ArticlePDF Available

Ontology-driven Natural Language Access to Legacy and Web Services in the Insurance Domain Ontology-driven Natural Language access to Legacy and Web services in the Insurance Domain Mikhail Simonov

Authors:
Article

Ontology-driven Natural Language Access to Legacy and Web Services in the Insurance Domain Ontology-driven Natural Language access to Legacy and Web services in the Insurance Domain Mikhail Simonov

Abstract and Figures

Ontologies can play an important role in industrial systems by enabling access to Legacy resources in a transparent and distributed way. Ontologies are developed in order to provide machine-processable semantics of information that is exchanged between different agents, either humans or software. The same mechanism can be used for Legacy service discovery, as well as for automated reasoning about the content of answers obtained from that service. We have created an information system offering a natural language, web-based access to distributed Legacy systems in the insurance domain. The application offers a communication channel that actualizes a new model of business interaction between an end user and a service provider, an insurance company in our case. Our system leverages on the knowledge management framework developed within the Eureka funded research project IKF, which has been adapted to the Insurance Domain by building a dedicated Domain Ontology named IES. The core component, named MetaDiscoverer, implements an ontology-based filtering of user queries in order to discover the intended Legacy service. Our application is able to perform information retrieval, knowledge extraction, automated reasoning, service location and dynamic invocation. The solution is RDF and XML-compliant, according to the recommendations given by the W3C Consortium.
Content may be subject to copyright.
Ontology-driven Natural Language Access to Legacy and Web Services in the Insurance Domain
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
Ontology-driven Natural Language access to Legacy and Web services in the
Insurance Domain
Mikhail Simonov
Nomos Sistema S.p.A.
Simonov@nomos.it
Aldo Gangemi
LOA, ISTC-CNR
A.Gangemi@istc.cnr.it
Massimo Soroldoni
Nomos Sistema SpA
Soroldoni@nomos.it
Abstract
Ontologies can play an important role in industrial
systems by enabling access to Legacy resources in a
transparent and distributed way. Ontologies are
developed in order to provide machine-processable
semantics of information that is exchanged between
different agents, either humans or software. The same
mechanism can be used for Legacy service discovery, as
well as for automated reasoning about the content of
answers obtained from that service. We have created an
information system offering a natural language, web-
based access to distributed Legacy systems in the
insurance domain. The application offers a
communication channel that actualizes a new model of
business interaction between an end user and a service
provider, an insurance company in our case. Our system
leverages on the knowledge management framework
developed within the Eureka funded research project
IKF, which has been adapted to the Insurance Domain
by building a dedicated Domain Ontology named IES.
The core component, named MetaDiscoverer,
implements an ontology-based filtering of user queries
in order to discover the intended Legacy service. Our
application is able to perform information retrieval,
knowledge extraction, automated reasoning, service
location and dynamic invocation. The solution is RDF
and XML-compliant, according to the recommendations
given by the W3C Consortium.
1. Introduction
Current Insurance business still heavily relies on
Legacy applications. Back office management activities
in the insurance domain are performed by huge and
well-developed Legacy systems. Through an intranet, an
insurance agent usually deals with a Legacy system in
order to obtain needed information, or to answer typical
queries asked by insured people that visit the agency or
call her by phone. The agent typically deals with the
back office when more complex questions are raised or
real-time calculations are required. This process is time
consuming and automation through semantic
capabilities would be helpful. Unfortunately, software
with such automated reasoning capabilities is not
available in commercial systems.
On the other hand, users typically encounter a web-
based application during an insurance product
purchasing, or when an "automated" CRM is contacted
for FAQs. Managing such relations is not
straightforward, because it requires access to Legacy
systems, understanding of user queries, composition of
insurance regulations and guidelines with user data or
even profiles, etc.
An industrial CRM application can provide natural
language access if it is able to perform semantic parsing
of the user query, and to answer that query by means of
automated ontology-based reasoning [e.g. Jarrar et al.,
2003 for a review and a proposal concerning Customer
Complaint Management].
Conforming to that, IKF-IES is an enterprise
Business Information System, including a knowledge
management system, and a domain ontology for
insurance (IES-Core) that is focused on life insurance
and Unit-linked policies.
Semantic parsing techniques are used in order to
obtain basic lexical disambiguation capabilities for
Italian and English, although other languages can be
handled, using appropriate tools e.g. available from the
Natural Language Understanding Consortium, known as
NLUC1.
IKF-IES is mainly used for for web-based, self-help
purpose. It also supplements human-centered
communication, since Call Centers can be coupled with
a semantic support coming from the system.
The application offers (Fig. 1):
1 NLUC members are listed on http://www.nluc.com
2
BUSINESS INFORMATION SYSTEMS – BIS 2004
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
Figure 1. The IKF-IES architecture
- A document management and information fusion
system (the IKF framework2 services)
- A natural language-based access to Legacy
insurance systems
- An ontology driven query answering
- An ontology-driven Legacy service discovery
- An ontology-supported Legacy service
invocation that employs an intelligent parameter
filtering.
In the following, we will briefly describe some of the
functionalities and components of IKF-IES.
2. From an insurance business model to
the domain ontology
Setting experts' intuition to the minimum, an
insurance business could be modeled according to the
following simplified schema:
Customer –has– Policy (1..*)
Policy –expects– Receipt (1..*)
Policy –has– Claim (0..*)
Claim –has– Payment (1..*)
Policy –envisages– Payment (1..1)
2 IKF is an international EUREKA co-funded project that
develops frameworks to allow knowledge management in
various business settings; http://www.ikfproject.com
The schema means that a customer has at least one
policy, for which at least one receipt is expected, and at
least zero claims can be expected. A claim expects at
least one payment. Life insurances also involve a final
reimbursement, or the possibility to reclaim the
surrender value, as well as many other details, but here
we limit the example to the above schema.
Following such schema, user’s data are represented
in a physical Legacy database, e.g. DB2 is widely
adopted in the use case.
Our application relies on the Legacy database, but it
also provides other ways to access insurance-related
information, and combines Legacy data with that
information by integrating IES domain ontology with
the DB.
In order to develop the domain ontology, a task force
of domain experts, software engineers, and ontology
specialists has reached a design agreement. The domain
ontology has been built from domain documents and
experts’ advice according to the ONIONS methodology
[Gangemi et al. 1999] provided by one of IKF partners,
and the DOLCE-Lite+ foundational ontology [Masolo et
al., Gangemi&Mika], developed in the context of the
WonderWeb3 project, and maintained at the Laboratory
for Applied Ontology4 of the Italian National Research
3 WonderWeb is an European research project co-funded by
5th Framework Program; http://wonderweb.semanticweb.org
4 http://www.loa-cnr.it
Ontology-driven Natural Language access to Legacy and Web services in the Insurance Domain
3
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
Council. The resulting design requires the specification
of IES ontologies into a highly expressive description
logic.5
An example of the analyses performed by experts in
order to produce IES Core is shown in Table 1.
5 An alternative and possibly complementary approach in
building an insurance ontology can be found in [Kietz et al.].
4
BUSINESS INFORMATION SYSTEMS – BIS 2004
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
Table 1. A sample table used for experts’ drafting of the core ontology
Figure 2. A sketch of the kernel derived from the draft tables
From experts’ view, it results that traditional
insurance ontology is focused on the notions of Insurer,
Policy, Insured, and Claim. A further step is to make a
sketch of the tabular drafting, distinguishing concepts
and relations, and then drafting a preliminary kernel for
IES Core (Fig.2).
Next step in IES Core building has been the logical
formalization of the preliminary kernel and of the
tabular drafts, and their integration within the
foundational ontology. DOLCE-Lite+ contains hundreds
of concepts and relations organized around typical
ontology design patterns. The most general ones
concern the topmost partition of categories (e.g. object,
event, region), and their basic relations (e.g. part,
participation, localization). A more specific design
pattern has been developed in order to represent
procedural and regulatory knowledge. It is called
Descriptions and Situations (D&S [Gangemi&Mika],
Fig. 3) because it is defined around the basic relation of
satisfaction between a description context and its viable
realizations (situations). D&S has been extensively used
for representing IES Core because of its flexibility.
Based on D&S, IES-Core reflects the procedural
approach of business modeling widely adopted by
Insurers.
Examples of description contexts in the insurance
domain are policies (as content) and risks (as analytical
specifications). Description contexts have a structure:
DOLCE-Lite+ Class
Informal description
Informal constraints
Non-physical object
(Description context)
A policy is issued by an insurer for an
insured and covers certain losses given
certain risks
Held_by Insured
Issued_by Insurer
Covers Insured
Non-physical object
(Functional role)
An insurer issues a policy to an insured
which covers losses given certain risks
Issues policy
Pays or disclaims claims
Non-physical object
(Functional role)
An insured is covered by a policy against
certain losses given certain risks
Covered by a policy
Activity
An insured makes a claim against a
policy for a loss
Made by insured
Paid or disclaimed by Insurer
Non-physical object
(Description context)
Risk is the likelihood of a certain loss for
a certain insured
Contemplated by a policy
Non-physical object
(Description context)
A loss exists when an insured suffers
some injury to person or property. The
loss is said to be covered if it was part of
the risk contemplated by the policy
Suffered by insured
Covered (or not) by a policy
Activity
A disclaimer is made by an insurer when
a loss is not covered by a policy, usually
when not a contemplated risk
Made by insurer
Ontology-driven Natural Language access to Legacy and Web services in the Insurance Domain
5
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
for example a policy includes roles such as Insurer or
parameters such as Minimum age. Situations are settings
in the real world that comply with (“satisfy”) a context if
the constituents (e.g. a company or an age value) of the
setting comply with the components of the context.
Figure 3. A UML class diagram depicting the Descriptions and Situations ontology design pattern within the DOLCE-
Lite+ foundational ontology.
3. Tuning the domain ontology with task
and application ontologies
Further development of IES Core has been based on
ontological requirement analysis, which uses
competency questions [Grüninger and Fox] to be asked
to experts about the use case for the application making
use of the ontology. In other words, a domain ontology
is considered adequate to the application if it supports
the tasks in the use case without overspecifying entity
types. Moreover, the existing or expected application
components must be taken into account as well. The
final result is an application ontology that explicits both
domain and tasks according to available software
components.
Since the application is a CRM one, customer-
oriented query types constitute a basic set of
requirements to be fulfilled. For example, in order to
obtain answers to FAQs, the following query
classification is adopted:
1. Queries that do not require the access to the
Legacy system, i.e. that can be answered by the
information retrieval component alone; for
example “which documents are required to start
a new life insurance?”
2. Queries that require access to the Legacy system,
when no calculation is needed; for example
“who is the insured?”
3. Queries that require an access to the Legacy
system associated with a specific calculation; for
example “when can I ask the surrendering of my
life insurance?” or “I wish to know the surrender
value of my life insurance”.
The second type of query can be answered easily if
the mapping between the query and the business
modeling of Legacy database is available (see below).
6
BUSINESS INFORMATION SYSTEMS – BIS 2004
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
The last type of query is the most difficult to manage.
There are two possible approaches: to store the pre-
calculated values in the database, or to invoke
dynamically, “on the fly”, the needed Legacy service. If
the last approach is adopted, the enterprise will be
invited to solve problems like: how to discover the
correct Legacy service, how to prepare the IN area, or
which data from the OUT area must be taken and
shown.
Whatever query type we consider, the IR component
or the wrapper to the Legacy system must use explicit
knowledge about the domain represented in the CRM
database, in the query, or in the Legacy schema.
Our business case involves two distinct applications:
the Legacy application of the Insurance Company and
the CRM application used for support to the call center.
A first (and typical) scenario includes a customer
calling an insurance agent, who calls back the customer
after examining all needed Legacy documents, and after
performing due calculations.
A second scenario includes the background (or batch)
pre-calculation of all needed values, for instance once a
week or every month, and the immediate
communication of results to the customer.
A third scenario enables the new communication
channel of Self Web Help, in order to allow the
customer to formulate a natural language query.
The main challenge consists in the semantic
processing of the query, including its disambiguation. In
this procedure, the exact semantic meaning is filtered,
and the available Legacy service is located and invoked.
Next, the obtained result is semantically enriched
through the ontology, and then delivered to the end user.
A stringent temporal constraint must be respected,
since the web session has a time out and the remote user
is waiting for an answer only for a limited time.
IKF-IES uses a semantic parser in order to process
the natural language query, and to extract noun- and
verb-phrases.
IES-Core and the insurance Domain Ontology are
currently expressed in the Loom-Ontosaurus knowledge
representation system [Swartout et al.] that implements a
very expressive description logic, but they will be ported
to both KIF and OWL (whose versions already exist for
DOLCE-Lite+), in order to benefit from both logical
programming engines, and Semantic Web inference
engines. The assertional part of the knowledge
containing textual occurrences has been encoded in a
specific knowledge repository specified in a dedicated
frame-based knowledge representation system. For
details cf. the web portal of the IKF project
(http://www.ikfproject.com).
IKF-IES ontologies currently include 506 concepts
and 79 relations.
4. The framework
4.1 The Knowledge Management Framework
The Knowledge Management Framework supports
the following functionalities:
the semantic parsing of the query
the invocation of the inference engine
(performing automated reasoning)
the discovery of the Legacy (or Web) service
the interaction with the ontology server
the query tunneling towards the knowledge
repository, the information retrieval component,
and the Legacy system.
The last process is an asynchronous one, since, as
generally recognized, query elaboration times can have
different magnitudes: the Legacy call is synchronous,
the IR answering is a fast process, and the KR access is
a time consuming process.
4.2 NLP tools (Parser and Link Elicitor)
IKF-IES can elicit the knowledge from free text,
semi-structured, and structured documents. NLP is
performed by means of a commercial parser developed
by Expert System, Italy. The commercial solution
adopted in the Italian version of IKF-IES uses an
embedded lexicon called the Sensigrafo6, an enhanced
dictionary originally based on WordNet [Miller et al.].
The lexicon includes a taxonomy of synsets ("synonym
sets", equivalence classes of word senses), and the
following supported relationships among synsets:
noun synset - hypernym (is a)
noun synset - meronym (part-of)
verb synset - subject
verb synset - object
verb synset entailment
adjective synset - class
adjective synset - quality
synset - antonym
synset - troponym
synset disjointness
synset covering
Sensigrafo currently has one of the best coverings
of the Italian language.
6 Sensigrafo is a trademark of Expert System S.p.A., Italy:
http://www.expertsystem.it/. Expert System is a member of
NLUC.
Ontology-driven Natural Language access to Legacy and Web services in the Insurance Domain
7
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
Since the semantic parser is lexicon-driven, but not
ontology-driven7, IKF uses an additional module named
Link Elicitor that performs knowledge elicitation driven
by the IES Core ontology. The Link Elicitor component
also manages uncertain knowledge (by means of fuzzy
rules), and produces RDF(S) triples (see next section).
The second processing stage (actually a post-
processing) stores the extracted triples in a dedicated
Knowledge Repository, represented in an XML/RDF
compliant frame-based language.
An analysis of the syntactical structures is adopted in
order to establish the degree of plausibility of triples
retrieved by means of the ontology. A plausibility=1 is
equivalent to the presence of a qualified semantic
relation.
4.3 Metadiscovery and Legacy calls
IES Core and the domain ontology of insurance
attached to it are not only used for display purposes, but
also for automation, integration and reuse of data across
applications.
For example, the IES domain ontology is used in
order to perform link elicitation, i.e. to discover relations
between concepts or individuals underlying terms within
documents or records in databases. Link elicitation is
performed by deriving a set of relevant RDF(S) triples
out of the ontology. Triples constitute a “lightweight”
semantic network that is then translated into a suitable
Legacy service invocation of the Metadiscovery module.
The Metadiscovery module is then ontology-driven and
performs a brokering service between the conceptual
model of the ontology and the business model adopted
in the Legacy database.
The Metadiscovery module performs reasoning and
unification between the part of the semantic network
coming from the user’s query and the object model of
the Legacy part of the enterprise application.
After Metadiscovery has resolved the Legacy service,
this module physically performs the call and handles
marshalling data vs. Legacy and OUT area elaboration.
Some special heuristics are used in this process which
are protected industrial know how.
Legacy calls are performed after the correct Legacy
service has been discovered, and the COMAREA8
system component has been properly initialized with
correct values. This process is also ontology-driven,
7 Although Wordnet are sometimes considered ontologies, we
distinguish here on the basis that current wordnets are not
formal, axiomatic theories [cf. Gangemi et al. 2002.].
8 Cobol-written Legacy applications uses the raw data
exchange through the shared area known as “common
area” or simply “COMAREA” for which the data
marshalling and unmarshalling are managed by
middleware.
since the correspondence between slots and values
comes from the knowledge frames represented in
RDF(S).
4.4 Processes and workflows
Initially, the Information Repository and the
Knowledge Repository are empty, but the system needs
an insurance-related corpus to be loaded, and some
knowledge elicitation has to be performed for each
introduced document. Since the relationship between
information objects and documents is maintained
during query answering, the knowledge elicitation
process creates the appropriate links, and populates the
Knowledge Repository with instances of classes
describing domain relevant objects.
The query answering adopted by IKF-IES involves
three separate answers: the first one comes from a
Knowledge Retrieval component, the second one from
an Information Retrieval one, and the last, most
interesting one is the answer coming from the Legacy
system.
Since the enterprise system is constrained by some
legal issues, the corpus introduced into the Information
Repository must be controlled by a human. To this
purpose, a Document Advisor role has been introduced.
The IKF-IES corpus consists of:
Codice Civile (Italian Civil Code), Circolari
ISVAP (specific laws and extensions issued by
Italian supervising authority), Condizioni
Generali di polizza (terms of contract),
Condizioni Particolari di polizza (specific
restrictions),Nota Informativa (product
description), Fondi e Gestione Separata (Unit
Linked-related), Performance monitoring
(periodic).
Once the Information Repository is populated by
documents, and the extracted knowledge is loaded into
the Knowledge Repository, the query can be answered.
The process can be summarized as follows:
1. Input query
2. Query expansion and disambiguation
(automatic)
3. Meta discovery (Legacy service or web service)
4. Search: information retrieval, knowledge
retrieval (unification) and Legacy call
5. Report generation
4.5 User interface and NLP
The system is web-based and the user interface
consists of a set of jsp and servlets.
A registered user receives the combination
login/password, and this mechanism allows us to build a
correct Legacy call in order to retrieve only the
instances of managed objects that can be retrieved
8
BUSINESS INFORMATION SYSTEMS – BIS 2004
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
according to the privacy law: only a Customer (Insured)
of an insurance company registered for online access
having an ID assigned can deal with the web based
system.
Hence, User = Jack Martin and Pass = *** means
UniqueID = 123 and the List of managed contracts =
Policy N. = 987. Furthemore, we can retrieve general
information about the above mentioned customer in both
ways: by Legacy Call or by unification between the the
instance of a human having a slot UniqueID = 123,
Name = Jack Martin.
Let us assume that the above mentioned customer is
37 years old, the policy N. 987 in the only managed
contract (otherwise, we will be invited to disambiguate),
and the current Tariff is RXII. We also know that
"Today" is 10th December 2003 and the next renewal or
surrendering will be at the end of the year, i.e. on 31st
December 2003. The remaining duration of the policy
can be calculated and will give us the remaining
duration = 6 years.
The abovementioned customer can now post a
natural language question through the web interface,
e.g.: "Vorrei conoscere il valore di riscatto della mia
polizza vita", i.e. "I would like to calculate the surrender
value of my life insurance policy". We handle noun-,
verb-phrases, and the links between them according to
the ontology.
Next, a query processing component performs the
stages of (1) Semantic parsing, (2) Disambiguation, (3)
Query expansion. The next step is service discovery: (1)
Semantic Discovery, (2) accessing the service
descriptor, (3) Translation vs. Legacy or Web service
call, (4) Service Invocation at a correct location.
The above mentioned query could look like the
following one:
CALC. VAL. RISCATTO Is-a <LEGACY
SERVICE>, <Invoked by INV07> <from CICS03> <on
server \\urano>. It has 6 parameters:
1st Cust.Id, 2nd Age, 3rdPolicyNumber, 4th
TariffId, 5th Date, 6th Duration
The binding between actual and formal parameters is
performed in order to ensure the correspondence
between the service required and the available Legacy
service, and so that the correct code base is called by the
Invoker. Furthermore, the COMAREA managing
component for the Legacy call invocation is run.
Hence the following natural language query:
or the same phrase in English:
To calculate the surrender value of my life insurance policy
VP NP
NP NP
To calculate the surrender value of life insurance policy .
V V
N AP NP
PPP
art
my
P
adj N
is disambiguated and translated into a parse tree in
order to be processed by our Link Elicitor.
NLP is assisted by algorithms that establish the fuzzy
degree of plausibility of the triples according to the
semantic structure of the text fragment.
For example, IES Core allows to infer the following
triples including concepts:
Policy – Surrendering – Surrender_Value
Insurer – Calculate – Surrender_Value
Insured – Own – Policy
At this point, we know that someone (actor) wants to
know (action) the surrender value (patient), but the
above mentioned value will be known only after the
calculation (process) to be performed by an Insurer
(actor).
Now we have to discover the Legacy service that
calculates the surrender value for an instance (my life
insurance policy).
Let us assume that the procedure named
“LF_GetValRisc()“ allowing to calculate the surrender
value is available. Once located, the only issue is to
replace parameters with effective values. In order to
perform this task, we use the information carried in
context, i.e. we perform the mapping between concepts
and managed objects (a problem similar to the mapping
of the OO model to ER model or vice-versa).
Finally, the dynamically generated call has the
following syntax:
CALL LF_GetValRisc( Id = 123, Eta = 37, Pol =
987, Tar = RXII, Dal = 31-12-2003, Anni = 6)
Another added value offered by this system is a
semantically enriched answer containing the explanation
of the procedure, the exact meaning of the concept
"surrendering" and the set of links to relevant documents
coming from the Information Repository, carefully
Ontology-driven Natural Language access to Legacy and Web services in the Insurance Domain
9
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
selected in order to answer the question in the most
proper way.
Legacy answers:
The SURRENDER VALUE calculated at the next
renewal date, i.e. (31-12-2003) of YOUR Life insurance
contract, i.e. (N. 987) will be the sum of (14.000,00)
Euro.
Knowledge Retrieval:
SURRENDER VALUE is “The amount an insurance
company will repay to a policyholder who wishes to
terminate his policy early.” The ontology concept is:
#6666; its parent is: #555; its relatives are: #777, #888,
#999:
Information Retrieval:
Link to "SURRENDER" is present in docs. (Doc.
#222, #333, #444)
Link to "SURRENDER VALUE" is present in docs:
(Doc. #555, #666)
Link to "SURRENDERING" is present in docs.
(Docs. #777, #888).
The most important achievement of this system is the
seamless, everywhere, everytime, realtime access to the
situation stored in the Legacy system, since all
calculations are performed on the fly (on demand).
4.6 Business scenario
Life insurance business can be depicted by an
additional sketch:9
Insurer
Life insurance Insured
Value
issues covers
generates Asks (to know)
cash
pays premium
Calculates
Pays
Units
Manage*
(buy, sell)
Represents (is represented by)
The transition from traditional business in the
insurance domain to eBusiness requires a new specific
communication channel. We use a web-based
communication in order to offer a semantic sensitive
eCRM on a self help basis. Registered users receive a
combinatorial UserId and a password. Users access the
9 We consider the management of gathered funds as an
external activity, in order to avoid the inclusion of the
whole Stock Exchange dictionary.
enterprise web portal, which enables to track the user’s
behavior and to provide the best service available. An
ontology-driven, natural language based Q-A dialog is
supported.
A typical user types the UserId and a password to
enter as an IES-User. Firstly, the user history is retrieved
and an authentication is performed, in order to allow the
knowledge retrieval of related objects from the Legacy
applications.
The list of managed policies is retrieved and
generated. At this point we can answer some typical
queries and FAQs. Mostly important, a Legacy call is
performed which computes user-related datatype values,
without a static batch pre-calculation..
Next, users can formulate a natural language query.
The eCRM system manages the query as a free text, i.e.
passes the string to the semantic parser that tags and
disambiguate relevant textual chunks. The resulting
parse tree is passed to the Link Elicitor in order to elicit
the knowledge.
The data flow is in XML, a frame based knowledge
representation is adopted and the set of data includes:
an Instance of the Classs (Object) “Insured” having
following attributes – values:
<Id 999>
<Family Name XXX>
<Name XXX>
<DateOfBirth DDMMYYYY>
<Policy 1234>, <Policy 1235>, <Policy 1236>
<Query: “Phrase in N.L. to be parsed”>
At this point the query is disambiguated and the
system tries to discover the needed Legacy (or Web)
service to be invoked. In order to facilitate reasoning,
disambiguation is aided by some domain-oriented
heuristics concerning typical constructions, e.g.
“to know” is interpreted as “to calculate”, “how to
do” as “procedure for” or “rules concerning” etc.).
The query is then submitted to the IR component, the
Knowledge Retrieval and the Legacy application. The
first returns textual chunks, the second returns the
ontological data and instances of information objects,
such as regulations. The last one returns the required
calculation for a given instance.
After the Legacy service is determined, the call must
be prepared appropriately. The formal parameters list
will be replaced with real data coming from the context.
Finally the IKF Invoker performs the Legacy call.
For instance, a swap between two life insurance
products involves some expenses, therefore the
following Legacy call will be performed:
GET_VAL_SWAP ( Id = 999, Age = 37, Policy = 1234,
Tarif = RXII, From = 01-01-2004, Years = 6 )
10
BUSINESS INFORMATION SYSTEMS – BIS 2004
Witold Abramowicz (ed.), Business Information Systems, Proceedings of BIS 2004, Pozna_, Poland
Finally, the IKF Demon receives the data from the
Legacy, and generates a report as an HTML/XML page.
This cycle enables a user, to know a value calculated
“on the fly” by the Legacy, to access to the links
returned by the IR component, and to read any
regulation contained in the corpus, by following
knowledge items and their links. In addition, users can
fully understand the operation performed, since the
reference conceptual model defined in the ontology is
shown as well.
5. Conclusions
We have illustrated a concrete enterprise system
developed using innovative techniques and algorithms.
Our aim has been to demonstrate the feasibility of an
ontology-driven system for intelligent service discovery,
Legacy application integration, and dynamic query
answering, as opposed to traditional techniques relying
on heavyweight, batch pre-processing.
The paper has presented some aspects of the IKF-IES
application concerning its architecture, its underlying
core and domain ontology, and its functionalities.
The IKF framework has been only briefly described,
since IES design is partly independent from it. As a
matter of fact, the core functionality of IES is ensured
by its ontological component, which is customizable,
reusable, and is still growing to cover other subdomains
within the insurance domain.
Besides architectural and technical choices, including
system components and specification languages, the
glue of IES is provided by its formal conceptual model,
which enables query reformulation, database integration,
and provision of background domain knowledge within
a common framework. Formal specification of insurance
workflows (by using the D&S design pattern) allows a
formal matching between procedural domain knowledge
and the system components. This peculiar use of
ontologies is exploited in related work carried out within
the IKF project for different domains, for example anti-
money laundering regulations in the banking domain
[Gangemi et al. 2001], and service-level agreements.
Further developments of the insurance ontology and
related applications will include the matching between
legal insurance regulation and policy enactment.
6. Acknowledgments
Some of achievements described in the article are
inherited from Eureka co-funded project named IKF.
We thanks the European Eureka Institution for financial
help provided. The work described in this paper draws
upon the contributions of many people, to whom the
authors are indebted. Of course the authors are the sole
responsible forany possible mistake. The authors would
like to thank the IKF Consortium for its support.
7. References
Gangemi A, Pisanelli DM, Steve G, “An Overview
of the ONIONS Project”, Data and Knowledge
Engineering, 31, 1999.
Gangemi A., Pisanelli DM., Steve G., A Formal
Ontology Framework to Represent Norm Dynamics.
Proc. of Second International Workshop on Legal
Ontologies, Amsterdam, 2001.
Gangemi A., Guarino N. Masolo C. Oltramari A.,
Schneider L. “Sweetening Ontologies with DOLCE”,
Gomez-Perez & Benjamins (eds.), Proceedings of
EKAW02, 2002.
Gangemi A., Mika P. 2003, Understanding the
Semantic Web through Descriptions and Situations,
Meersman R, et al. (eds.), Proceedings of ODBASE03,
Springer, Berlin, 2003.
Gruninger M. and Fox M.S., "The Logic of
Enterprise Modelling", in Modelling and Methodologies
for Enterprise Integration, pages 140--157, 1996.
Jarrar M, Verlinden R, Meersman R, Ontology-based
Customer Complaint Management, Proceedings of
WORM03 Workshop at ODBASE03, Springer, Berlin,
2003.
Kietz, J.-U., Maedche, A. and Volz, R., Extracting a
Domain-Specific Ontology Learning from a Corporate
Intranet, Conference on Natural Language Learning
(CoNLL'2000). Lisbon, Portugal, September 13-14,
2000.
Masolo C., Borgo S., Gangemi A, Guarino N,
Oltramari A, Schneider L, Deliverable D18: “The
WonderWeb Library of Foundational Ontologies”,
http://wonderweb.semanticweb.org, 2003.
Miller G.A., Beckwith R., Fellbaum C., Gross D. &
Miller K., "Introduction to WordNet: An On-Line
Lexical Database", http://www.cosgi.princeton.edu/~wn;
August 1993.
Swartout B., Patil R., Knight K. and Russ T.,
"Toward Distributed Use of Large-Scale Ontologies", in
Proceedings of the Tenth Knowledge Acquisition for
Knowledge-Based Systems Workshop, Banff, Alberta,
Canada, 1996.
... It is used in the fields of semantic Web and database systems [Gruber, 2008], library sciences [Ding and Foo, 2002], routing systems [Winter and Tomko, 2006], Information Systems and software development [Mavetera, 2007; Aβmann et al., 2006; Dristas et al., 2005; Corcho et al., 2006], to mention but a few. Ontology has also been used in accessing legacy resources [Simonov et al., 2004] through an ontology-driven natural language Web-based access system. More interestingly, Soffer et al. [2001] used ontologies to bridge the gap between business requirements and off-the-shelf Information Systems software capabilities in order to adapt the business to software capabilities. ...
... When dealing with legacy systems that problem emerges even more, due to the time elapsed between the requirements specification for each developed module and the present. Usually, each application has its own configuration, store its own data and is guided by its own business rules [1]. One of the major costs in software is its maintenance, as it becomes more complex and expensive thru time. ...
Conference Paper
Software development in highly variable domains constrained by tight regulations and with many business concepts involved results in hard to deliver and maintain applications, due to the complexity of dealing with the large number of concepts provided by the different parties and system involved in the process. One way to tackle these problems is thru combining software product lines and model-driven software development supported by ontologies. Software product lines and model-driven approaches would promote reuse on the software artifacts and, if supported by an ontological layer, those artifacts would be domain-validated. We intend to create a new conceptual framework for software development with domain validated models in highly variable domains. To define such a framework we will propose a model that relates several dimensions and areas of software development thru time and abstraction levels. This model would guarantee to the software house traceability of components, domain validated artifacts, easy to maintain and reusable components, due to the relations and mappings we propose to establish in the conceptual framework, between the software artifacts and the ontology.
Conference Paper
Full-text available
New technologies have improved the ability of electronically storing, transferring and sharing medical data; they also create serious questions about who has access to this cross-media content and how it is protected and distributed. Our aim is to support healthcare professionals in this changing environment, by providing a set of software tools that help reducing the time and associated costs to collect the information and knowledge required, and in making the best use of it for a more informed decision making (diagnoses, therapies, protocols). An appropriate and consistent level of information security consists of protecting information both in written, spoken, electronically recorded, or printed form from accidental or malicious modification, destruction, or disclosure by applying the strictest requirements for data security and privacy and complying with ethical standards. By Doc@Hand project we experiment the distribution of computer-based patient record, natural language queries and the contained knowledge in a secured way.
Article
Full-text available
Large scale knowledge bases systems are difficult and expensive to construct. If we could share knowledge across systems, costs would be reduced. However, because knowledge bases are typically constructed from scratch, each with their own idiosyncratic structure, sharing is difficult. Recent research has focused on the use of ontologies to promote sharing. An ontology is a hierarchically structured set of terms for describing a domain that can be used as a skeletal foundation for a knowledge base. If two knowledge bases are built on a common ontology, knowledge can be more readily shared, since they share a common underlying structure. This paper outlines a set of desiderata for ontologies, and then describes how we have used a large-scale (50,000+ concept) ontology develop a specialized, domain-specific ontology semi- automatically. We then discuss the relation between ontologies and the process of developing a system, arguing that to be useful, an ontology needs to be created as a "living document", whose development is tightly integrated with the system's. We conclude with a discussion of Web-based ontology tools we are developing to support this approach.
Article
Full-text available
WordNet is an on-line lexical reference system whose design is inspired by current psycholinguistic theories of human lexical memory. English nouns, verbs, and adjectives are organized into synonym sets, each representing one underlying lexical concept. Different relations link the synonym sets.
Article
Full-text available
This paper describes our actual and ongoing work in supporting semi-automatic ontology acquisition from a corporate intranet of an insurance company. A comprehensive architecture and a system for semi-automatic ontology acquisition supports processing semi-structured information (e.g. contained in dictionaries) and natural language documents and including existing core ontologies (e.g. GermaNet, WordNet). We present a method for acquir- ing a application-tailored domain ontology from given heterogeneous intranet sources.
An Overview of the ONIONS Project
  • A Gangemi
  • Dm Pisanelli
  • Steve
Gangemi A, Pisanelli DM, Steve G, “An Overview of the ONIONS Project”, Data and Knowledge Engineering, 31, 1999
Toward Distributed Use of Large-Scale Ontologies
  • B Swartout
  • R Patil
  • K Knight
  • T Russ
Swartout B., Patil R., Knight K. and Russ T., "Toward Distributed Use of Large-Scale Ontologies", in Proceedings of the Tenth Knowledge Acquisition for Knowledge-Based Systems Workshop, Banff, Alberta, Canada, 1996.