KONSULTANT

A Knowledge Base for Automated Interpretation of Profit Values

Bojan Tomić and Tanja Milić

Faculty of Organizational Sciences, University of Belgrade, Jove Ilića 154, Belgrade, Serbia

Keywords: Expert system applications, Knowledge base, Rule-based system.

Abstract: Modern reporting systems and business intelligence tools provide various reports for everyday (business)

use. Unfortunately, it seems that these reports contain mostly data and little or no information. The

consequence is that users need to manually analyze and interpret large quantities of data in order to get

information on how the business is doing. A potential solution for this problem is presented in this paper. It

is a knowledge base for automated interpretation of annual profit values for enterprises.

1 INTRODUCTION

Increased usage of reporting systems and business

intelligence tools (BI onward) has lead to improved

reporting throughout various enterprises. However,

some experts in the BI area state that the generated

reports are not that useful as it seems that they

contain mostly data and little or no information.

A potential solution for this problem is presented

in this paper. The idea is to automate the process of

transforming data into information. The information

generated this way can then be presented along with

data, thus making reports more useful.

The result is a knowledge base for automated

interpretation of annual profit values for enterprises -

Konsultant. This simple prototype takes relevant

business data and creates natural-language-like

conclusions about the state of business.

2 RELATED WORK

Recent magazine articles and interviews with

professionals in the BI area reveal that there might

be an abundance of data and lack of information in

reports generated by BI systems. One article

(Whiting, 2002) suggests that users don't have easy

access to the information they need because there is

too much information to sift through. Gnatovich

(2005, §2) stated that “one would think this

avalanche of data would be delivering huge benefits,

but executives report that they are no closer to

making good use of this mountain of information”.

In an interview, Wise said that “BI is so data heavy

that it's often very difficult in terms of being able to

get the information that people want”(All, 2008, §3).

Data can be defined as “the raw material that is

processed and refined to generate information”

(Silver and Silver, 1989, pp. 6). Data becomes

information when meaning is added to it (Floridi,

2005, pp. 353). This can be presented as an equation

(Checkland and Scholes, 1990, pp. 303):

INFORMATION = DATA + MEANING (1)

When the user gets a report (it usually contains a

few spreadsheets together with some charts), he/she

takes a look and interprets the data presented within

by using his/her knowledge. Information about the

current state of business is the result (equation (2)).

INFORMATION = DATA +

INTERPRETATION KNOWLEDGE

(2)

Finally, by consulting equations (1) and (2), it

can be deducted that it is the users' interpretation

knowledge that adds meaning to data thus making

information. This leads to equation (3) which is

described in detail in a paper by Tomić (2009).

INTERPRETATION KNOWLEDGE =

MEANING

(3)

Hence, the user is the one who manually

interprets data, adding meaning to it and turning it

into information. Ideally, this data-to-information

transformation process should to be automated. The

420

Tomi

c B. and Mili

c T. (2010).

KONSULTANT - A Knowledge Base for Automated Interpretation of Proﬁt Values.

In Proceedings of the 12th International Conference on Enterprise Information Systems - Artiﬁcial Intelligence and Decision Support Systems, pages

420-423

DOI: 10.5220/0002968904200423

 SciTePress

first step then is to formalize this interpretation

knowledge and create a knowledge base.

The knowledge base presented in the following

sections is just one major milestone in our project,

while the other includes developing an explanation

facility which could provide natural-language-like

explanations in various output formats and

languages (Tomić, Horvat and Jovanović, 2010).

3 MODEL

The knowledge for the Konsultant knowledge base

was extracted from literature and experts. Literature

included several university textbooks (Milićević and

Ilić, 2005; Žarkić Joksimović, 2005; Jovanović,

2005) and an expert was called on to verify the

model for correctness and completeness.

The process of analyzing annual profit values

was the basis of the knowledge base model. This

process can be divided into four steps.

The first step is to interpret profit value for the

current year. No other data is taken into account, not

even profit values for previous years. The only

information that can be inferred is whether profit is

positive, negative or neutral (close to zero). But,

depending on the size of the enterprise, the industry

sector and other factors, it can be hard to determine

if profit is neutral or positive (a profit of ten

thousand dollars may be considerate for some

enterprises, but insignificant for others). Each

enterprise has its own set of referential values, so it

was decided that the boundary values for neutral

profit should be entered together with other data.

Positive profit is always considered to be a good

thing. But, negative or neutral profit may not be

interpreted as bad in all situations. New enterprises

tend to achieve negative profit in the first few years

of business. This is why, in the second step, it is

important to check whether the enterprise is new and

if the return-on-investment (ROI) period is over. It is

bad if negative or neutral profit persists after the

projected ROI period is over. Established enterprises

should almost always be able to achieve positive

profit, except if they undergo a major investment.

The third step includes comparing current profit

with average profit in the industry sector. Average

profit values are never good indicators by

themselves as individual profit values may be very

scattered. But, negative or neutral average sector

profit can indicate a serious crisis or disruption in

the market and thus enable finding root cause for

negative profit values. If the annual profit is larger

than the average in sector, the enterprise is doing

better than most of its competitors and vice versa.

Since it is hard to determine accurately if the profit

is significantly larger or smaller than the average

profit in sector, the solution we employed was to

enter a percentage value to act as a boundary.

The final step is to compare profit value for the

current year with values from previous years. This

time series analyses is always limited to a few years

in the past (usually three to five) and helps

determine whether profit is on the rise, oscillatory or

declining. The analyses is made by comparing

differential profit for two consecutive years. If, for

example, profit is on the rise, it can be determined if

the growth trend is exponential, asymptotic or nearly

linear. Declining profit can have a parabolic,

asymptotic or nearly linear trend. Oscillatory profit

trends are most common. Their interpretation starts

by determining if profit is on the rise since the

previous year or not. Then, profit values for the

current year and two years before get compared in

order to see if the oscillation has lead to an overall

growth or decline.

The knowledge model derived from this process

contains four parts (which correspond to the process

steps). The acquired knowledge is formalized by

using production rules and objects. Forward

chaining is the inferencing technique of choice as

there is a need to infer as much information as

possible from a limited set of data. A few example

rules can be seen in the following listing.

IF enterprise.currentProfit > 0 AND

enterprise.ProfitNearZero = false

THEN “Enterprise is making money and

profit is significantly large

which is good.”

IF enterprise.currentProfit > 0 AND

enterprise.diffProfit1 > 0 AND

enterprise.diffProfit2 > 0 AND

enterprise.diffProfit1 >

enterprise.diffProfit2

THEN “The profit has had an exponen-

tial growth trend in the past two

years which is excellent.”

4 IMPLEMENTATION

The Konsultant knowledge base is implemented by

using the Drools Expert rule engine (JBoss

Community, 2009). This free tool uses Java objects

as facts and has support for forward chaining. The

implementation consists of 73 rules, two of which

can be seen in the following figure (Figure 1).

KONSULTANT - A Knowledge Base for Automated Interpretation of Profit Values

421

Figure 1: Knowledge base implemented in Drools Expert.

All rules are divided into “agenda groups”. These

groups provide a way of controlling which rules are

active during the inference process and, together

with rule importance (“salience”), can help control

the order in which inferences are made. Each part of

the model is implemented as an agenda group.

Second, all rules use regular Java objects as facts

in their “if” part (in Drools Expert, this is the

“when” part). Business data is represented by using

an instance of the “Enterprise” class. For example,

attributes “currentProfit” and “profitNearZero”

contain values which are used to determine whether

the “if” part of the rule evaluates as true. Private

methods do all of the necessary calculations before

the inference process begins. For example, the

“profitNearZero” attribute is set to true or false in

advance by a method that checks whether current

profit is inside the defined neutral profit boundaries.

The actions in the “then” part of each rule enable

creation of information in the form of natural-

language-like explanations. These explanations are

generated by commands that output text on the

standard Java output stream - computer screen. Also,

these commands do not alter existing facts nor create

new facts to be used for future inferences.

Up to this point there has been no mention about

data sources used for Konsultant. The data source is

a small OLAP implemented in the MySQL relational

database (Oracle Corporation, 2009). The data is

structured in the form of a “star schema”, meaning

that it is not normalized and contains redundant,

precalculated data suitable for reporting. Time is the

only dimension (in years), and data tables contain

annual profit, average sector profit and differential

profit values. Additional data provided by the user

like neutral profit boundaries and near average profit

boundaries are stored in separate data tables.

The connection between the OLAP data source

and the knowledge base is made by using the

Hibernate object-relational mapping framework

(JBoss Corporation, 2008). A single instance of the

“Enterprise” class is created and populated with data

from the OLAP data source. This instance is then

inserted into the working memory, and the

inferencing can begin.

An example of a report generated by Konsultant

can be seen in the following figure (Figure 2). The

report consists of simple, unformatted text with

some data values inserted at certain places.

Information it contains can simply be read and

understood without further interpretation. Since data

is also present, the user can ensure that the

information is correct. Currently, there is no

possibility of including data tables or charts in the

report, or transforming the report to XML or PDF.

Figure 2: A report example generated by Konsultant.

5 EVALUATION

The reports Konsultant provides are very basic, and

we believe that an evaluation with end users would

be premature and strongly influenced by limitations

of the report presentation techniques. Therefore, this

section summarizes experiences that the expert and

knowledge engineers gained during development.

On the positive side, business data interpretation

knowledge is not difficult to acquire because it is

standardized and can be found in university

literature. The knowledge base can easily be

maintained and upgraded thanks to the modular

design. The reports contain both information and

data, thus enabling the user to assert that the

information is correct. Also, the information is given

in the form of natural-language-like sentences which

ICEIS 2010 - 12th International Conference on Enterprise Information Systems

422

can just be read and require no additional technical

skills on behalf of the user. Finally, the inferred

information is not biased and is of constant quality

(uninfluenced by fatigue, lack of knowledge etc.).

The solution has some negative aspects as well.

First, graphical charts and data tables cannot be

inserted into the reports at this time. Also, there is no

standard way of defining fuzzy facts such as “near

average”, “close to zero” etc. Currently, Konsultant

uses “hard” boundaries (usually represented by

constants) to define these facts. Finally, the

knowledge base is not able to work with missing or

incorrect data. The consequence is that the inferred

information can prove to be incomplete or incorrect.

6 FUTURE WORK

Konsultant is a part of a larger project, so future

work is dictated by the overall project plan and

findings presented in the evaluation section.

First of all, issues regarding presentation of fuzzy

facts, working with missing and incorrect data as

well as graphical representation of data need to be

resolved. Second, the explanation facility which is

being developed (Tomić, Horvat and Jovanović,

2010) should be utilized. This facility should be able

to insert graphical representations of data into

reports thus addressing one of the issues. Only then

should an evaluation with end users be performed.

7 CONCLUSIONS

A knowledge base for automated interpretation of

annual profit values is presented in this paper. It

transforms business data into information which is

then presented as natural-language-like sentences.

Positive aspects of this solution include easy

maintenance and upgrading through availability of

standardized knowledge. The provided information

is unbiased and constant in quality. On the negative

side, the generated reports cannot contain any data

tables or charts, there is no standard way of defining

fuzzy facts and the issue of missing and incorrect

data has not yet been properly addressed.

Future work includes resolving all issues and

utilizing the explanation facility that is being

developed. The final step should include an

evaluation with end users.

REFERENCES

All, A. (2008). With Broader BI Usage, Who Owns the

Data? IT Business Edge. Retrieved January 9, 2008,

from: http://www.itbusinessedge.com/item/?ci=37391

Checkland, P. B. and Scholes, J. (1990). Soft Systems

Methodology in Action. NY: John Wiley & Sons.

Floridi, L. (2005). Is Semantic Information Meaningful

Data? Philosophy and Phenomenological Research,

70(2), 351-370.

Gnatovich, R. (2005). Closing the Analysis Gap. Imedia

Connection. Retrieved January 10, 2008, from:

http://www.imediaconnection.com/content/6060.asp

JBoss Community. (2009). Drools Expert. Retrieved

September 1, 2009, from:

http://labs.jboss.com/drools/drools-expert.html

JBoss Corporation. (2008). Hibernate: Relational

Persistence for Java and .NET. Retrieved November

13, 2008, from http://www.hibernate.org/

Jovanović, P. (2005). Upravljanje investicijama (5th ed.).

Belgrade: Faculty of organizational sciences.

Milićević, V. and Ilić, B. (2005). Ekonomika preduzeća:

fokus na savremeno poslovanje (2nd ed.). Belgrade:

Faculty of organizational sciences.

Oracle Corporation. (2009). MySQL. Retrieved December

3, 2009, from: http://www.mysql.com/

Silver, G. A. and Silver, M. L. (1989). Systems Analysis

and Design. MA: Addison Wesley.

Tomić, B. (2009). Automated interpretation of key

performance indicators by using rules. In Giurca, A.,

Gasevic, D. and Taveter, K. (Eds.) Handbook of

Research on Emerging Rule-Based Languages and

Technologies: Open Solutions and Approaches (pp.

625-646). IGI Publishing: Hershey.

Tomić, B., Horvat, B. and Jovanović, N. (2010). JEFF:

Java Explanation Facility Framework. Proceedings

from YU INFO 2010: The 16th Computer Science and

Informatics Conference. Kopaonik.

Whiting, R. (2002). Analysis gap. Information Week

Retrieved January 10, 2008, from: http://

www.informationweek.com/story/IWK20020418S000

Žarkić Joksimović, N. (2005). Upravljanje finansijama:

osnove i principi (2nd ed.). Belgrade: Faculty of

organizational sciences.

KONSULTANT - A Knowledge Base for Automated Interpretation of Profit Values

423