Data
Analytics and Collaborative Computing Group– Soft Computing, Data Mining,
Information Visualisation –
Applications
Applications
Here is a sampling of some of the applications developed by
our group.
Stream 1: Soft Computing
Simulation-based Evaluation of Workflow Escalation Strategies (2009)
Workflows in the service industry sometimes need to
deal with multi-fold increases in customer demand
within a short period of time. Such spikes in
service demand may be caused by a variety of events
including promotional deals, launching of new
products, major news or natural
disasters. Escalation strategies can be incorporated
into the design of a workflow so that it can cope
with sudden spikes in the number of service requests
while providing acceptable execution times. In this
research, we develop a method for evaluating
escalation strategies using simulation
technology. The effectiveness of the proposed method
is demonstrated on a workflow from an insurance
company.
Temporal Exception Prediction for Loops in Resource
Constrained Concurrent Workflows (2009)
Workflow management systems (WfMS) are widely used
for improving business processes and providing
better quality of services. However, rapid changes
in business environment can cause exceptions in WfMS
leading to deadline violation and other
consequences. In these circumstances, one of the
crucial tasks for a workflow administrator is to
detect any potential exceptions as early as possible
so that corrective measures can be taken. However
such detections can be extremely complex since a
workflow process may consist of various control flow
pattern and each pattern has its own way of
influencing temporal properties of a task. In this
research, we describe a novel approach for
predicting temporal exceptions for loops in
concurrent workflows which are required to share
limited identical resource pools. Our approach is
divided into two phases; preparation phase and
prediction phase. In the preparation phase, temporal
and resource constraints are calculated for each
task within the workflow schema. In the prediction
phase, an algorithm is used to predict potential
deadline violations by taking into account
constraints calculated from the preparation phase.
Modeling Support for Simulating Traffic Intensive
Web Applications (2009)
Business Process Simulation (BPS) has been widely
accepted as a modus operandi for evaluating business
applications and for understanding the strength and
weakness of the implemented system. However, less
attention has been devoted to applying BPS for
simulating traffic intensive applications hosted on
the Internet. These systems are inherently different
from typical enterprise-wide business applications
since the number of hits received at the application'
s Web site can fluctuate to a large degree depending
on the external factors (e.g. political unrest
causing overbooking in airlines). In addition, every
hit does not necessarily lead to a successful
completion of a transaction since users can navigate
away to other Web sites during the browsing. The
proposed system utilizes audit trail data and Web
server logs which contain vital information about
the interaction between clients and the Web
application.
Contact Tracing for Infectious Disease Control and
Quarantine Management (2009)
Highly infectious diseases such as SARS (Severe
Acute Respiratory Syndrome), Avian Influenza (Bird
Flu), Small Pox, and currently Swine Flu, to name
but a few, pose a significant threat to the global
population. Detection and prevention of infectious
diseases is notoriously complex and problematic due
to the ever increasing number of international
travelers. In addition, the risk of being infected
with an infectious disease in densely populated
urban areas tends to be much higher compared to
rural areas. When an outbreak occurs, the detection
of source of infection (or index case), clusters of
cases and transmission routes in a rapid manner is
crucial in preventing the infectious disease from
further spreading. Contact tracing has proven to be
helpful for these detections. Traditionally, contact
tracing is a field work of the medical personnel
with little assistance of IT (Information
Technology), if any. During the worldwide outbreak
of SARS in 2003, HCIS (Health Care Information
Systems) were built to facilitate contact
tracing. However, contact tracing, and thus the
detection process, is not a fully automatic process
in these systems. In this research, with SARS as a
case study, we realize detection as an automatic
process by applying algorithms and data mining
techniques in the patients' activities and social
interaction together with characteristics of the
infectious disease.
Fuzzy Adaptive Agent for Supply Chain Management
(2006)
Recent technological advancement in electronic
commerce has fuelled the need for designing
effective strategies for supply chain
management. Two crucial tasks in supply chain
management are the planning of raw material
acquisition for inventory and competing for customer
orders. Maintaining a flexible yet an adequate
inventory level is a complex task due to the
fluctuation in suppliers' production capacity and
changing customers' demand. Designing an effective
strategy for bidding customer orders is also a
difficult problem due to the intense competition in
fast changing market environment. In this research,
we describe the strategies of a supply chain
management agent which adaptively adjusts its target
inventory level and customer order bidding price
based on Fuzzy Logic reasoning. The agent has
competed in the 2006 Trading Agent Competition for
Supply Chain Management and has achieved good
result.
Stream 2: Data Mining
Integrated Enhancement of OLAP and Data Mining (2009)
OLAP performance and its data visualization can be
improved using different types of enhancement
techniques. Previous research has taken two separate
directions in OLAP performance improvement and
visualization enhancement respectively. Some recent
works have shown the benefits of combining OLAP with
Data Mining. Our previous work presents architecture
for the enhancement of OLAP functionality by
integrating OLAP and Data Mining. In this project,
we proposed a novel architecture that not only
overcomes the existing limitations, but also
provides a way for an integrated enhancement of
performance and visualization. We have developed a
prototype and validated the proposed architecture
using real-life data sets. Experimental results show
that cube construction time and its interactive data
visualization capability can be improved
remarkably. By integrating enhanced OLAP with data
mining system a higher degree of enhancement is
achieved which makes significant advancement in the
modern OLAP systems. Finally, we emphasize the
coupling of OLAP and Data mining in a framework that
supports both integrated OLAP performance
improvement and visualization enhancement.
Mining for Performance Evaluation of e-Government
Portal (2009)
An e-Government portal represents not only a
reputable public image of the sovereignty of a
region, it is responsible for serving many users
from local citizens and beyond, in a reliable
way. The requirement for robustness of e-Government
portal becomes relatively stringent. In view of
this, it would be very useful to have an independent
server that serves as a 24/7 watch-guard to monitor
the performance of the service portal. In this
project, we designed a Web-based performance
monitoring system (WMS) for checking the health
status and some extras, of the service portal in
real-time. Some basic server statuses are usually
provided by the web hosting company at which the
service portal is hosted. Usually they are simple
statistics such as the number of hits and server
loads over a selected period of time. Further
details regarding the performance of the portal
however may not be available or they come in
expensive prices. This chargeable service is either
by some third-party proprietary commercial software
tools on the market or by provision of external
analytic jobs of consultancy. In contrast, the
advantages of WMS are additional performance checks
and open source in Java for easy future
development. It features web usage mining from web
log files in addition to the computation of
statistics offered by the hosting company. A higher
level of analytic insights can be obtained too when
the two combined. The functions of WMS include (1)
Web Log Analysis, (2) Web Usability Analysis, (3)
Website Performance Benchmarking, (4) Web Link
Validation, and (5) Performance Reporting. WMS can
be used as a generic monitoring tool for virtually
all kinds of e-business models.
Time Series Trend Analysis in Stock Market (2008)
Trend-following (TF) strategies use fixed trading
mechanism in order to take advantages from the
long-term market moves without regards to the past
price performance. In contrast with most prediction
tools that stemmed from soft-computing such as
neural networks to predict a future trend, TF just
rides on the current trend pattern to decide on
buying or selling. While TF is widely applied in
currency markets with a good track record for major
currency pairs, the strategies are doubted if they
do apply universally. In this project a new TF model
that features both strategies of evaluating the
trend by static and adaptive rules, is created from
simulations and later verified on Hong Kong Hang
Seng future indices. The model assesses trend
profitability from the statistical features of the
return distribution of the asset under
consideration. The results and examples facilitate
some insights on the merits of using the trend
following model.
Decision-support for Optimizing Supply Chain
Formation (2008)
Supply chain formation problem is one of the
important research topics in e-Commence. In an
e-Marketplace where buyers and sellers meet and
trade online, dynamic supply chains can be formed
among them by mediating agents. SET and CSET are two
typical make-to-order supply chain models. CSET
represents a scenario that has a central authority
in charge of the formation, management and
dissolution of a supply chain. The principal
authority selects the partners under certain
principles which may either aim for maximizing
profits of the whole supply chain or for ensuring
every partner to receive a job for communal
prosperity. In SET, every supply chain partner uses
local knowledge to compete for jobs at each supply
chain level. We have implemented a Java-based
simulator for simulating the process of dynamic
supply chain formation. The simulator can operate in
both modes whose results are useful in
decision-support in supply chain planning.
Detecting Suspicious Patterns in Secure Physical
Environment (2007)
Security in physical environments has become
increasingly important in the wake of terror and
criminal activity, particularly over the past
decade. One of the challenges is to identify
activities that may not be outright illegal or
breaches of security, but that are suspicious,
i.e. where there is a possibility that these
activities may lead to breaches of
security. Technology such as RFID is used to track
the access and movement of people in highly security
physical environments. This project searches for
methods of detecting patterns of suspicious activity
in logs collected by such physical access control
systems. It also outlines methods of predicting
future suspicious activities based on such
logs. Based on this concept, a real-time suspicious
access patterns detection system is to be developed,
which can provide a powerful security measure to
RFID systems, in a closely monitored physical
environment. The proposed system may monitor the
physical object movements and hunt for potential
security threats out of a large number of normal
object movements.
An Automated University Admission Recommender System
for Secondary School Students (2007)
University or college admission is a complex
decision process that goes beyond simply matching
test scores and admission requirements. Past
research has suggested that students' backgrounds
and other factors correlate to the performance of
their tertiary education. However, almost all
admission and enrollment studies are based on the
perspective of universities or colleges, and only
few studies are based on the perspective of
secondary schools. This project presents a hybrid
model of neural network and decision tree classifier
that serves as the core design for a university
admission recommender system. The system was tested
with live data from sources of Macau secondary
school students. In addition to the high prediction
accuracy rate, flexibility is an advantage such that
the system can predict suitable universities that
match the students' profiles and the suitable
approaches through which the students should
enter. The recommender can be generalized into
making different kinds of predictions based on the
students' histories.
GA-Based Collaborative Filtering For E-Commerce
Recommenders (2007)
Many e-Commerce websites have widely adopted recommenders systems to automatically suggest products or services to customers for enhancing their online experiences. Amazon.com is a classical example. Recommenders assist users to narrow down their choices for making a purchase decision from a large pool of items. A simple recommendation can be generated based on the top selling products. This however assumes every customer has a common taste by a 'one-size-fits-all' approach. A more advanced approach is known as one-to-one marketing, which considers the profile of a particular user and searches for a personal match of product that is predicted to be most likely he will favour best. The technical core mechanism in e-Commerce recommenders is usually implemented in software programs by three types of filtering techniques, namely: Content-based filtering, Collaborative Filtering (CF) and Hybrid Models.
Since different types of filtering techniques have their shortcomings but advantages in certain aspects, combining them may be a promising solution provided there is a novel way of overcoming a large amount of input variables. When the filtering techniques are fused to work together, almost all the information and the relative weights about the products, users, and users' activities would be required. Genetic algorithm (GA) is an ideal optimization search function, for finding a best recommendation out of a large population of variables. Here, we present a GA-based approach for supporting hybrid modes of CF. We show how the input variables can be coded into GA chromosomes in various modes. Insights of how GA can be used in e-commerce recommenders are derived through our experiments.
Data Mining GSM Network Resources for Supporting QOS
of Mobile Payment (2006)
In mobile commerce, short-message-service (SMS) is
an important technique for delivering payment
instruction. A payment model "SMS Credit" was
proposed earlier by us. Such payment service or
similar relies on the transmission of SMS; it is
needed to reduce the occurrence of packet losses,
delay and to improve the quality of packet
transmission services (QOS) in the network. This
project investigates on how the payment service
operates in a configurable radio resource
environment via data mining. A Radio Resource
Management and Prediction Server equipped with data
mining algorithms will optimize the radio resources
for both voice and data services in order to provide
an optimized QOS. Specifically, data mining
techniques are applied to define traffic policy and
to calculate optimization result through traffic
profile analysis.
Web Watcher Agent for Marketing Information
Monitoring (2004)
World-Wide-Web is a huge pool of valuable
information for companies to know what their
competitors are doing and what products and services
they offer up-to-date. Companies can gather business
intelligence from the Web for planning
countermeasures strategies. Hence it is crucial to
have the right tool to effectively gather such
information from the Web. Many information retrieval
and monitoring technologies have been developed. But
they are more for generally tracking changes and
downloading the whole websites for offline
browsing. This project is to study on specifically
the design of a Web monitoring system for gathering
business information relevant to a company. The
Watcher Agent is a server-based system that is built
with two main parts, namely Price Watcher and Market
Watcher. The system will assist company users in
price information collection, news information
filtering, and product ranking estimation, thus
saving time and effort for them.
Stream 3: Information Visualisation
Wiki Recent Changes Visualization (2011)
This work has developed a visualization that allows
the display of current and recent editing activity
in Wikipedia, enabling an administrator to quickly
obtain an overview of some noteworthy editing
patterns occurring in the Wikipedia at the present
moment.
Wiki Category Radial Visualization (2011)
This project has developed a visualization of the
Wikipedia content (i.e. all Wikipedia articles) in a
radial representation which highlights latent
connections between co-assigned categories of these
articles.
Map-like Wiki Category Visualization (2011)
This work has developed a visualization of the
Wikipedia category hierarchy in the form resembling
a geographic map.
Wiki Category Matrix (2010)
This work was our first effort to create an overview
visualization of category assignments in
Wikipedia. In Wikipedia authors are free to classify
articles into categories, making as many such
category assignments as they wish. Categories
themselves are hierarchically organized, with
sub-categories down any number of levels. We were
interested to know the distribution of content over
categories, expecting to find gaps or relatively
sparse areas, and other areas with a large
concentration of content. In order to test this
hypothesis we decided to create a graphical
representation of category assignment that would
allow us to visually identify such patterns. To do
so the first step was to simplify the category
hierarchy, which contained both loops and multiple
parents, into a tree structure. We then calculated
similarity between categories based on a cosine
similarity function. Having similarity values, we
arranged categories in a row, with the most similar
categories next to each other. This resulted in a
one-dimensional spectrum of categories which we
mapped onto both the x and y axes of a
matrix. Thereafter we traced the top-level and
second-level categories that each article belonged
to by traversing the tree from the article's
assigned category up to the root. Taking each
article's first two assigned categories (which we
assumed to be most representative of its content),
we represented the number of articles assigned to a
given pair of categories as a proportionally sized
dot aggregating all articles assigned to these two
categories on the corresponding x,y coordinate in
our matrix. Moreover, second-level category
assignments ("small dots") belonging to the same
top-level category are aggregated to a single
top-level category dot ("big dots"). The result is
shown in a matrix like in the above figure (as the
matrix is symmetric mirrored along its diagonal we
only draw half the matrix). Our hypothesis is indeed
confirmed. There are some categories that
consistenly have many articles with every other
category (taken articles' first/second category
assignment), whereas other categories have a much
less even distribution of articles over second
categories. The figure shown here is from the
Swedish Wikipedia and shows some highlights of
top-level categories with sub-categories (a); and
numbers of articles by first/second category
visualized as dots (b).
BambooGarden (2009)
This work developed an information visualization
application for visualizing a user's instant
messaging history. Users may want to reflect on
their communication patterns with their chat
partners: obtaining a quick overview of their
message history, who the main chat partners are in
terms of bulk of chat conversations, whether chat
messages are happy or sad in content, whether
conversations tend to be short or drawn out, whether
they are sporadic or regular daily exchanges, and
many other social aspects connected with their
person-to-person instant messaging communication. In
order to reveal these aspects of the communication
we have employed the bamboo as a visualization
metaphor: each chat partner (i.e. IM buddy) is shown
as a bamboo plant, and the group of all chat
partners is shown as a bamboo garden. Each chat
conversation is shown as a branch on the bamboo
trunk. As the vertical dimension corresponds to time
(growing from bottom up, like the bamboo itself),
the position on the trunk indicates the
conversation's position in time. The length of the
branch correponds to the length of the conversation,
with each chat message represented as a leaf on the
branch, its position corresponding to the time of
the message during the conversation. Leaf size
corresponds to message size. This part of the
visualization is somewhat similar to PeopleGarden, developed by Xiong and Donath at
MIT in 1999 and which gave an inspiration to our
work. However, BambooGarden goes beyond PeopleGarden
in analyzing and visualizing the emotional state of
the chat messages. We look for emoticons (smiley
faces) and certain keywords in the chat text to
infer one of four different emotional states:
happiness, sadness, anger and a neutral state. These
emotional states are mapped to four different
colours.
Wiki Analysis (2009)
This project developed an extension to MediaWiki,
the wiki engine used to run the popular Wikipedia site and
other wiki sites. This extension provides analytical
abilities, tightly integrated in the MediaWiki software and
user interface. For any page, the user of the Wiki Analysis
extension can do three kinds of analysis: (1) text
comparison, to compare differences between versions,
which includes not only insertion and deletion of text which
are standard in MediaWiki but also a GA-driven matching
function for finding similar text; the ability to detect old
text that has been replaced with new text; detecting text that
has been moved up or down relative to other text; and a
highlighted text display showing which word was contributed by
(or "belongs" to) which author. (2) page evolution,
which shows the evolution of both the article page and its
attached discussion page over time, in absolute size; and
allows the display of an individual author's contribution and
their evolution over time, both as all cumulative
contributions as well as all surviving contributions after
being edited. (3) contribution summary which displays
the amount of contributions by different authors to different
parts of the wiki (article and discussion, volume and number
of times of contribution). (4) user-to-page
contribution which visualizes the temporal patterns of
one or more users contributing to different wiki pages over a
given period of time.
MediaWiki Co-Authors and Expert Finder (2008)
In this work targeted at MediaWiki, the system
underlying the Wikipedia collaborative
user-contributed encylopedia, we developed a method
and algorithm for determining the degree of
co-authorship among the users who contribute to wiki
articles. Unlike in conventional authoring where
co-authoring is explicit and usually spans the
entire writing process, in wikis co-authoring is
implicit and can be temporally disjoint, thus the
co-authoring relationship can be of different
strength. Our algorithm measures the degree of
co-authorship among a pair of authors. We have
implemented this as a MediaWiki extension that can
find the co-authors of a given user. A further
MediaWiki extension builds on this to uncover
expertise groups on a given topic in the wiki
system, given the extent of involvement of
individual authors as well as their significant
co-authors.
Wiki Text Analyzer (2007)
In wiki systems users collaborate with each other in
the production of content. To uncover detailed
aspects of these processes we have developed a
visualization system that allows us to perceive user
involvement, contribution, process evolution, and
patterns of engagement with the text and each
other's contributions.
Visualization for Google Desktop Search Engine
(2006)
Modern desktop computers contain thousands, possibly
tens of thousands of files. Desktop search engines
help in locating files that the user knows are
"somewhere there". However, search results
themselves can be overwhelmingly large. Therefore
this system interfaces with the Google desktop
search engine to present search results in graphical
form as a visualization of pertinent information. A
2-dimensional display of the directory hierarchy of
the user's desktop computer is generated, and search
results are highlighted in the display. The
development of this system gave us the opportunity
to explore layout problems of large tree-structured
hierarchies.
WikiVis - Wikipedia Visualization (2005)
Wikipedia is a popular online user-contributed
encyclopedia with numerous different language
editions. This project has developed a visualization
of the Wikipedia information space, primarily as a
means of navigating the category hierarchy as well
as the article network. The project is implemented
in Java, utilizing the Java 3D package.