An Improved AC-BM Algorithm for Monitoring Watch List

TELKOMNIKA Indonesian Journal of Electrical Engineering
Vol.12, No.1, January 2014, pp. 379 ~ 387
DOI: http://dx.doi.org/10.11591/telkomnika.v12i1.4141  379
Received June 25, 2013; Revised August 26, 2013; Accepted September 19, 2013
An Improved AC-BM Algorithm for Monitoring
Watch List
Tang Jun, Mo Yiwen*
Zhongnan University of Economics and Law
Wuhan, PR China (430074)
e-mail: evenmomo59@gmail.com
Abstract
With the expanding of database of the watch list of anti-money laundering, improving the speed in
matching between the watch list and the database of account holders and clients’ transaction is especially
important. This paper proposes an improved AC-BM Algorithm, a matching algorithm of subsection, to
improve the speed of matching. Experiment results show the time performance of the improved algorithm
is better than traditional BM algorithm, AC algorithm and the AC-BM algorithm. It can improve the
efficiency of on-line monitoring of anti-money laundering.
Keywords: watch list, matching algorithm, anti-money laundering
Copyright © 2014 Institute of Advanced Engineering and Science. All rights reserved.
1. Introduction
Some international organizations, including the UN, and some related departments of
anti-money continuously release the organizational and personal watch list of suspicious money
laundering and terrorist financing, which require close monitoring by countries’ financial
institutions. Because the data of monitoring system updates constantly and the size of data
expands day by day, the financial institutions have to spend much time in matching the
database of account holders and clients’ transaction, which affects the normal transaction and
the on-line monitoring of anti-money laundering.
The earliest algorithm of matching single pattern is the algorithm of Brute-Force (the
algorithm of BF), which is algorithm of matching in order and the efficiency is low. In 1977,
Knuth D.E, Morris J.H and Pratt V.R proposed an algorithm of matching single pattern, the
algorithm of KMP, which eliminate the problem of the comparison of backtracking. In the same
year, Boyer and Moore proposed the algorithm of BM [1], which can skip by the rule of bad
character and good postfix. The algorithm of BM performs efficiently in the algorithm of
matching single pattern. In 1975, Aho and Corasick proposed the algorithm of AC [2], which
uses the finite state machine to match strings and can match all the pattern strings by scanning
text strings once. However, the algorithm of matching single pattern scans text strings once, and
it only can match one pattern string. In 1993, Fan Jang-Jong proposed the algorithm of AC-BM
[3], which uses the idea of the finite state machine of AC in preprocessing and uses the idea of
skipping of BM in scanning. The algorithm highly improves the traditional the algorithm of AC in
the part of scanning and matching.
In order to improve the speed of matching in watch list, this paper compares and
analyses some kinds of essential pattern matching algorithms, such as BM algorithm, AC
algorithm and AC-BM algorithm, and proposes an improved AC-BM Algorithm, a matching
algorithm of subsection, which not only combines with the advantages of BM algorithm and AC
algorithm, but also performances better than the traditional AC-BM algorithm. The speed of
matching is highly improved, so that it can satisfy the demand of on-line monitoring of anti-
money laundering. At last, this paper designs a model of the monitoring system, and structures
the module of matching, which is one of the submodules in the system, with the algorithm of the
improved AC-BM in detail. The structure can perfect the whole system of on-line monitoring of
anti-money laundering well.

 ISSN: 2302-4046
TELKOMNIKA Vol. 12, No. 1, January 2014: 379 – 387
380
2. Watch List and Matching Process
Watch list is a list about the staffs, organizations and classes of business who may lead
financial institutions to face the risk and the financial institutions need to pay close attention to.
The information of watch list comes from the international organization of anti-money laundering
(such as Financial Action Task Force on Money Laundering, Basel Committee on Banking
Supervision, Egmont Group, International Criminal Police Organization and so on), public and
media, the report of the financial internal monitoring system, the risk information of government
departments, while some professional institutes also collect the related information and release
it regularly. It is a legal duty of financial institutes to implement the on-line monitoring of watch
list. Figure 1 gives an example of watch list comes from the Bank of England, which shows
some important persons of Taliban. In addition, the watch list also includes senior officials of
government departments, corrupt officials, and tax evasion personnel and so on.
Figure 1. The watch list of Bank of England
The matching process of watch list is a matching between the watch list and the
database of account holders and clients’ transaction. Usually this process costs lots of time
because there are plenty of personal properties of the clients requiring matched and the
database is keeping expanding. Consequently, decreasing the consumed time of the matching
process is especially important to financial institutes when they are doing normal businesses
and the on-line monitoring of money laundering.
The watch list that this paper discusses is the basic data of the system monitor of
finance. Because of the large system of finance, a lot of subsystems will use the personal
information to confirm some businesses, such as opening an account, transaction, transfer and
so on. Because of the large volume of business, it is easy to cover the facts of crime. The
system of watch list achieves the goal of anti-money laundering by paying close attention to the
staffs that are in the high risk. When a person, who is in the watch list, opens an account or
trades with the personal information in the system of finance, the system monitor will sound the
alarm. Like this, the suspicious criminals will be discovered in time. Obviously, it simplifies the
process of anti-money laundering and it is convenient to supervise the money laundering.
The watch list mainly involves the data module of personal information. The systems
of account holders and clients’ transaction have close relations with it. When the operators help
clients open accounts or trade, the behaviors will involve the personal information. So the watch
list that this paper discusses is mainly about the systems of account holders and clients’

TELKOMNIKA ISSN: 2302-4046 
An Improved AC-BM Algorithm for Monitoring Watch List (Mo Yiwen)
381
transaction. Figure 2 shows the structure of matching process of watch list. Supervise the
money laundering by matching the information of the watch list. Judge the person whether is
suspicious by matching the client’s personal information with the watch list.Figure 1. Effects of
selecting different switching under dynamic condition
Figure 2. The structure of matching process of watch list
3. Study of Matching Algorithm
According to the requirement of filtering of the watch list’s content, this paper mainly
discusses the matching algorithm. On the basis of the number of pattern string, it can be
classified as algorithm of matching single pattern and algorithm of matching multiple patterns.
The former one matches only one string every time, while the latter one matches several strings
every time. The algorithm of matching single-pattern is represented by the BM algorithm; on the
other hand, the algorithm of matching multiple patterns is represented by the AC algorithm.
We assume P is a string with size m, P[0…m-1], then the P[i…j] is a substring of P,
P[i…j] is a prefix of P, and P[i…m-1] is a postfix of P. (0≤i≤j≤m-1).
Let T is a text string with size n and P is a pattern string. The characters of the P and
the T belong to a limited character set Σ; Therefore, the goal of the matching is to find out the
equal substring between P and T.
3.1. BM Algorithm
BM algorithm (Boyer-Moore) [1] is a kind of matching algorithm based on postfix.
Matching backwards from right side to left side is a distinctive characteristic of BM algorithm.
When there is a character that fails to match in text string T, T[i] =c, it will skip in the text string
to speed up the pattern string moving. If c is a string in P, finding out the rightmost c in P and
then moving P to make T[i] align it; if c isn’t a string in P, moving P to make P[0] align T[i+1].
The fundamental of BM algorithm: matching one character at a certain distance in T,
and then determining whether skip to the right side and the distance of skipping according to the
character, which is matched at present, whether appears in P or appears in which location. If it
doesn’t skip, it will compare with P from the right to the left in the present location. Otherwise it
will skip to the next location of P according to the skipping distance that has been computed in
advance.
The algorithm also includes two parts, preprocessing and scanning. In the part of
preprocessing, it just consider the pattern string p and the set of strings Σ. We introduce a
function of shift (c). It shows the situation that every character c appears in pattern string P, and
the distance that the pattern string moves. The function value of shift is also saved in an array.
The following is a method of computing the array of shift. If c appears in the pattern string P, find
out the location of c in the rightmost of P, and the location is index. The value of function is the
distance between the c of rightmost and the rightmost of P, it is m-1-index; if c doesn’t appear in
the pattern string P, the value is the length of P.
Supervise
the watch
list
The
process of
opening
account
The
process of
transaction
The data
set of
watch list
Supervise
money
laundering

 ISSN: 2302-4046
382
For example, Σ={a,b,c,d} and the pattern string P = abacab, we can compute it the
Shift(a)=1，，，Shift(b)=0 Shift(c)=2 Shift(d)=6. And the time complexity of shift is O(m+|Σ|).
Although the average efficiency of BM algorithm is high, it doesn’t performance well in
the worst case. The complexity of the worst case is O(mn), such as, a text string T=aaa…a, a
pattern string P=baa…a.
BM algorithm is designed as a kind of algorithm that matches the single pattern string
in text. Among the algorithms of matching single-pattern, BM algorithm is proved to performance
best. However, when there are various kinds of key words to match in the filtering and matching
of content, the BM algorithm has to match every kind of pattern. The time complexity of the BM
algorithm is O (n) when matching the single pattern, but it is O (kn) when matching the multiple
patterns.
3.2. AC Algorithm
AC algorithm (Aho-Corasick) [2] is a kind of algorithm that based on finite state
machine. It preprocesses the set of pattern strings to form a state machine in a tree before start
matching. It just scans the text string T once, and then it can find out the all patterns that match
with T in P.
The algorithm also includes two parts, preprocessing and scanning. It generates three
functions: the transition function: goto, failure function: failure and output function: output. The
following is the matching process of AC algorithm: starting from state zero, picking up a
character from the text string, and then going to the next state with goto function and the failure
function. When the output function of some state is not a null, it means that it successes to find
out the pattern string.
It structures the finite state machine in the part of preprocessing, and the time
complexity of structuring transition function is O(M), M is the total length of all patterns. The time
complexity of structuring the failure function is O (M), too. In the part of scanning, it scans every
character of the text T on the basis of the finite state machine that has structured before. Every
character just has one transition function, and the time complexity of scanning is O (n). So the
total time complexity of the algorithm that based on the finite state machine is O (M+n). The
time complexity is related to the length of text and pattern, but is not related to the content of
text and pattern. It means that the average time of scanning in the best case or the worst case
is the same, O(M+n).
The disadvantage of AC algorithm is that the demand of space is large. Too many
matching patterns will occupy plenty of space; even maybe make the system crash. So AC
algorithm can satisfy the demand and performances well in the case of a few patterns.
However, it doesn’t skip when it scans the text. It inputs the text in order, which means it can’t
skip the unnecessary comparison. Obviously AC algorithm is not the best matching algorithm in
practical process of matching.
3.3. An Improved AC-BM algorithm
3.3.1. AC-BM Algorithm
AC-BM algorithm [3] combines the AC algorithm and BM algorithm. It includes two
parts, preprocessing and scanning. The part of preprocessing structures a finite state machine
likes AC algorithm. The part of scanning based on the idea of skipping likes BM algorithm, and
matches the text string with the finite state machine. So it improves the speed of AC algorithm
efficiently in the part of scanning.
3.3.2. An Improved AC-BM Algorithm
On basis of the traditional AC-BM algorithm, this paper proposes an improved AC-BM
algorithm, a matching algorithm of subsection. Figure 3 shows the flow chart of the algorithm.
a. Let P align the left side of T, and start matching with BM algorithm. If matching
successfully once, we call it match completely, and then return the result. When Pi≠Ti, pick up
the Pi as L1, so L1={P1…Pi }.
b. Continue to match Pi+1 and make subsection when they are inequality.
c. Do step a and step b repeatedly, we get L1,L2…Lk.
d. Divide T into t1,t2, …tm, and { t1,t2, …tm }∈Σ.
e. Let { L1,L2…Lk } match { t1,t2, …tm } with the technology of the multi-thread AC
matching, and then return the result.

383
Figure 3. The flow chart of the improved algorithm
The improved AC-BM algorithm above uses the technology of subsection, so the length
of string can’t affect it when it is matching the single pattern. However the length of the string it
is, it affects the speed of matching less. In the matching of multiple patterns, the improved AC-
BM algorithm is not only able to match multiple patterns, which likes AC algorithm, but also skip
the unnecessary comparison, which likes BM algorithm. In addition, it improves the traditional
AC-BM algorithm and the improvement makes it better than the traditional one in the speed of
matching.
4. Experiment
In order to test the efficiency of the algorithms we mentioned above, we extract the
100MB data from the informational database of clients. And we extract randomly the pattern
strings from the database of watch list to test the algorithms. The system configuration is
2.4GHz Intel CPU and 2G RAM. We generate two sets of experimental data, and the size of the
data is 100MB. The first data set has no repetitive strings, and the second one has plenty of
repetitive strings. The length of pattern strings includes 10, 20, 30, 40, 50, 100, 500, 800, 1000.
Figure 4 shows the time of the first data set that every algorithm takes. Figure 5 shows the time
of the second data set that every algorithm takes.
Start
BM(P，T)
Pi ≠ Ti
YES…
L ={ L1，L2…Lk }
AC（，）L T
Result
Quit
Match completely
NO
Match successfully
L1={P1…Pi }；（BM Pi+1，）T
L1，L2={Pi+1…Pi+j }; BM（ Pi+j，T）

 ISSN: 2302-4046
384
Figure 4. The result of the first data set
Figure 5. The result of the first data set
It shows that the improved AC-BM algorithm performs best. When the number of
pattern sets increasing, the time of all the algorithms increase. Because the finite state machine
is larger and it has to take more time to transform; meanwhile, the times of matching increase.
Using the SMA algorithm, it can match by 100MB bytes per second and 1000 kinds of patterns.
So it improves the performance of the matching algorithm of key words well. In addition, as the
number of pattern strings increasing, the improved AC-BM algorithm is affected least. It means
that it’s convenient to add key words, and we don’t need to worry about that it may affect the
speed of matching. Obviously it is one of advantages of the improved AC-BM algorithm.
5. A Model of the Monitoring Watch List
5.1. The Structure of the Whole Monitoring System
The whole monitoring system should include the management and control center, the
module of preprocessing, the module of matching and the module of on-line monitoring and
analyzing. Figure 6 shows the structure of the system.

385
Figure 6. The structure of the monitoring systemt
The management and control center has a nice user interface, and shows the risk index
of the list that is detected at present. The module of preprocessing deals with the information of
watch list, and makes it standardization. The module of matching is used to match clients’
information with the watch list. The module of on-line monitoring and analyzing is used to
monitor the data of transactions in time.
5.2. The Structure of the Module of Matching
This paper mainly discusses the structuring of the module matching. Improving the
efficiency of this module will improve the operating efficiency of the whole system and help it
achieve the goal of on-line monitoring. The module of matching uses the algorithm of improved
AC-BM that this paper proposes to match and detect the clients’ information. If the module
discovers it there is someone suspicious or someone who maybe have some relationship with
suspicious organization after matching accurately, the system of monitoring will sound alarm. If
the module discovers it there isn’t any suspicious person or organization, it will save the clients’
information to the clients’ dataset. Figure 7 shows the workflow chart of this module.
Preprocessing
The dataset
of Watch list
The module of matching
Clients’ information
Alarm the suspicious data
Clients’
dataset
The module of on-line monitoring and
analyzing
The management and control
center
The situation of system

 ISSN: 2302-4046
386
Figure 7. The workflow chart of matching module
6. Conclusion
This paper focuses on the problems that anti-money laundering meets, and deliver
related research about the algorithms of matching pattern strings. On basis of BM algorithm, AC
algorithm and traditional AC-BM algorithm, we propose an improved algorithm, the matching
algorithm of subsection. The experiment shows that application of the algorithm to the matching
of watch list of anti-money laundering, could improve the speed of matching significantly. Even
the database of the watch list is expanding and the number of the key words is increasing, the
speed of the matching algorithm is affected least. These advantages can increase the speed of
matching watch list and meet the demand of on-line monitoring of anti-money laundering. In
addition, this paper designs a model of the monitoring system of anti-money laundering.
Applying the algorithm of the improved AC-BM to the module of matching in the monitoring
system will highly improve the efficiency of the whole monitoring system of anti-money
laundering.
Acknowledgment
This work is supported by National Social Science Foundation of China Project “Studies
on Suspicious Financial Transaction Monitoring System based on Behavior Pattern
Recognition” (09BTJ002).
References
[1] RS Boyer, JS Moore. A fast string searching algorithm. Communications of the ACM. 1977; 20(10):
762-772.
[2] AV Aho, MJ Corasick. Efficient string matching: an aid to bibliographic search. Communications of the
ACM. 1975; 18(6):333-340.
[3] Jang-Jong Fan, Keh-Yih Su. An Efficient Algorithm for Matching Multiple Patterns. IEEE Transactions
on Knowledge and Data Engineering. 1993; 5(2): 339-351.
[4] Alba LM. Evolution of Methods of Money Laundering in Latin America. Journal of Financial Crime.
2002.
NO
Imputing the clients’ information
Clients’ dataset
Sound alarm
Further monitoring
Watch list
Discover suspicious
information?
YES
Matching with the algorithm of subsection

387
[5] Alvaor E. Monge. An Adpative and Efficien Algorithm for Detecting Approximately Duplicate Database
Record. Califonia State University, Janeg. 2000.
[6] Jun Tang, Huan Liang, The building and perfection of watch list of the financial institution. Xinan
finance. 2011.
[7] Guogen Wan, Zhiguang Qin, Improved AC-BM Algorithm for Matching Multiple Strings. Journal of
University of Electronic Science and Technology of China. 2006; 35(4): 531-541.
[8] Siwei Zhou, Yong Cai. Mend of AC- BM A lgorithm and Application in Intrusion D etection Technique.
Microcomputer Applications. 2007; 28(1): 27-31.
[9] Brooks B. Money Laundering and Gambling: The New Zeal and Experience. Journal of Money
Laundering Control. 2003.
[10] Baker R. CaPitalism’s Achiles Heel: Dirty Money and How to Renew the Free-Market System. John
Wiley and Sons, 2005.
[11] Lu Jun, Zhang Zhuo, Mo Juan, LV xingfeng. Multi-pattern Matching Methods Based on Numerical
Computation. TELKOMNIKA Indonesian Journal of Electrical Engineering. 2013; 11(3); 1497-1405.
[12] Ye Peixin, Wei Xiujie. Lebeseguetype Inequality for Orthogonal Matching Pursuit for Microcoherent
Dictionaries. TELKOMINIKA Indonesian Journal of Electrical Engineering. 2013; 11(1); 213-226.
[13] N.Cereone, A,Tsuehiya eds. Special Issue on Learning and Discovery in Knowledge-BasesDatabase.
IEEE Transactions on Knowlddge and Data Engineering.1993.
[14] Dwyer T. ”Harmful” Tax Competition and the Futrue of offshore Financial Centers. Journal of Money
Laundering Control. 2000; 15(1); 48-69.
[15] Alexander R. The Role of Whistleblowers in the Fight against Economic Crime. Journal of Financial
Crime. 2004; 12(2); 131-138.
[16] Doudou La Loudouana, Mambobo Bonouliqui Tarare. Data Set Selection. Journal of Machine
Learning Gossip. 2003; 1: 11-19.
[17] Haddad E. Load distribution optimization in heterogeneous multiple Processor systems. Proceedings
of the Workshop On Heterogeneous Processing. 1993; 42-47.
[18] Jiawei Han, Mieheline Kamber. Data Mining Concepts and Techniques. Morgan Kaufinann Publisher.
2000.

An Improved AC-BM Algorithm for Monitoring Watch List

More Related Content

Similar to An Improved AC-BM Algorithm for Monitoring Watch List (20)

More from Nooria Sukmaningtyas (20)

Recently uploaded (20)

An Improved AC-BM Algorithm for Monitoring Watch List