literature, existing (and future) DSS, as well as exploiting existing methods for evaluating semantic technologies. Q2 will use experiences and software from IKS (the project leader is actively involved in IKS and software is already available), together with existing results from other research projects, e.g. LarKC. The previous research of the project leader, concerning ODPs and related methodologies will be the starting-point for answering Q3. Catalogues of ODPs are also available, e.g. such as the ones collected in a bottom-up fashion by online community portals.

Five hypotheses, based on our experience of semantic technologies and DSS, respectively, were initially given (for the revised list, see top of this page). The list will be subject to change based on literature, industry requirements, and empirical evidence. Each hypothesis includes the general task, and initial ideas for semantic-based solutions:

Situation detection - Given a stream of data, e.g. sensor data, semantically representing it utilizing ODPs, and detecting complex events (c.f. CEP) by performing reasoning on the stream.
Information filtering and integration - Given a large set of data from heterogeneous sources, filtering the data based on ODPs able to express a unified view of the relevant parts, i.e. matching the original models to the ODP model, and using it to query the data (c.f. [16] sect. 6).
Information enrichment - Given an ODP as a view on data, extracting additional data for that ODP from text or structured sources, i.e. using the ODP model for Information Extraction.
Model extension - Given a set of input data, semi-automatically develop or extend a formal model of that data, tailoring it to a specific DSS task by applying ODPs (c.f. method in [19]).
Tracking and sharing decisions - Given a decision-making process, tracking, sharing and comparing decisions, and applying metrics on the process, e.g. efficiency of information flow, c.f. the preliminary work of the Decision and Decision-making W3C incubator.

For assessing the nature of the improvement (c.f. Q1), both quantitative and qualitative evaluations will be used. Quantitative methods include evaluating effectiveness of information delivery, e.g. through precision, recall, and quantifying usability, e.g. through SUS [17], as well as efficiency, e.g. performance and scalability measures. The SEALS platform provides unbiased evaluation datasets, measures, benchmarks and an evaluation software framework. Qualitative methods relate to user satisfaction, perceived usefulness, and ease of use of the semantic technologies for certain tasks.

The first year focuses mainly on the first hypothesis, i.e. situation detection, sometimes called Complex Event Processing (CEP) [5,18]. The hypothesis is based on observations from several organizations. For instance, in situation awareness and monitoring systems for municipalities (as developed by Saab), low-level sensor data needs to be aggregated and transformed into "situations", which make sense to a user. In a combat management system (as observed at Försvarsmakten), the situation is continually monitored by officers leading the combat, to plan their next move and predict future developments. Finally, during training of emergency managers (as done by VSL Systems) there is a need for monitoring the development of an exercise, e.g. to assess if a training session is proceeding according to plan and to detect deviations. Our hypothesis is that these problems could be supported by semantic CEP and stream reasoning. Concretely, the problem can be described as a process characterized by the following input and output requirements:

Input:: (1) A stream of data representing the state of the environment, as well as (2) background knowledge of the domain, which in turn includes a characterization of a set of situations (expressed by means of ODP models) that are relevant from a user perspective.
Output:: (1) A stream of complex events representing the current situation in a way that is useful and makes sense to the end user, and where (2) any user-relevant "situations" are marked.

After the first year, a prototype based on Semantic Web technologies, e.g. ODPs, will be presented. During the second year, the prototype will be evaluated (according to Q1). Parts of the prototype will also be generalized, to constitute a first component of the general software framework. In parallel, the other four hypotheses (tasks) will be analysed in detail, to determine (i) their relevance for DSS, as well as (ii) detailed plans for including such functionalities in the project results. Thus, the following tangible results will be delivered after the first year:

ODPs, similar to the ones in [10,19], to support detection of situations, i.e. complex events, and a specialization of an existing methodology, e.g. [14], to create the ontologies needed.
A software prototype realizing CEP for DSS, e.g. based on [5,18] in combination with ODPs.
A component-based architecture design for the general semantic software framework.
A revision of the hypotheses (including priorities), and a project plan for the following years.

Industry Involvement and Research Environment

Several organizations and companies have expressed interest in the proposed line of research, and will be actively involved during the project. The founders of the previous W3C Incubator group for Decisions and Decision-making, i.e. the Space and Naval Warfare Centre Pacific and MITRE corporation (both in the US), will provide practical experiences and challenges. Nationally, the project includes collaboration with FOI, VSL Systems AB, and Saab AB, which are all involved in research and development of different aspects of DSS.

The project will be conducted at the MDA lab (IDA), but will connect other researchers at IDA who are already working on Semantic Web-related topics, e.g. ontology matching (IISLAB), logical formalisms and rules (TCSLAB), language processing (NLPLAB), and stream reasoning (KPLAB).

Project staff constitutes the applicant (40% of full time), and a PhD student (80% of full time, partly financed by CENIIT). The (long-term) goal is to create a research group for semantic DSS, acting as a focal point for semantic technology research at LiU.

In an international perspective the research will complement, and benefit from, several current research projects, as mentioned. The research will also be conducted in close collaboration with international research groups, e.g. STLab at ISTC-CNR (Italy), who are partners of the IKS project. The project is also related to other CENIIT projects, e.g. Stream-Based Reasoning Grounded Through Sensing. While the existing project studies the basic mechanisms of stream reasoning, without restricting the underlying logical formalism, our project aims to focus on data represented through Semantic Web standards, e.g. RDF/OWL.

References

T. Berners-Lee, J. Hendler, and O. Lassila: The Semantic Web. In: Scientific American, V. 284 , N. 5, 2001.
J. Spira, and C. Burke. Intel's War on Information Overload. Basex, 2009.
J. Doomen. (2009), Information Inflation. In: Journal of Information Ethics, 18 (2), 2009.
J. Urbani, S. Kotoulas, J. Maassen, F. van Harmelen, and H. Bal. OWL reasoning with WebPIE: calculating the closure of 100 billion triples. In: Proceedings of ESWC 2010, Heraklion, Greece, Springer, 2010.
D. F. Barbieri, D. Braga, S. Ceri, E. Della Valle, Y. Huang, V. Tresp, A. Rettinger, H. Wermser. Deductive and Inductive Stream Reasoning for Semantic Social Media Analytics. In: IEEE Intelligent Systems, V. 25, N. 6, 2010.
T. Heath, J. Domingue, and P. Shabajee. User Interaction and Uptake Challenges to Successfully Deploying Semantic Web Technologies. In: Proc. of The 3rd Intl. Semantic Web User Interaction W.s., 2006.
T. Kazic. Factors Influencing the Adoption of the Semantic Web in the Life Sciences. In: Semantic Web: Revolutionizing Knowledge Discovery in the Life Sciences, Springer, 2007.
A. L. Rector, and R. Stevens. Barriers to the use of OWL in Knowledge Driven Applications. In: Proceedings of OWLED 2008. Vol. 432 CEUR-WS.org, 2008.
F. Scharffe, Y. Ding, and D. Fensel. Towards correspondence patterns for ontology mediation. In: Proceedings of The Second International Workshop on Ontology Matching, 2007.
A. Gangemi, and V. Presutti. Ontology Design Patterns. In: Handbook of Ontologies, 2nd edition, Springer Berlin, 2009.
F. van Harmelen, A. ten Teije, and H. Wache. Knowledge Engineering Rediscovered: Towards Reasoning Patterns for the Semantic Web. In: Proceedings of The Fifth International Conference on Knowledge Capture. ACM, September 2009.
E. Blomqvist and K. Sandkuhl. Patterns in Ontology Engineering: Classification of Ontology Patterns. In: Proceedings of the International Conference on Enterprise Information Systems, Miami Beach, Florida, 2005.
A. Gangemi. Ontology Design Patterns for Semantic Web Content. In: The Semantic Web - ISWC 2005, LNCS Vol. 3729 Springer, 2005.
V. Presutti, E. Daga, A. Gangemi, and E. Blomqvist. eXtreme Design with Content Ontology Design Patterns. In: Proceedings of WOP 2009 Vol. 516 CEUR-WS, 2009.
E. Blomqvist, V. Presutti, E. Daga, and A. Gangemi. Experimenting with eXtreme Design. In: Proceedings of EKAW 2010. LNCS Vol. 6317 Springer, 2010.
A. Nuzzolese, A. Gangemi, V. Presutti, and P. Ciancarini. Encyclopedic Knowledge Patterns from Wikipedia Links. To appear in: Proceedings of the International Semantic Web Conference 2011, Springer LNCS, 2011.
J. Brooke. SUS: a "quick and dirty" usability scale. In: Usability Evaluation in Industry. Taylor and Francis, 1996.
D. Anicic, S. Rudolph, P. Fodor, and N. Stojanovic. Stream Reasoning and Complex Event Processing in ETALIS. To appear in: The Semantic Web Journal, IOS Press, 2011.
E. Blomqvist. OntoCase-Automatic Ontology Enrichment Based on Ontology Design Patterns. In: Proceedings of the International Semantic Web Conference 2009, Springer LNCS, 2009.

Page responsible: Eva Blomqvist
Last updated: 2012-09-27

Department of Computer and Information Science