KIDS

A Model for Developing Evolutionary Database Applications

Zhen Hua Liu

, Andreas Behrend

, Eric Chan

, Dieter Gawlick

and Adel Ghoneimy

500 Oracle ParkWay, Redwood Shores, CA 94065, U.S.A

Universität Bonn, Institut für Informatik III, Römerstr. 164, 53117 Bonn, Germany

Keywords: Knowledge Management, Information, Data Management, Event Processing, Workflow.

Abstract: Database applications enable users to handle the ever-increasing amount and complexity of data, knowledge

as well as the dissemination of information to ensure timely response to critical events. However, the very

process of human problem solving, which requires understanding and tracking the evolution of data,

knowledge, and events, is still handled mostly by human and not by databases and their applications. In this

position paper, we propose KIDS as a model that reflects the way human are solving problems. We propose

to use KIDS as a blueprint to extend database technologies to manage data, knowledge, directives and

events in a coherent, self-evolving way. Our proposal is based on our experience of building database cen-

tric applications that require comprehensive interactions among facts, information, events, and knowledge.

1 INTRODUCTION

Current database technologies are supporting many

modern mission critical applications. There are two

major reasons: a declarative interface and a wide

range of important operational characteristics. The

declarative interface supports queries and data

manipulations without the need to understand

implementation details. The operational

characteristics provide the support for OLTP

applications that serve many users concurrently, high

security, and high availability with no loss of data or

service. With parallel query processing, data

partitioning, data mining, business intelligence, and

decision support databases can analyze large amount

of data, recognize patterns, and extract useful

information from raw data upon which intelligent

decisions can be made. Active DB technologies,

such as triggers, rule processing, continuous

execution of registered queries, and streaming query

processing, can be used to recognize and monitor

interesting events and secure timely responses .

With modern databases individuals and

institutions can solve problems by collecting and

storing factual data, using knowledge to analyze

such facts to develop directives for follow up

activities, and finally, performing such directives

and capturing their effects in the form of additional

facts for further analysis. Such analysis either

indicates that the problem has been resolved (goal

achieved) or further analysis is required. This cycle

of inspecting results, specifying directives and acting

upon them will be repeated until the problem is

resolved. Although modern DB based applications

help human at every step of the process, it is human,

not the database system itself, who understand, track

and follow up the process. The human mind can

handle simple problems without support. On the

other hand, complex problems requiring many

repeated iterations of the process with large amount

of evolving data, knowledge directives, and the

collaboration of teams, require an infrastructure that

tracks problem solving activities in a scalable

manner. As a side effect, the effective management

of generated facts, information, and directives can

lead to enriching the existing knowledgebase.

Since problem solving relies heavily on tracking

the state of facts, knowledge, information, and

activities, we believe that database technologies

provide the right foundation since their traditional

focus and strength is on providing scalable state

tracking and management service in a declarative

fashion. In this position paper, we propose

enhancements to database technologies to create a

comprehensive problem solving platform. We

propose the KIDS model for describing the life cycle

of data management tasks. The acronym KIDS

stands for the most important elements of this model

129

Hua Liu Z., Behrend A., Chan E., Gawlick D. and Ghoneimy A..

KIDS - A Model for Developing Evolutionary Database Applications.

DOI: 10.5220/0004002501290134

In Proceedings of the International Conference on Data Technologies and Applications (DATA-2012), pages 129-134

ISBN: 978-989-8565-18-1

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

by means of Knowledge, Information, factual Data,

Directives and Services. The interrelationships

among the elements of KIDS are presented in Figure

Figure 1: High Level KIDS Model.

KIDS distinguishes among three classes of data

(facts, information, and directives) and three classes

of knowledge (classification, assessment, and

enactment). Solving problems entails the capturing

and the reduction of emerging and historical facts

into information by applying classification

knowledge. Then such information is used to assess

the situation and prescribe/describe the directives for

dealing with the situation. Finally, the directives

have to be executed by applying enactment

knowledge. As directives are enacted, newly

emerging facts will again be classified to determine

whether the situation has been resolved or not. By

implementing KIDS as a database centric platform,

we will not only enable applications to function as

an eco-system that can track the evolution of data,

knowledge, directive that facilitate effective problem

solving processes, but also make the entire

application itself query-able with time traversal

service, “what if” analytical service and provenance

tracking using modern DB temporal and provenance

technologies.

Techniques of modeling data and process logic

are not new. EEML (Krogstie, 2008) provides

modeling languages to express data, process,

resource and goal modeling logic. However, our

focus is to integrate data and process tracking deeply

into DB technologies whose strength is to provide

scalable and declarative state tracking service with

time and provenance support. To our best knowledge,

this is the first position paper that makes the

following contributions:

• This paper proposes the KIDS model and its

underlying concepts and shows how to leverage the

strength of DB technologies to provide KIDS as a

database centric service.

• This paper derives a blueprint to organize and

extend DB technologies, based on the challenges

encountered in supporting a database centric KIDS

platform.

The remainder of the paper is structured as

follows: Section 2 describes in detail the major use

case – patient care that motivates KIDS, section 3

describes KIDS concepts in detail, section 4 is a

functional illustration of how KIDS can be

supported using Object SQL, section 5 discusses the

challenges of supporting KIDS in DB, and section 6

discusses proof points with conclusion in section 7.

2 PATIENT CARE USE

Health care providers are confronted with an

increasing amount and complexity of data and

knowledge as well as an increasing amount of rules

and regulations.

The capturing of patient data as EMRs

(Electronic Medical Records) is becoming common

place. The management of EMRs requires storing a

wide variety of data without any loss while

providing instant access at any time as well as

comprehensive security.

The review and the interpretation of medical data

is becoming increasingly time consuming and

controversial. Therefore, modern patient care

applications have to handle such challenge; i.e.,

doctors need a system that transforms EMRs into

compact information, applying the codified medical

knowledge and providing all possible interpretations.

This must be done on demand as well as proactively

in real time to alert doctors and nurses about adverse

and time critical situations. The number of false

positive and negatives has to be kept as low as

possible to avoid the alert fatigue that is observed

with practically any existing alerting systems.

The system should leverage any encoded medical

knowledge to reduce vitals, the blood chemistry, the

radiology, other observation, or any mix thereof into

one or a few classes of patient state and alert doctors

if such classification indicates the presence of one or

more possible adverse situations that need attention.

This transformation has to be well documented,

and should be easily personalizeable to the

preference of each doctor while observing the

institution’s constraints.

Once the doctor is alerted of the situation and

supplied with the information summarizing the

patient condition along with the relevant facts, s/he

assesses the situation and decides on the course of

action, including any changes to existing treatment.

The application should guide doctors using the

standard of care, models, or any other means to

Enactmen

Assessment

Classification

Facts Informatio

Directives

DATA2012-InternationalConferenceonDataTechnologiesandApplications

130

advice about the best course of action; e.g., indicate

which medicine or combination of medicines has

been must successful with patients in a similar

situation and also which tests are advisable to reduce

the level of uncertainty of the diagnosis. Once the

orders are submitted, the system needs to help in the

supervision and documentation of the execution.

In essence, doctors need comprehensive support

in all phases of the treatment including: the

capturing of the facts (EMRs), the extraction of

information from these facts, the assessment of the

relevant information and facts (diagnosis), determine

the course of action (directives - orders), and the

enactment of such directives. The system should

allow doctors to personalize the application

according to their preferences, be able to integrate

new knowledge easily, contribute to the

development of knowledge, and still leave any

crucial decision in the hand of the doctors.

3 KIDS CONCEPTS

KIDS provides three base classes: the Data,

Knowledge, and Actor classes. An in-depth

discussion of the Actor classes will be covered in a

different paper.

3.1 The Three Classes of Data

Data is factored into Facts, Information, and

Directives that are defined as follows:

Facts represent anything that is observed and

captured; examples are the stream of data coming

from monitors in an ICU, a doctor’s observation, and

an x-ray or ultrasound. Facts provide a quantitative

representation of what has been observed.

Information represents an interpretation of the

facts; examples are qualitative values assigned to a

vital (the blood pressure is normal, severe, or critical)

or the assignment of a name to a pattern – a

diagnosis. Information represent are a qualitative

interpretation of what has been observed.

Directives represent a description of activities to

be carried out; examples are prescriptions for

medicine and orders to test certain aspects of the

blood chemistry. Directives move the task on hand

ahead and trigger the acquisition of new facts.

3.2 The Three Classes of Knowledge

The three classes of data are handled by

corresponding three classes of knowledge:

Classification, Assessment, and Enactment. These

three classes of knowledge provide the necessary

ingredients for abstracting most activities of any

medical, scientific, or business institution. The

definitions of knowledge classes are as follows:

Classification knowledge reduces newly arrived

quantitative facts into qualitative information. An

example is the association of a severity to a vital,

such as the blood pressure, is critical based on the

blood pressure measurement and the general

condition of the patient. Another example is the

identification of diseases based on the available

observation. The goal of the classification is to

derive a compact representation of important facts.

Assessment knowledge analyzes the new

information to formulate a hypothesis about the root

cause and finally render a set of directives to

respond to the situation. An example is the selection

of medicine to treat a patient in which the case that

the directives are the prescriptions. This assessment

could be purely manual or computer guided; e.g.,

using models that show the best selection for

patients in a similar situation. The goal of the

assessment is to development the best directives

based on goal.

Enactment knowledge interprets directives and

carries out their intent. Examples would be the

application of medicine, a single or repeated blood

test, the activation or change of health monitors, and

a surgery. The goal of an enactment is to solve a

problem, and/or capture additional facts.

3.3 Evolution with FID Loop

The three types of knowledge support the evolution

of facts, information, and directives; we call this the

FID loop. The FID loop is the ‘heart beat’ to provide

live process control of KIDS concepts. With time

and provenance tracking, the evolution of the data

and the knowledge in a FID loop can be completely

reviewed, queried, analyzed for auditing, for “what

if” analysis, for knowledge enhancements etc.

4 KIDS SERVICE FROM

DATABASE SYSTEMS

Databases shall provide services to allow any KIDS

instance to be stored and accessed declaratively.

Additionally, databases shall keep every version of

each instance in support of temporal and provenance

access. By storing multiple instances of KIDS, many

users shall share and work the same or multiple

KIDS instances collaboratively. Whenever

KIDS-AModelforDevelopingEvolutionaryDatabaseApplications

131

knowledge gets added, changed, or deleted, the

existing data elements will be processed again. This

ensures that the most current knowledge is always

applied to the most current data.

In this section, we show how KIDS concepts can

be accessed declaratively using OR-SQL.

/* This creates an object table serving as a container to hold all

KIDS object instances */

CREATE TABLE MYCLINICS OF KIDS_TYPE;

/*now populating two KIDS instances into the table containers */

/* creation of KIDS instance for patient ‘John Smith’ */

INSERT INTO MYCLINICS VALUE(

KIDS_TYPE(‘John Smith Case’,

FACTS_TYPE(SELECT PATIENT_OBJ

FROM PATIENTS

WHERE PATIENT_NAME = ‘John Smith’),

INFORMATION_TYPE(BLOOD_CHEM_REPORT_VIEW,

CARDIOLOGY_REPORT_VIEW),

DIRECTIVE_TYPE(ORDER_BLOOD_PRESSURE_CONTROL_PROC)

as d1,

KNOWLEDGE_CLASSIFICATION_TYPE(BLOOD_CHEM_ANA_FUN

C, CARDIOLOGY_ANA_FUNC, BLOOD_PRESSURE_ANA_FUNC),

KNOWLEDGE_ASSESSMENT_TYPE(WHEN

BLOOD_PRESSURE_ANA_FUNC() = “HIGH” EXEC d1));

/* creation of KIDS instance for patient ‘Mary Dunn’ */

INSERT INTO MYCLINICS VALUE(

KIDS_TYPE(‘Mary Dunn Case’,

FACTS_TYPE(SELECT PATIENT_DOC

FROM PATIENT_DOCS

WHERE PATIENT_NAME = ‘Mary Dunn),

INFORMATION_TYPE(BREAST_CANCER_PROGRESS_VIEW),

DIRECTIVE_TYPE(ORDER_CHEMO_PROC) as d1,

KNOWLEDGE_CLASSIFICATION_TYPE(BRST_CANCER_ANA_FUN

C),

KNOWLEDGE_ENACTMENT_TYPE(BRST_BIOPSY,SURG_REMOVA

L_PROC p1)

KNOWLEDGE_ASSESSMENT_TYPE(WHEN p1.status = ‘done’ EXEC

d1));

SQL Code 1: KIDS container and object instances.

SQL code 1 illustrates the creation of table to store

KIDS object type instances.

SELECT OBJECT.GET_FID_LOOP(‘PROVENANCE’)

FROM MYCLINICS

WHERE OBJECT.DESCRIPTION = ‘Marry Dunn Case’

SQLCode 2: Provenance Query.

Provenance Retrieval: SQLCode 2 retrieves the

provenance for patient ‘Marry Dunn’ by calling the

method GET_FID_LOOP() in ‘PROVENANCE’

mode. This method shows the entire FID loop

execution sequence of treatments for the respective

patient. This includes data and knowledge

classification rules that are used to derive Mary’s

cancer progress report, e.g. which surgery and

chemo-therapy have been applied. Such provenance

output allows doctors to examine and navigate the

FID loop process so that they can judge the

trustworthiness of the diagnosis and treatment plan

information. Provenance computation relies on the

full version history of the underlying KIDS

components. It supports time traversal aspect of

KIDS.

Knowledge Evolution: Knowledge evolves over

time. Doctor may be interested, in the case of patient

‘John Smith,’ to use the latest medical knowledge

advancement for blood chemistry analysis.

SQLCode 3 shows how this can be accomplished

using a conditional update statement. First, in a

workspace, the update statement is employed to set

the knowledge classification rule to use the latest

version of blood chemistry analysis function for

patient ‘John Smith’. Afterwards, it calls the method

GET_FID_LOOP() using the mode ‘PROJECTED’.

In this way, unchangeable historical data can be

reprocessed using “what if” kind of analysis.

BEGIN WORKSPACE;

UPDATE MYCLINICS

SET OBJECT.KNOWLEDGE_CLASSIFICATION

= BLOOD_CHEM_ANA_FUNC(LATEST)

WHERE OBJECT.DESCRIPTION = ‘John Smith Case’;

SELECT OBJECT.GET_FID_LOOP(‘PROJECTED’)

FROM MYCLINICS

WHERE OBJECT.DESCRIPTION = ‘John Smith Case’;

END WORKSPACE;

SQLCode 3: Knowledge Evolution Analysis

Personalization and Collaboration:

Knowledge application is not always absolute; e.g.,

different doctors may derive different information

from the interpretation of the same data. KIDS shall

allow users to customize knowledge application to

their preferences. That is, KIDS is personal context

aware. Furthermore, KIDS shall promote user

collaboration. With permission, multiple doctors

shall be able to see each other’s contexts and the

derived information and directive due to these

contexts. This facilitates knowledge collaboration

among doctors. SQLCode 4 shows how this can be

accomplished declaratively by calling the method

GET_FID_LOOP() and passing the ‘PROJECTED’

mode with Dr. Ute Gawlick’s knowledge.

DATA2012-InternationalConferenceonDataTechnologiesandApplications

132

BEGIN WORKSPACE;

UPDATE MYCLINICS

SET OBJECT.KNOWLEDGE = (SELECT KNOWLEDGE

FROM KOWLEDGETAB

WHERE

DOCTOR_NAME=’Ute Gawlick’)

WHERE OBJECT.DESCRIPTION = ‘John Smith Case’;

SELECT OBJECT.GET_FID_LOOP(‘PROJECTED’)

FROM MYCLINICS

WHERE OBJECT.DESCRIPTION = ‘John Smith Case’;

END WORKSPACE ;

SQLCode 4: Knowledge Personalization & Collaboration

Query.

5 DATABASE CHALLENGES TO

PROVIDE KIDS SERVICE

This section illustrates the challenges of hosting

KIDS service from DB perspective.

Knowledge as First Class Citizen: Modern

RDBMSs already support concepts from expert and

knowledge-based systems by means of RDF, OWL

and further logical reasoning capabilities (Das,

2009). Additionally, machine learning techniques

have been incorporated into modern RDBMSs in the

form of data mining functions (Agrawal et al, 1994),

(Milenova et al, 2005). However, compared with the

declarative way of querying data, database systems

are still weak in terms of providing the same support

for querying knowledge. In particular, a user cannot

declaratively query application knowledge coded in

view definitions, stored procedures, triggers, or

event- processing handlers. Being able to classify,

query, search, browse and validate knowledge is

necessary to make knowledge a first class citizen of

a RDBMS just like data is. To this end, all forms of

knowledge, whether represented in form of inference

rules, statistical classifications, learning algorithms,

conditional expressions, query qualifications, or

procedural code, ought to be indexable and its

modification should be automatically monitored and

version tracked as if they were plain data.

Active Knowledge Application: Classical

database systems require users to play an active role

applying knowledge to facts. In contrast, active

knowledge application refers to applications with

knowledge actively looking for facts that are needed

to achieve the actor’s goals. This is done by

monitoring fact updates. Additionally, registered

queries and real-time scoring can be used for

automatically deriving new information from

evolving facts. However, answering questions like

what knowledge or facts are missing or needed to be

changed in order to derive certain information and to

execute certain directives to fulfill actor's goal

require abductive reasoning techniques (Denecker et

al, 2002) which are not yet available in database

systems.

Full Version History and Provenance

Awareness: Data in the form of facts, information,

and directives, as well as knowledge in the form of

classification, assessment, and enactment are

intrinsically temporal and thus need to be versioned

tracked. KIDS service requires temporal database

support with snapshot isolation to access consistent

versions of data and knowledge to extract

provenance. Multi-version index structures (Becker

et al, 2005) with declarative temporal expressions

are necessary for efficiently tracking the

development of data and knowledge over time.

Although data provenance is already a research topic

(Karvounarakis et al, 2010), the integration of

provenance with workflow and process management

is still a challenge. Such general form of provenance

enables users to navigate within the FID control loop

and examine the actual instances of data and

knowledge that have been used at each loop step. In

this way, it is feasible to provide time traversal of

KIDS instances so that user can understand how

historically conclusions were reached and decisions

were made. Furthermore, although history is not

alterable, it is feasible to do “what if” analysis by

generating a new branch of history with application

of latest knowledge to historically collected facts.

Quality Control and Collaboration for KIDS:

KIDS instances have to enable groups of people to

help and learn from each other. Users shall be able to

share parts of a FID loop enabling other users to be

engaged. This type of collaboration approach

supported by version tracking and provenance

allows for improving the quality of data and

knowledge (Richardson et al, 2003).

6 PROOF POINTS

The KIDS concepts have been used for guiding the

implementation of a patient care prototype for an

SICU (Surgical Intensive Care Unit) at the

University of Utah (Guerra et al, 2011). The

prototype consists of a single data repository that

combines a highly configurable rule-based system;

(push-based) alerts, data mining models and an

intuitive user interface. Everything is highly

customizable to the preferences of doctors and the

KIDS-AModelforDevelopingEvolutionaryDatabaseApplications

133

specific circumstances of patients. The prototype is

able to predict life-threatening events (e.g., cardiac

arrest).

The ideas of this prototype have the potential to

significantly increase the value of healthcare

information as the database can store and analyze

healthcare records continuously, find critical

situations, explain why they are critical, allow for a

detailed investigation and provide recommendations

- even for diseases or critical situations that a

specific doctor has never encountered or does not

even know about. An important side effect of this

approach is that the number of false positive and

negative alerts has been significantly reduced;

reducing significantly the familiar alert fatigue. This

approach has the potential to save lives and improve

health care not only in ICU settings by also for any

inpatient and outpatient services.

Another subset of KIDS concepts concerning

knowledge elicitation, structuring, maintenance, and

evolution, has been empirically proven for

troubleshooting and resolving software and

hardware issues. A case-based approach to

knowledge elicitation from subject matter expert

was a key ingredient to the success of our

knowledge elicitation method. The agile and

economic maintenance of such models is facilitated

by case-based automated regression testing

framework. Finally, we found out that leveraging

models based on structured knowledge is more

effective when the information to such models is

evaluated automatically, rather than relying on

extracting such information from users.

7 CONCLUSIONS

In this position paper, we have illustrated the KIDS

concepts and FID loop that models human problem

solving process. Since individual pieces of KIDS

concepts, such as fact, information, directives and

knowledge, have already been managed by DB

system, it is time to use KIDS as a blueprint to

extend DB technologies to understand and manage

the whole KIDS FID loop process and to provide

declarative KIDS service so as to assist human

problem solving in a scalable manner. The challenge

is ahead of us: deep integration of knowledge

management and process management into DB

database technologies are essential to the support of

KIDS abstractions. It is our vision that KIDS

enabled database technology shall be able to host

applications that track evolution of data, knowledge,

directives, events, in a scalable, consistent and self-

evolving manner.

REFERENCES

M. Richardson, P. Domingos: Building large knowledge

bases by mass collaboration. K-CAP 2003: 129-137

R. Agrawal, R. Srikant: Fast Algorithms for Mining Asso-

ciation Rules in Large Databases. VLDB 1994: 487-

499

D. Guerra, U. Gawlick, P. Bizarro, D. Gawlick: An Inte-

grated Data Management Approach to Manage Health

Care Data. BTW 2011:596-605

G. Karvounarakis, Z. G. Ives, V.Tannen: Querying data

provenance. SIGMOD Conference 2010: 951-962

E. N. Hanson, C. Carnes, L. Huang, M. Konyala, L. Noro-

nha, S. Parthasarathy, J. B. Park, A. Vernon: Scalable

Trigger Processing. ICDE 1999: 266-275

S. Das, J. Srinivasan: Database Technologies for RDF.

Reasoning Web 2009: 205-221

B. L. Milenova, J. Yarmus, M/ M. Campos: SVM in Ora-

cle Database 10g: Removing the Barriers to Wide-

spread Adoption of Support Vector Machines. VLDB

2005: 1152-1163

Bruno Becker, Stephan Gschwind, Thomas Ohler, Bern-

hard Seeger, Peter Widmayer: An Asymptotically Op-

timal Multiversion B-Tree. VLDB J. 5(4): 264-

275(1996)

M. Denecker, A. C. Kakas: Abduction in Logic Program-

ming. Computational Logic: Logic Programming and

Beyond 2002: 402-436

J. Krogstie: Using EEML for Combined Goal and Process

Oriented Modelling: A Case Study: EMMSAD 2008

DATA2012-InternationalConferenceonDataTechnologiesandApplications

134