Semantic Search and Query Over SBVR-based Business Rules using

SMT based Approach and Information Retrieval Method

Kritika Anand, Sayandeep Mitra and Pavan Kumar Chittimalli

TCS Innovation Labs, Pune, India

Keywords:

Business Rules, First Order Logic, SBVR, SMT Solvers, Information Retrieval.

Abstract:

Presently, business organizations are regulating their activities with the aid of Business Rules (BR’s). A

single rule set of an organization contains large and diverse categories of BR’s , thereby making it difﬁcult

for Business Analysts and end users to analyze and extract relevant BR’s. Rule Search with natural language

terms fail due to their inability to capture logical semantics present in BR’s. In this paper, we present a novel

approach to give correct and complete sets of SBVR (Semantics of Business Vocabulary and Business Rules)

based BR’s based on a speciﬁed query. We integrate conventional Information Retrieval Approach of text

based searches over the rule base and corresponding meta-data with a SMT (Satisﬁability Modulo Theory)

based approach capturing the higher ﬁrst order logic of the rules. The major applications of this approach are

change impact analysis when rules are added, deleted or modiﬁed from a rule set, identifying the candidate

set of rules affected due to change in the rule set and during match and gap analysis where we compare

two sets of BR’s identifying similarity and difference in business functionality between them. We show the

implementation of our tool along with its performance on industry level datasets.

1 INTRODUCTION

There is a major paradigm shift in business-system

design and development by the introduction of BR ap-

proach (Ross, 2003). BR’s are the logical constraints

imposed by enterprise business organizations to regu-

late their business activities. BR’s are imperative for

ﬂexibility and efﬁcacy of business systems. A BR’s

Approach is a methodology to mine, represent, auto-

mate, and change rules from a strategic business per-

spective (Von Halle, 2001).

BR’s are subject to redesign depending upon ex-

ternal market conditions, government policies or with

attempt to maximize revenue. Also, due to the IT

transformation which include business practices like

mergers and acquisitions, upgrades, incorporation of

a new application, etc, the Business Analysts need

to revisit their BR’s. BR’s and the ability to change

them effectively are fundamental to improving busi-

ness adaptability (Ross, 2003). BR’s are rarely in-

dependent, the removal or changing of facts in BR

repository may impact other rules in the system.

There is a need of system that can minimize the

impact of IT transformation and the external factors

on the BR. BR engine is required along with busi-

ness modeling for planned agility of business systems

(McCoy and Sinur, 2006). In order to maintain agility

at a competitive level, there is a need to focus on 2

main problems:

i To study the change impact when a rule is added,

modiﬁed or deleted from a rule repository.

ii Semantic search and query over the large reposi-

tory of BR’s.

Typically, a BR set consists of extensive and wide

variety of rules. For example in a car rental com-

pany, rule knowledge base may consist of rules con-

cerned with Rental Reservation Acceptance, Car Al-

location for Advance Reservations, Walk-in Rentals,

Handover, No Shows, Early Return, etc. As a result,

Business Analysts and end users ﬁnds it difﬁcult to

analyze and extract relevant BR’s. Therefore, we need

a mechanism to query and retrieve relevant rules from

rule repository or knowledge base keeping in mind the

user’s intent.

Most of the operational Information Retrieval (IR)

systems in existence today use Boolean logic dur-

ing search (Frants et al., 1999). In a search query,

Boolean logic helps us to deﬁne the logical relation-

ship between multiple search terms. The operators

used to express the relationship are AND, OR and

NOT. The IR search engines can be keyword based

(Google, Yahoo), semantic based (Hakia) or hybrid of

Anand, K., Mitra, S. and Chittimalli, P.

Semantic Search and Query Over SBVR-based Business Rules using SMT based Approach and Information Retrieval Method.

DOI: 10.5220/0007710700470058

In Proceedings of the 14th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2019), pages 47-58

ISBN: 978-989-758-375-9

two approaches (T

umer et al., 2009). The keyword-

based search is based on the occurrence of words in

a document. Semantic search in information retrieval

engines seek to improve search accuracy by under-

standing searcher’s intent and the contextual meaning

of terms in the query (Malve and Chawan, 2015). The

logical nature of BR’s makes the rules inter-connected

and dependent on one another. Therefore, query and

search on BR’s must take into consideration the logi-

cal relationships among BR’s.

Our current approach focuses on the rules ex-

pressed in SBVR (Team et al., 2006), an OMG stan-

dard that works as a bridge between business and

IT. The main reason for choosing SBVR for the rule

modeling is due to its declarative nature, natural lan-

guage representation and support for the ﬁrst order

logic. The SBVR meta-model is used to represent

business knowledge as (1) Specifying business vo-

cabularies, (2) Specifying BR’s. In our literature re-

view, we could not identify much work done in the

ﬁeld of searching and querying over the BR’s. The

work (Sukys et al., 2012) describes the transformation

framework from questions in SBVR into SPARQL (Har-

ris et al., 2013) queries over the ontologies deﬁned in

Web Ontology Language (OWL2) (Hitzler et al., 2009)

and SWRL (Karpovic and Nemuraite, 2011) analyzes

the subset of Semantics of Business Vocabulary and

BR’s (SBVR) for a comprehensive representation of

ontological knowledge deﬁned using OWL2. The paper

(Hassanpour et al., 2010) captures logical relation-

ship between SWRL and the OWL ontology language to

form a dependency graph. Based on the graph, (Has-

sanpour et al., 2010) cluster nodes within a layer if

they have similar dependencies. The paper (Anand

et al., 2018) detects inconsistencies in BR’s based

on model checking that exploits the First-order logic

(FOL) (Smullyan, 2012) basis of SBVR speciﬁcation.

In this paper, we propose that to analyze the change

impact on BR’s, we need to query the business repos-

itory taking into account all the 3 aspects of BR’s :

semantic, logical and keyword based.

We also propound one of the other major applica-

tion of search and query is the area of Match and Gap

Analysis. The technique (Mitra et al., 2018) compares

a set of BR’s of a particular organization with the rules

of a reference model, to get a measure of similarity

among the business functionality of the two. The tool

also measures the functionality gap that is present be-

tween embedded logic in the reference BR set and the

other rule set.

The rest of the paper is organized as follows. Sec-

tion 2 presents the list of challenges that motivated the

need of the work. Section 3 provides a detailed classi-

ﬁcation of queries. Also, the section discusses in pro-

fundity how the approach of conventional IR method

and SMT-Libv2 is used in Semantic search and query

in SBVR based BR’s. Section 4 and 5 discusses the

application of our approach in Change Impact Analy-

sis and Match and Gap Analysis respectively. Finally,

experimental studies and discussions are provided in

Section 6.

2 MOTIVATION

Enterprises remove, add and modify the rules with-

out analyzing the impact on the existing rules. This

may result in inconsistency and inclusion of unﬁrable

rules. Consider the set of rules {R

, R

: It is necessary that if purchase price of customer

is greater than 2500 and less than 5000 then

customer is Bronze Customer.

: It is necessary that if purchase price of customer

is greater than 5000 and less than 10000 then

customer is Silver Customer.

: It is necessary that if purchase price of customer

is greater than 10000 then customer is Gold

Customer.

Now, a Business Analyst plans to introduce a new

category i.e “Platinum customer” for the customers

whose purchase price is greater than 15000 to already

existing categories (Bronze, Silver and Gold). The ad-

dition of rule R

causes rule R

to get changed to R

: It is necessary that if purchase price of customer

is greater than 15000 then customer is Platinum

Customer.

: It is necessary that if purchase price of customer

is greater than 10000 and less than 15000 then

customer is Gold Customer.

Clearly, there is a need of BR engine that can auto-

mate the change impact for planned agility of business

system.

Let us take queries from Insurance domain:

: ‘Retrieve me all the rules related to vehicle pur-

chase price is 40000.’

: ‘Retrieve me all the rules for which the product

Group code is Motor Home Plus.’

The following rule is relevant and should be retrieved

from the given Insurance rule base:

ENASE 2019 - 14th International Conference on Evaluation of Novel Approaches to Software Engineering

“‘For other than EAE states during

a ‘New Business’, ‘NLOB’ and ‘First

Auto In Umbrella transaction’,

‘Cancel’ the coverage for the vehicle

with reason code ‘Decline based on

Application information without further

investigation’ when one of the below

condition is met:

• Vehicle type is ‘Motor Home’

• Product group code is other than

‘Facility Tier’ and ‘Bureau Rating’

• vehicle purchase price is greater

than 35000 and less than 60000.

As we can see query Q

maps to rule’s third condi-

tion i.e ‘vehicle purchase price is greater than 35000

and less than 60000’ and Q

falls into the category

of the second condition ’Product group code is other

than ‘Facility Tier’ and ‘Bureau Rating’. The rule set

retrieved for a query include the computational anal-

ysis and the domain-based semantics. Therefore, the

keyword-based IR techniques will not present the cor-

rect result.

The General Financial Rules (GFRs) are a compi-

lation of rules and orders of Government of India to be

followed by all while dealing with matters involving

public ﬁnances. It consists of 324 rules from ‘De-

falcation and Losses’, ‘Submission of Records and

Information’, ‘Control of expenditure against bud-

get’, ‘Classiﬁcation of transaction in Government ac-

counts’. If an end user wants to retrieve the rules re-

lated to ‘Local Fund’ or ‘service is agent of private

body’ or ‘cost of project is greater than Rs 100 crore

or above’, the conventional information retrieval tech-

niques will not be able to reclusively retrieve the re-

sults.

To tackle this limitation, we propose that to ana-

lyze the change impact on BR’s, we need to query the

business repository taking into account all the 3 as-

pects of BR’s : semantic, logical and keyword based.

Our paper aims to retrieve relevant rules from a set

of SBVR based BR’s when certain rules belonging that

particular set is targeted for a change.

3 CLASSIFICATION OF QUERIES

We identify 6 classes of queries: taxonomic,

metadata-based, logical, complex query involving

quantitative analysis and domain based semantics,

data-rule compliant and question-answering based

query. A query can fall into any one of the category

or the combination of two or more categories. We

discuss each type of query in detail in this section.

Business Rule

Query

Logical Query

Complex Query involving

Quantitative Analysis &

Domain based Semantics

Data-Rule Compliant

Query

Question-Answering

based query

Metadata based

Query

Taxonomic Query

Figure 1: Classiﬁcation of Queries on BR’s.

Examples accompanying most of these queries for il-

lustration and better understanding have been taken

from real expert system applications.

1. Taxonomic Query :

Such type of queries take into consideration the

hierarchical aspects of concept types involved in

the SBVR vocabulary. The concepts can be indi-

vidual noun concepts (e.g., Germany to Poland

car movement) or general noun concept (e.g., in-

ternational car movement) as seen from the block

diagram in Figure 2. The Figure 2 is a sub-set of

the block diagram used in EURent (KDM Analyt-

ics, 2016), a car rental company data. Figure 3

consists a small sub-set of rules taken from the

EURent dataset.

We keep a dictionary of SBVR terms used in the

SBVR vocabulary. Then for each term, we have a

list that records the rules in which the term oc-

curs in. Such a list is conventionally known as a

posting list and all the postings lists taken together

are referred to as the postings (Christopher et al.,

2008). The dictionary with pointer to posting list

is kept in memory and the posting list is stored in

disk. The idea is very much similar to the index-

ing the vocabulary used in information retieval for

accessing the relevant documents.

The lexicons obtained after tokenization and

stemming have been replaced by SBVR vocabulary

and the SBVR rules are used instead of documents.

Figure 4 represents the posting list for sub-set of

terms used in rules in Figure 3.

First, let us consider the query term only consists

of single SBVR term. The retrieved set of rules

relevant rule, as an output to the query consists of

Semantic Search and Query Over SBVR-based Business Rules using SMT based Approach and Information Retrieval Method

i Rules present in the posting list of query term q

ii Rules present in the posting list of terms present

in upward hierarchy of q (q

) = {q

, q

,....,

} .

iii Rules present in the posting list of terms present

in downward hierarchy of q (q

) = {q

, q

,....,

} .

iv Rules present in the posting list of synonym of

q, q

and q

relevant rule (Q) = posting list(q) ∪ posting list(q

)

∪posting list(q

) ∪ posting list(synonym(q))

∪posting list(synonym(q

))

∪posting list(synonym(q

))

Consider the query when the user is interested

in retrieving the rules associated with ‘one way

car movement’ from set of rules shown in Fig-

ure 3. The users should be presented with all the

rules relating to ‘one way car movement’, the up-

ward hierarchy i.e car movement, the downward

hierarchy i.e local car movement, in-country car

movement and international car-movement and its

specialization Germany to Poland car movement.

The conventional keyword based or entity based

search will not give the desired rules as it does not

consider the hierarchy important to the rule set se-

mantics.

In General, Q can consist of the query terms {q

, q

,....q

}, Q can be the conjunction or disjunc-

tion of q

or its negated form ¬q

Figure 2: Building block showing taxonomic relation of

SBVR concept ‘car movement’ (planned movement of a

rental car of a speciﬁed car group from a sending branch

to a receiving branch).

For Conjunctive Query Q= {q

and q

and ... q

relevant rule (Q) = relevant rule (q

)∧

relevant rule(q

) ∧ ......

..... ∧ relevant rule (q

)

For Disjunctive Query Q= {q

or q

or ... q

relevant rule (Q) = relevant rule (q

)∨

relevant rule(q

) ∨ ......

..... ∨ relevant rule (q

)

For Query Q= {q

and not q

} , the relevant rule

set will consists of rules which are present in post-

ing list of q

or present in posting list of synonym

of q

but are not present in the downward hierar-

chy of posting list of q

( q

). For Q = ‘round trip

car movement and not one way car movement’,

rule R

, R

and R

will be retrieved from a set

of rules in Figure 3. Rule R

is retrieved in the

result set as the Noun Concept rental movement is

synonym of the Noun Concept car movement.

2. Logical Query : Sometimes, the execution of

one or more rules in business system can trigger

other rules. This logical nature makes the rules

interconnected and dependent on one another. It

is very important for business analysts to analyze

and comprehend such interconnected rules. So

we propose the idea of construction of Rule De-

pendency Graph to map the logical relations tak-

ing into consideration the semantic and syntactic

properties of rules. The approach will explore the

interactions between SBVR rules and check how a

rule is affected by execution of other rules.

SBVR is based on ﬁrst-order logic and can also

adopt higher-order logic restricted to Henkin se-

mantics (e.g., for dealing with categorization

types). In general, standard higher-order logic al-

lows quantiﬁcation over uncountably many pos-

sible predicates (or functions)(SBVR, 2008). The

rule R

in Figure 5 encompasses an obligation for-

mulation which ranges over an implication logical

formulation . The implication logical formulation

scopes over 2 logical operands. The ﬁrst logi-

cal operand is conjunction and the second logical

operand ranges over an universal quantiﬁcation.

This universal quantiﬁcation introduces a variable

‘rental’ and ranges over another universal quan-

tiﬁcation. The latter universal quantiﬁcation intro-

duces a second variable that ranges over concept

‘car exchange’ and scopes over an atomic formu-

lation ‘rental incurs car exchange’.

Let’s take a SBVR rule of the form : ‘if f a

and f a

and..... f a

then f c

and f c

and.... f c

’

ENASE 2019 - 14th International Conference on Evaluation of Novel Approaches to Software Engineering

car movement

international car

movement

one way car

movement

rental movement

Figure 4: Posting List for sub-set of terms used in rules in

Figure 3.

where f a

, f a

,.... f a

are the atomic facts as-

sociated with antecedent and f c

, f c

, ... f c

are

atomic facts associated with consequent. We con-

struct the the fact tree of atomic facts involved in

the rule. For example, the forest of Fact tree as-

sociated with rule R

(of Figure 5) corresponding

to atomic facts : f a

: ‘driver has driving li-

cense’ and f c

: ‘age of driver is greater than 21’

is shown in Figure 6.

As we can see in Figure 5, rule R

is clearly de-

pendent on Rule R

as the atomic fact of conse-

quent of R

f c

) i.e ‘car is in need of ser-

vice’ matches with the atomic fact of antecedent

of R

. Therefore, there is an edge from R

to R

in Figure 7. Rule R

does not initially seem to

be dependent on either rule but if the underlying

SBVR vocabulary declares that rental has special-

ization as luxury rental, then a dependency can be

inferred. As the atomic fact of R

i.e R

f c

(age

of driver is greater than 18) subsumes atomic fact

of R

i.e R

f a

(age of driver is greater

than 21), there is a dependency from R

to R

The complete Rule Dependency Graph for set of

SBVR rules(Figure 5) is represented in Figure 7.

We detect the logical dependencies in SBVR rules

based on model checking that exploits the FOL

basis of SBVR speciﬁcation.

Test for checking dependency between rules:

We use the work described in (Anand et al., 2018)

to convert the SBVR rule to SMTLibv2 (Barrett

et al., 2010) mappings. The mappings are con-

structed based on Many-sorted logic and graph

reachability algorithm. Many-sorted logic is

the generalization of FOL in which the domain

of universe is classiﬁed into disjunct sorts (or

types). We present a SMT based method to de-

tect the the presence of edge between two SBVR

rules. Let’s ﬁrst simplify the notation of the nega-

tion of the formula obtained by smt(atomFact

)

→smt(atomFact

¬(smt(atomFact

) → smt(atomFact

))

≡ ¬(¬smt(atomFact

) ∨ smt(atomFact

))

.........[MaterialImplication]

≡ (smt(atomFact

) ∧ ¬smt(atomFact

))

.........[DeMorgan

sLaw]

A formula F subsumes another formula F

(FF

) if for each interpretation I, I |= F

implies

I |= F (Lukichev, 2010). We use this deﬁnition to

detect logical dependency present in SBVR rules.

To detect an edge between two rules, let us con-

sider 2 cases.

• Case I: Rules involving Numeric Range Quan-

tiﬁcation:

SBVR rules and deﬁnitions can consist of as-

pects in which a thing or property associated

with a Noun Concept is measurable in terms

of ‘greater than’ or ‘ less than’ or ‘equal to’

(as shown in Rule R

in Figure 5). Also,

sometimes the Noun Concepts are scoped by

numeric range quantiﬁcation like exactly-n-

quantiﬁcation, atmost-n-quantiﬁcation having

minimum or maximum cardinality as shown in

Rule R

in Figure 3. Figure 5 depicts rules re-

lating to car rental company. The SBVR vocab-

ulary and rules used in Figure 5 is an extension

of EURent data, to provide clear insights of our

: It is necessary that rental is open if estimated rental charge is provisionally charged to credit card of

renter.

: It is necessary that each car movement has exactly 1 receiving branch and sending branch.

: It is necessary that each rental movement speciﬁes exactly 1 car group.

: It is necessary that car transfer has a transfer drop off branch if transfer drop off branch is receiving

branch of one way car movement that is included in car transfer.

: It is necessary that rental is open if rental has international car movement then rented car satisﬁes legal

requirements and emission requirements of visiting country.

: It is permitted that rental is open if each driver of rental is not barred driver.

: Round trip car movement has pick up branch that is same as return branch of car rental.

: It is necessary that each car movement has exactly 1 movement-id.

Figure 3: SBVR rules related to EURent car rental company.

Semantic Search and Query Over SBVR-based Business Rules using SMT based Approach and Information Retrieval Method

Figure 6: Forest of fact trees { f

, f

} corresponding to

the Rule R

Figure 7: Rule Dependency Graph for the set of rules in

Figure 5.

approach.

We construct the FOL and SMTLibv2 for

each atomic fact involved in a rule. The FOL

corresponding to atomic fact R

i.e ﬁrst

consequent of R

: age of driver is greater than 21.

fol(R

): ∀x driver(?x ) ∧ ageInt(?x) > 21

smtlib :

(assert (forall ((x Cluster Driver) (y Int))

( =⇒ (driver domain x)(> (ageInt x) 21))))

The FOL corresponding to atomic fact R

i.e ﬁrst antecedent of R

: age of driver is greater than 18.

fol(R

) : ∀x driver(?x ) ∧ ageInt(?x) > 18

smtlib:

(assert (forall ((x Cluster Driver) (y Int) )

( =⇒ (driver domain x) (> (ageInt x) 18) )))

We then check the satisﬁability of negation of

=⇒ R

) i.e., forml = (smt(R

))

∧ ¬(smt(R

)). An instance of driver that

has age greater than 21 is bound to have

an age greater than 18. SMT solver will fail

to ﬁnd a solution that satisﬁes R

but not sat-

isfy R

. When we assert the forml to be true

and run the SMT solver, solution will be UN-

SAT. The presence of unsatisﬁable core (UN-

SAT) will tell there is an edge from R

to R

• Case II: Rules not involving Numeric Range

Quantiﬁcation:

Rules R

and R

in Figure 5 falls under this cat-

egory. We construct the SMTLibv2 coresspond-

ing to atomic fact R

as shown below:

: rental is open.

fol(R

) : ∀x rental(?x ) ∧ isOpen(?x)

smtlib :

(assert (∀ ((x Cluster Rental) )

( =⇒ (rental domain x)(isOpen x)))).

The FOL corresponding to atomic fact R

i.e ﬁrst consequent of R

: luxury rental is open.

fol(R

) :

∀x luxury rental(?x) ∧ isOpen(?x)

smtlib :

(assert (∀ ((x Cluster Rental) )

( =⇒ (luxury rental domain x)(isOpen x))).

Germany to Poland car movement

Deﬁnition: international car movement whose source country is Germany and destination country is

Poland

Germany to Poland car rental

Deﬁnition: car rental that includes Germany to Poland car movement

: It is necessary that if rental includes international car movement then rented car satisﬁes legal require-

ments and emission requirements visting country.

: It is necessary that if driver has driving license then age of driver is greater than 21 years.

: It is necessary that driver is authorized in visting country if age of driver is greater than 18 years and

issuing date of driver license is before scheduled pick up date of rental by atleast 1 year and rented car

satisﬁes legal requirements and emission requirements visting country.

: It is necessary that if rental is open and car is in need of service then rental incurs car exchange.

: It is necessary that if driver of rental is not barred driver then luxury rental is open.

: It is necessary that if service reading of car is greater than 5500 miles then car is in need of service.

Figure 5: SBVR rules related to car rental company.

ENASE 2019 - 14th International Conference on Evaluation of Novel Approaches to Software Engineering

We then check the satisﬁability of negation of

=⇒ R

) i.e. forml = (smt(R

))

∧ ¬(smt(R

)). As the luxury car rental is a

specialization of rental, it is not possible that all

rentals are open but the luxury rental is not open.

SMT solver will fail to ﬁnd a solution that satisﬁes

but not satisfy R

. The presence of un-

satisﬁable core (UNSAT) will tell there is an edge

from R

to R

We will check the dependency between the rules

by performing the test between with SMTLibv2

corresponding to each fact tree (R

) of an-

tecedent of rule R

with SMTLibv2 associated with

each fact tree (R

) of consequent of rule R

. If

the solver returns UNSAT, then there is an edge

present between rules R

and R

Figure 7 depicts the complete Rule Dependency

graph for the set of rules shown in Figure 5. As

seen from the graph, deﬁnition ‘D

’ is trigger-

ing the rule R

as ‘Germany to Poland rental’ is

specialization of rental as inferred from D

and

‘Germany to Poland car movement’ is an instance

of ‘international car movement’ as deduced from

. As a result, there is an edge present from D

to D

and in turn from D

t o R

3. Metadata based Query

BR’s are often associated with the contextual in-

formation known as meta-data. The information

is usually structured that describes the data like

date-timestamp when the rule was created or mod-

iﬁed, the expiry date of rules, states or products on

which the rule is applicable or the user ID of the

person who created the rules. Queries on meta-

data are normally database or SQL (Date and Dar-

wen, 1989) queries like ‘List the rules that are

applicable for state ‘AX’ or ‘List the rules for-

mulated in last 1 year’.

4. Complex Query involving Quantitative Analysis

and Domain based Semantics

The rule set retrieved for a query may include the

computational analysis and the domain-based se-

mantics. Let us take query from Insurance do-

main which we consider in Section 2.

: ‘Retrieve me all the rules related to vehicle pur-

chase price is 40000.’

: ‘Retrieve me all the rules for which the product

Group code is Motor Home Plus.’

We devise SMT based approach to retrieve the

rules for such complex queries. We formu-

late SMT formula for the the query q i.e smt(q)

and for the atomic facts associated with an-

tecedents and consequents : smt(R) = smt( f a

smt( f a

),......smt(fc

), smt( f c

).... Then,

∀ forml ∈ smt(R)

assert conjunction of smt(q) and negation of

smt(forml)

i.e (assert ( smt(q) ∧ ¬ smt(forml)))

If the SMT solver returns UNSAT then R is a part

of result set.

5. Data-Rule Compliant Query

It is very important for data to abide by both

industry regulations and government legisla-

tions. Legislation laws and business policies are

changed regularly by the government, as a re-

sult of which businesses continually have to re-

spond to changes in the legal framework e.g. the

“Sarbanes-Oxley Act of 2002” is a United States

federal law that was passed in response to a num-

ber of corporate accounting scandals that occurred

between 2000-02 (Sarbanes Oxley Act, 2002).

Business Analysts are also interested to check the

validity of the records or data present in database

with business rules. The data compliance en-

sures that the data is in accord with established

guidelines or speciﬁcations. Such type of queries

are useful to ensure the data is consistent with

demand of dynamically changing regulatory

environment and the operational policies. These

queries checks the user’s given data with the rules

and gives the answer in Boolean format. For

instance, for a query q : “Check the validity of

‘customer has age that is equal to 65 years and

that customer holds a SBI credit card”’. The

query will return false as it contradicts with the

rule r: ‘To avail SBI credit card, customer must

lie within age bracket of 21 to 60 years’. We

convert the query q to following SMT formula :

fol : ∃x, customer(?x) ∧

holds(?x, SBI Credit Card) =⇒ ageInt(?x) = 65

smt(q) :

(assert(exists(

(x Cluster Customer)(y Cluster Card))

(and (customer domain x)

(implies

(and (customer domain x) (holds x y)

(= y SBI Credit Card) )

(= (ageInt x) 65))))

The SMTLibv2 formula has been formulated based

on Many Sorted Logic and Graph Reachability

approach as described in (Anand et al., 2018).

smt(r) : (assert(forall(

(x Cluster Customer)(y Cluster Card))

(and (customer domain x)

(implies

(and (customer domain x) (holds x y))

(= y SBI Credit Card) )

Semantic Search and Query Over SBVR-based Business Rules using SMT based Approach and Information Retrieval Method

(or (> (ageInt x) 21)

(< (ageInt x) 60))))))

The SMT solver will produce an unsatisﬁable core

as the SMT-Libv2 formula corresponding to the

query i.e. smt(q) contradicts with SMT-Libv2 for-

mula corresponding to the rule i.e. smtlib(r).

6. Question-Answering based Query

The Question-Answering queries demands an an-

swer to be given with a short string snippets ex-

pressing named entities, temporal expression, lo-

cation or a numerical value (Aunimo et al., 2007).

The queries can be

q1 : ‘What is the rental duration of rental ?’

q2 : “What is the maximum limit of service read-

ing of a rental car ?’

The result of the query can retrieved from a single

rule or have to be extracted from the combination

of two or more rules.

(

Sukys et al., 2017) deﬁned 9 model transforma-

tion rules to transform SBVR questions to SPARQL

queries. The work (

Sukys et al., 2017) handles

different SBVR questions that are transformed in

the solution: i. questions to ﬁnd individuals of

certain type (e.g., Find persons) ii. simple ques-

tions with roles bound to variables or individuals

(e.g., What states that border Illinois?) iii. count-

ing questions (e.g.,How many states border Illi-

nois?) iv. questions with cardinality restriction

(e.g., Find states that border at least 3 states.) v.

questions with numerical comparison (e.g., Find

cities that have population greater than 100000.)

vi. questions to ﬁnd minimum or maximum val-

ues (e.g., Find state that has largest population.).

4 APPLICATION OF SEARCH

AND QUERY OVER

SBVR-BASED RULES

4.1 Change Impact Analysis

Change Impact Analysis provides accurate under-

standing of the implications of a proposed change on

a rule set. Sometimes, the rules are inter-dependent or

interleaved with each other. Modiﬁcation in one rule

can impact other rules in business system. Therefore,

it is necessary to identify the rules and vocabulary that

might have impacted if business analysts performs a

desired change. The technique of semantic and logi-

cal search and query can be adapted to use in Change

Impact Analysis. It is imperative to detect candidate

set of rules which can be triggered or affected when a

business rule is targeted for a change.

The following instances highlight the importance of

querying in Change Impact Analysis:

1. Instance I: Let’s suppose business owner of car

rental company wants to remove the term ‘in-

ternational car movement’ from vocabulary as

shown in Figure 2. From Figure 2, it can be de-

picted all the rules related to ‘international car

movement’ and its specialization (like ‘Germany

to Poland International car movement’) should be

removed from the system. To achieve this, we

ﬁre a taxonomic query i.e. Q = ‘international car

movement and not one way car movement’. The

query Q will retrieve all the rules (relevant rule

(Q)) related to international car movement and

its specialization. The relevant rule (Q) will not

cover the rules related to one way car move-

ment, local car movement, in-country car move-

ment, round-trip car movement or car movement.

Therefore, to remove the term ‘international-car

movement’, all the rules present in (relevant rule

(Q) should be removed from the system.

2. Instance II: Consider set of rules R

, R

and R

introduced in section 2, an introduction of rule R

causes the rule R

to changed to R

. To ﬁnd set

of impacted rules by the addition of rule R

, let’

s ﬁrst ﬁnd the atomic facts involved in the rules.

By parsing the SBVR XMI, we obtain atomic fact

: ‘purchase price of customer is greater than

15000’ and R

: ‘customer is Platinum cus-

tomer’. The rule R

to be changed can be retrieved

if we ﬁre a fact based query on atomic fact R

The atomic fact of R

i.e R

‘purchase price of

customer is greater than 10000’ subsumes atomic

fact R

3. Instance III: Consider a scenario when we want

to add a rule R

to a set of rules in Figure 5. Sup-

pose R

:‘if age of driver is greater than 16 then

driver has driving license’. As we can see from

the set of SBVR rules, the newly added rule contra-

dicts with the rule R

. This type of anomaly can

be achieved with the help of logical query. We as-

sert and append the rule R

to the existing SBVR

vocabulary and rules involved in Figure 5. Then,

we construct a the Rule Dependency Graph. The

graph will help to analyze the set of dependen-

cies present between the rules. There will be an

edge present from R

to R

as the atomic fact ‘age

of driver is greater than 21’ is subsumed by the

atomic fact ‘age of driver is greater than 16’. By

analyzing and comprehending the incoming and

outgoing edges of R

, we can know the candidate

set of rules that can be affected by the addition of

ENASE 2019 - 14th International Conference on Evaluation of Novel Approaches to Software Engineering

4.2 Match & Gap Analysis

In modern business model, which undergoes con-

tinuous change comparing two sets of rules to ﬁnd

matches and gaps among them is a very important

operation. The work done by (Mitra et al., 2018)

presents an intelligent and automated way to perform

Match & Gap Analysis on BR’s mentioned in SBVR.

We believe that application of Search and Query tech-

niques will aid their approach to perform more efﬁ-

ciently.

The work done in (Mitra et al., 2018) initially uses

NLP techniques along with domain knowledge to ini-

tially match the SBVR vocabulary of both rule sets,

followed by selecting a set of possible matches be-

tween the rule sets. The set of possible matches are

then matched using logical equivalence by converting

the rules to SMT-LIBv2 which gives us the matches

and gaps (if present). Presently the approach tries to

generate the set of possible matches by considering

linguistic similarity of facts only, which causes some

possible matches involving quantiﬁcation range and

stricter conditions to miss out. To ensure better pre-

cision, we propose the use of search and query to en-

hance the candidate set for logical equivalence.

For every rule r consisting of facts F =

{ f

, f

, · ·· f

} in rule set R

, we run a Complex Query

involving facts F on the other rule set R

. The query

generates the set of rules R

(r) ⊆ R

, which contain

rules having similar facts to F . We ﬁnd that replacing

the existing method of using fact similarity with this

approach generates richer candidate sets.

(1): It is necessary that if purchase price of cus-

tomer is greater than 2500$ then customer

is Bronze Customer and customer is eligible

for credit card.

(2): It is necessary that if purchase price of cus-

tomer is greater than 3500$ then customer

is Silver Customer and customer is eligible

for credit card.

(1): It is necessary that if purchase price of cus-

tomer is equal to 4000$ then customer is el-

igible for credit card.

Figure 8: Example Rule Sets R

and R

Figure 8 shows two rule sets R

& R

. As per the

present approach mentioned in (Mitra et al., 2018),

neither of R

(1) or R

(2) will be shown as possi-

ble matches to R

(1), even though they being possi-

ble matches. On the other hand, running a complex

based query which is a disjunction of two facts in-

volved in rule R

(1), i.e., f

: purchase price of cus-

tomer is equal to 4000$ or f

: customer is eligible for

credit card .

Another application of search and query on Match

and Gap Analysis is to identify the extent of gaps

present. Most of the times, two rules from different

rule sets will not be a perfect match. At present, the

approach described in (Mitra et al., 2018) does not

highlight the exact nature of gap present in approx-

imate matches. Figure 9 shows two rules from two

different rule sets which will show as a match as per

(Mitra et al., 2018), but the match is partial with Mer-

cedes and car are similar conceptually but not linguis-

tically.

Vocab

: car

driver

rented car

(1): It is necessary that if car has driver then

car is rented car

Vocab

: four wheeler

Mercedes

General Concept: four wheeler

driver

rented car

(1): It is necessary that if Mercedes has driver

then Mercedes is rented car.

Figure 9: Example Rule Sets RS

and RS

Using search and query we can ﬁnd out the term

hierarchy of Mercedes, and identify the term which

has the closest linguistic similarity to car, which in

this case is four wheeler. This allows us to pro-

pose to the business analyst that a rule containing

four wheeler instead of Mercedes will be a better

match.

5 EXPERIMENTAL STUDIES

AND DISCUSSIONS

To evaluate our approach, we built ‘Logic and Se-

mantic Rule Searcher’ on top of SBVR rule editor in

our BURRITO tool. The SBVR rule editor facilitates an

easy way to business analysts for specifying the SBVR

vocabularies and rules. The tool is allowed to gener-

ate the SBVR XMI based on the SBVR 1.2 meta-model,

that can be provided as an input to the Rule Searcher.

The Rule Searcher presents the user with the follow-

ing options :

• Term based Search: This option clusters the rules

on the basis of term speciﬁed by user. The in-

put to term based search can be the conjunc-

tion/disjunction of terms or their negation. For

Semantic Search and Query Over SBVR-based Business Rules using SMT based Approach and Information Retrieval Method

Figure 10: Snapshot of business rule editor of BuRRiTo tool.

instance, in a car rental company the user can be

interested the rules with ‘international car move-

ment and not one way car movement’

• Fact based Search: In this option, we retrieve the

rules based on the fact that the user has given as

input. The results are retrieved taking into account

the keyword, semantic and logical aspects of rules

and query.

Figure 11: Snapshot of Rule Dependency Graph created

from the BR’s depicted in Figure 10.

• Analyze Rule Knowledge Base: It constructs a

rule dependency graph that assists the user to

explore the interactions between SBVR rules and

check how a rule is affected by or affects the exe-

cution of other rules.

• Analyze the Change Impact: It detects the Candi-

date Rules impacted by addition/ deletion/ modi-

ﬁcation of rule/term.

Figure 10 depicts the screenshot of SBVR editor in

our BuRRiTo tool. The editor checks for the duplicate

facts and terms and also assists user to write the rules

correctly based on the deﬁned SBVR vocabulary. The

Rule Dependency Graph constructed from the rules is

shown in Figure 11.

To evaluate the usefulness and efﬁcacy of our

techniques, we used the tool to perform a case-study

using two real-world applications

1. EU Rent Car Rental (KDM Analytics, 2016):

It consists of 64 rules from a car rental company

concerned with Rental Reservation Acceptance,

Car Allocation for Advance Reservations, Walk-

in Rentals, Handover, No Shows, Early Return,

etc.

2. Rules from Industrial Insurance Application:

We obtained a set of 110 rules from the Industrial

case-study belonging to insurance domain. The

ENASE 2019 - 14th International Conference on Evaluation of Novel Approaches to Software Engineering

Table 1: Comparison of work on EURent and Insurance

Data.

EURent Insurance

Tool Manual Tool Manual

Term-based Search 10 8 5 4

Fact-based Search 4 2 3 2

Addition of rule/ term 1 0 2 1

Deletion of rule/ term 2 1 4 2

Modiﬁcation of rule 2 2 2 1

rules were complex containing the data related to

liability and package policy.

As per our knowledge, there is no study that has been

conducted on searching in SBVR based rules. We pro-

vided a researcher working in the ﬁeld of Natural Lan-

guage Processing with a query belonging to the EU-

Rental domain and another query to an experienced

business analyst from the Insurance Domain. In both

the cases we asked them to manually retrieve the re-

sults from the rule set based on the queries. When

compared with the results from our tool, our tool re-

trieved a richer set for both the queries than man-

ual searches in considerable lesser time. The most

promising observation was that the tool gave no false

positives in the results. We represent the results in

Table 1 for reference.

6 CONCLUSION AND FUTURE

WORK

We have presented a novel approach to give correct

sets of SBVR business rules for a user’s speciﬁed

query. We have integrated the conventional informa-

tional retrieval approach to perform text-based query

over knowledge base and meta-data and SMT-based-

approach to capture the higher ﬁrst order logic of

SBVR. The paper aims to retrieve the set of SBVR

rules for a user’s speciﬁed query taking into consider-

ation the logical, keyword and semantic. We build a

rule graph to analyze and visualize logical dependen-

cies present in SBVR rules. The method leverages the

transformation frameworks from SBVR to SMTLibv2

to incorporate the

• quantiﬁcations (universal, existential, at-most-n,

at-least-n and exactly-n),

• logical operators (logical negation, conjunction,

disjunction and implication)

• synonyms, synonymous forms and specialization

or taxonomic relations relations in SBVR.

The paper also discusses how the interaction of

different types of query can be adapted to be used to

give the potential candidate rules rules when a busi-

ness rule is targeted for a change (addition, deletion

or modiﬁcation). An intuitively appealing approach

therefore seems to be to enhance the ﬂexibility and

resilience of systems to cope with impact of changes

in the business rules. We also depict the application

of searching and querying in Match and Gap Analy-

sis to compare a set of Business Rules of a particular

organization with the rules of a reference model.

The generic framework sketched in the above ap-

proach supports decision making in the organization.

In terms of analysis, an important task is to better de-

tect subsuming rules, rules involving circularity, un-

ﬁrable rules and the duplicate or redundant rules. The

information can then help in designing the anticipa-

tory strategies that will soon be needed to detect such

kinds of rules. The future research in the ﬁeld of

searching and querying should also take into consid-

eration the approach for Question-Answering queries,

which have not been considered in the current paper.

We want to provide stronger experimental results to

showcase the efﬁciency of our tool. As mentioned

in (Mitra and Chittimalli, 2017), there is a lack of

strong datasets of SBVR Vocabulary and Rules. We

are presently working on generating an empirical sur-

vey covering different spectrum of SBVR analysis

which shall also contain precision and recall for this

tool. Therefore, there is a need of a standardized and

universally accepted case study which captures all the

complexities in business rules and can serve as bench-

mark data for all the future works.

REFERENCES

Anand, K., Chittimalli, P. K., and Naik, R. (2018). An auto-

mated detection of inconsistencies in sbvr-based busi-

ness rules using many-sorted logic. In International

Symposium on Practical Aspects of Declarative Lan-

guages, pages 80–96. Springer.

Aunimo, L. et al. (2007). Methods for answer extraction in

textual question answering.

Barrett, C., Stump, A., Tinelli, C., et al. (2010). The smt-lib

standard: Version 2.0. In Proceedings of the 8th Inter-

national Workshop on Satisﬁability Modulo Theories

(Edinburgh, England), volume 13, page 14.

Christopher, D. M., Prabhakar, R., and Hinrich, S. (2008).

Introduction to information retrieval. An Introduction

To Information Retrieval, 151(177):5.

Date, C. J. and Darwen, H. (1989). A guide to the SQL

Standard: a user’s guide to the standard relational

language SQL. Addison-Wesley.

Frants, V. I., Shapiro, J., Taksa, I., and Voiskunskii, V. G.

(1999). Boolean search: Current state and perspec-

tives. Journal of the American Society for Information

Science, 50(1):86–95.

Semantic Search and Query Over SBVR-based Business Rules using SMT based Approach and Information Retrieval Method

Harris, S., Seaborne, A., and Prud’hommeaux, E. (2013).

Sparql 1.1 query language. W3C recommendation,

21(10).

Hassanpour, S., O’Connor, M. J., and Das, A. K. (2010).

Visualizing logical dependencies in swrl rule bases.

In International Workshop on Rules and Rule Markup

Languages for the Semantic Web, pages 259–272.

Springer.

Hitzler, P., Kr

otzsch, M., Parsia, B., Patel-Schneider, P. F.,

and Rudolph, S. (2009). Owl 2 web ontology language

primer. W3C recommendation, 27(1):123.

Karpovic, J. and Nemuraite, L. (2011). Transforming sbvr

business semantics into web ontology language owl2:

main concepts. Information Technologies, pages 27–

29.

KDM Analytics (2016). EU-Rent Car Rental case study.

http://www.kdmanalytics.com/sbvr/EU-Rent.html.

Lukichev, S. (2010). Improving the quality of rule-

based applications using the declarative veriﬁcation

approach. International Journal of Knowledge Engi-

neering and Data Mining, 1(3):254–272.

Malve, A. and Chawan, P. (2015). A comparative study of

keyword and semantic based search engine.

McCoy, D. W. and Sinur, J. (2006). Achieving agility: The

agile power of business rules. Gartner, Special report

on Driving Enterprise Agility, 20.

Mitra, S. and Chittimalli, P. K. (2017). A systematic re-

view of methods for consistency checking in sbvr-

based business rules. In Joint Proceedings of the 3rd

Modelling Symposium (ModSym), Developmental As-

pects of Intelligent Adaptive Systems (DIAS), and Ed-

ucational Data Mining Practices in Indian Academia

(EDUDM) co-located with 10th Innovations in Soft-

ware Engineering (ISEC 2017), Jaipur, India, Febru-

ary 5, 2017.

Mitra, S., Prakash, C., Chakraborty, S., and Chittimalli,

P. K. (2018). Matgap: A systematic approach to per-

form match and gap analysis among sbvr-based do-

main speciﬁc business rules. In Asia Paciﬁc Software

Engineering Conference (APSEC).

Ross, R. G. (2003). Principles of the business rule ap-

proach. Addison-Wesley Professional.

Sarbanes Oxley Act (2002). Sarbanes Oxley Act.

https://www.thebalance.com/sarbanes-oxley-act-and-

the-enron-scandal-393497.

SBVR, O. (2008). Semantics of business vocabulary and

business rules (sbvr), version 1.0.

Smullyan, R. R. (2012). First-order logic, volume 43.

Springer Science & Business Media.

Sukys, A., Nemurait

e, L., and Butkien

e, R. (2017). Sbvr

based natural language interface to ontologies. Infor-

mation Technology And Control, 46(1):118–137.

Sukys, A., Nemuraite, L., Paradauskas, B., and Sinkevi-

cius, E. (2012). Transformation framework for sbvr

based semantic queries in business information sys-

tems. Proc. BUSTECH, 2012:19–24.

Team, S. et al. (2006). Semantics of business vocabulary

and rules (sbvr). Technical report, Technical Report

dtc/06–03–02, Object Management Group, Needham,

Massachusetts.

umer, D., Shah, M. A., and Bitirim, Y. (2009). An em-

pirical evaluation on semantic search performance of

keyword-based and semantic search engines: Google,

yahoo, msn and hakia. In Internet Monitoring and

Protection, 2009. ICIMP’09. Fourth International

Conference on, pages 51–55. IEEE.

Von Halle, B. (2001). Business rules applied: building bet-

ter systems using the business rules approach. Wiley

Publishing.

ENASE 2019 - 14th International Conference on Evaluation of Novel Approaches to Software Engineering