The First International Workshop on
Unified Data Mining Engine: Addressing Challenges

UDME 2007

Montréal, Canada, October 22, 2007
(in conjunction with OOPSLA 2007)

http://www.oopsla.org/oopsla2007 (OOPSLA 2007 Link)
http://www.oopsla.org/oopsla2007/index.php?page=sub/&id=160 (Workshop Link 1)
http://www.engr.sjsu.edu/~fayad/workshops/UDME07 (Workshop Link 2)
http://www.vrlsoft.com/workshops/UDME07 (Workshop Link 3)

Data mining is the discovery of knowledge of analyzing enormous set of data, by extracting the meaning of the data and then predicting the future trends. Data mining helps us to find out secret information from large databases, and also helps companies to take sound decisions, based on knowledge and information. 

If we closely take a look into any data-mining tool, we can see there are some common core logic, which are independent of the data and the applications, but most of existing implementations try to ignore that fact and concentrate on the specific problem, in that way the tool becomes limited to only to a particular set of data for specific application.

Data mining is also finding interesting patterns in data. The main challenge of any data-mining engine is how to apply different algorithms or different techniques, on different set of data, to find interesting pattern, which is very useful to business. It is extremely difficult to come with some standard way of analyzing the data. The enormous volume and the complexity of the data make it impossible to run same algorithms on different dataset. Nowadays, there are different vendors, who are trying to solve this problem, but mostly they support a subset of different algorithms. None of them has come up with any stable engine, which can work in any data set and in any domain.

In the last decade, the improvement in storage and CPU speed has created a huge opportunity for different data mining application, ranging from CRM to medical health care application. The evolution of data mining is shown in table 1. 

Now it is very difficult to develop a single application, which can take care all of these problems. It’s a dream even to think of an application, which can iterate through any data and will find pattern. Data mining also deals with useful pattern, not just patterns, now whether a pattern is useful or not, depends on the context where it is usually applied. Present day tools depend solely on the expert about what kind of algorithms to apply, and how to analyze the output, because most of them are generic, and there is no context specific logic is attached to the application.

Table 1. Steps in the Evolution of Data Mining [12].

Here is s summary of the problems that we face today in the existing data mining tools

1.       Difficult to use– Existing data mining tools try to cover all different data mining applications, thus it becomes very difficult to configure and run.

2.       Needs Expert to run the tool – No domain or problem specific logic is tied with the tool, therefore needs expert to run the to tool and analyze the result

3.       Difficult to add new functionality - Because of the size and complexity of each tool, it is very difficult to add any new feature.

4.       Difficult to interface -  There is no way  those algorithms developed by some other companies, can be integrated with the tool easily

5.       Short Lifetime -  There is no stable component in the tool  and with time the tool become obsolete, as new tools take the market, changing the exiting tool to incorporate new feature is difficult and require lot of changes.

6.       Limited Number of algorithms – Existing tool only provide limited number of algorithm and sometime use of multiple algorithms is very limited.

7.       Need lot of resources: Existing tools are not optimized for any specific application, therefore they need lot of resources, such as runtime memory, hard disk etc.

Thus, this workshop is driven forward by three main questions. First, “how can we develop a unified data mining engine {UDME)?” Second, “what kind of technologies and tools to  build such an Engine?” and third, “how can we overcome the existing problems?”

Building such an engine is not an easy exercise, specifically, when several factors can undermine their quality success, such as cost, time, and lack of systematic approaches. We would like to architect and develop a Unified Data Mining Engine (UDME), that has the some or all of the following properties:

1.       Ease of use– Multiple tools can be developed easily by focusing on specific problems, because they all can share the core services, that are provided by the UDME.

2.       No Need of Expert to run the tool – Domain specific knowledge such as verification, selection of tool etc, can be implemented in the tool itself, while developing the tool.

3.       Easy to add new functionality - – The application specific logic should be separate from the core logic, therefore new application specific functionality can be added easily, without making any change in the core logic.

4.       Easy to interface -  The design should be based on system of independent patterns, they can be developed by 3^rd party vendors.

5.       Long Lifetime -  The engine should be based on stable core logic, which has a long lifetime, the application logic should be loosely connected which can change over time.

6.       Multiple algorithms – The engine must support any number of algorithms.

7.       Fewer resources: The proposed engine should be developed by connecting several patterns or components. Depending on the application, a  domain the engine can use patterns or components, which are necessary therefore it needs less resources compare to existing tools.

8.       Stable: The engine should be stable over time, and provide a simple way to apply different data mining and data analysis algorithms on different sets of data in any domain.

9.       Isolation of Application logic: We must also isolate the stable knowledge from any application specific logic, therefore different applications can use the same core knowledge, which need not to be changed.

10.   Minimum Maintenance Cost – Maintenance cost of such an engine should be very minimal.

The workshop will address the unified data mining engine challenges and debate several issues  that are related to the following questions. We also want researchers, framework developers, and application developers to discuss and debate the following questions related to:

I.                   UDME Architecture

a.       What is the best approach for building such an engine?

b.       What are the bases of creating the engine architecture?

c.       Are there any guidelines, methodologies, and/or processes for an engine architecture creation and development?

d.       What are the components of the unified data mining engine architecture?

e.       What kind of patterns or components that appear in UDME ?

f.        Show how your engine architecture meets the above UDME properties.

II.                UDME Development

a.       What is the ultimate way to develop such an engine?

b.       What are the techniques and tools for developing such an engine?

c.       Show how to extend your engine to the new application logics?

More information will be available at:

http://www.oopsla.org/oopsla2007 (OOPSLA 2007 Link)

http://www.oopsla.org/oopsla2007/index.php?page=sub/&id=160 (Workshop Link 1)

http://www.engr.sjsu.edu/~fayad/workshops/UDME07 (Workshop Link 2)

http://www.vrlsoft.com/workshops/UDME07 (Workshop Link 3)

Detailed instructions for electronic paper submission and review process are found at http://www.compsac.org/. Developers and programmers, who are interested in participating in the workshop, are requested to submit a short position paper (3-5 pages), or regular workshop paper (limited to 6-15 pages, double spaced, including figures) by representing views and experiences that are relevant to the given discussion topic. The title page must include a maximum 150-word abstract, five keywords, full mailing address, e-mail address, phone number, fax number, and a designated contact author. Workshop papers will be selected depending on their originality, quality and relevance to the workshop.  All submitted papers will also be evaluated according to their originality, significance, correctness, presentation and relevance. Papers should be submitted electronically to the chair.  Please follow the instructions that are provided on the web page. Camera Ready manuscripts must be submitted following ACM SIGPLAN conference proceedings style and guidelines. We also encourage authors to present novel and fresh ideas, critiques of existing work, and practical studies.

Each accepted workshop paper must be presented in the person, either by the author or by one of the co-authors.  To foster and promote lively discussions, authors are encouraged to present open ended questions and one or two main statements for the purpose of discussion at the workshop.  Submissions must be made either in MS-Word or RTF formats (Please, DO NOT compress files).

Depending on the total number and spread of contributions, the scope may be further narrowed down  to ensure an effective communication and information sharing session. Accepted position papers will be distributed to the participants, just before the workshop and will be made generally available through the WWW and FTP.   Accepted papers will also be published in the Workshop Proceedings. At least one of the authors of each accepted paper must register, as a full delegate in the workshop.  Selected papers will be published in one of the future issues of the online International Journal Of Patterns (IJOP), www.ijop.org and/or  International Journal of Software Architectures (IJSA), www.ijsa.net

People who are interested in participating in the workshop, without making any submissions are requested to fill out the participation form and e-mail to any of the workshop chairs. 

-------------------------------------------------

PARTICIPATION FORM:

Name and Affiliation:

Position: 

Address:

E-mail:

URL:

Areas of interest:

Reasons for Attending?

-------------------------------------------------

Please note that registration is absolutely mandatory, in order to participate in the workshop.  An early registration discount is made available for all desired participants.  An overhead projector and a flipchart will also be made available to all participants.   

For more information please visit any of the following websites:

http://www.oopsla.org/oopsla2007 (OOPSLA 2007 Link)

http://www.oopsla.org/oopsla2007/index.php?page=sub/&id=160 (Workshop Link 1)

http://www.engr.sjsu.edu/~fayad/workshops/UDME07 (Workshop Link 2)

http://www.vrlsoft.com/workshops/UDME07 (Workshop Link 3)

You may also contact the organizers, either by e mail or by phone.

1. Welcome and introduction of participants. The organizers will first provide  a short overview of all open issues, and also of the main arguments arising out of the position papers. (Estimated time: 20-30 minutes)

2. Selected authors (who’ll be representing the main trends) will be allotted  20 minutes, to explain, how their position relates to other positions, and what each one of them sees as the three major issues. We are expecting about 5-10 position papers in this session.  (Estimated time: 120-130 minutes)

3. The organizers will also propose an identification process of the major issues, and the participants will then discuss, choose and select what they perceive are the hottest issues to be examined and analyzed. (Estimated time: 10-15 minutes)

4. The participants will work for 70-95 minutes in small groups, with a designated moderator assigned for leading each group. The groups will then individually deal with two identified,  but different hot issues, and will produce a summary note in the form of points and counterpoints, showing either how several views are irreducibly opposed or how they are complementary.  The total number of groups will depend mainly on the number of participants and issues selected; ideally there should be 3-5 people in each group. (Estimated time: 60-70 minutes)

5. Each group will be provided10-15 minutes to present its findings and inferences to the workshop.   A closing discussion will soon follow. The workshop report will be composed on the basis of these findings, and will include a clear cut agenda for future exploration and cooperation; this will be made available through the WWW and FTP. (Estimated time: 50-60 minutes for five teams)

(Total estimated time: 285-315 minutes, i.e. about five hours +/- 15 minutes; lunch and breaks are not included.)

IMPORTANT DATES -- Will be updated based on acceptance process.

Submission deadline                       September 14, 2007

Acceptance notification                   September 30, 2007

Camera-ready paper due               October 10, 2007

Workshop date: Starts at 8:30 a.m         October 22, 2007

Dr. M.E. Fayad  (Chair)

Professor of Computer Engineering

Computer Engineering Dept., College of Engineering

San José State University

One Washington Square, San José, CA 95192-0180

Ph: (408) 924-7364, Fax: (408) 924-4153

E-mail: m.fayad@sjsu.edu, mefayad@gmail.com

http://www.engr.sjsu.edu/fayad

Dr. Tarek Helmy (Co-Chair)

College of computer science and engineering,

Department of Information and Computer Science,

King Fahd University of Petroleum and Minerals,

Dhahran 31261, Mail Box. 413, Saudi Arabia.

Ph: 9663-860-1967 (Office)

E-mail: helmy@ccse.kfupm.edu.sa

Dr. Rami Bahsoon (Co-Chair)
School of Engineering and Applied Science
Aston University in Birmingham, Birmingham B4 7ET, United Kingdom
office: Main Building, Second Floor, MB 213E
Ph:  +44 (0) 121 204 3464
fax:  +44(0) 121 204 3681
URL: http://www-users.aston.ac.uk/~bahsoonr/index.htm

Professor Dilip Patel (Co-Chair)
Faculty of Business, Computing and Information Management
London South Bank University
103 Borough Road
London SE1 0AA, United Kingdom
TEL: +44 (0)20 7815 7429

Somenath Das (Co-Chair)

eBay, Inc.
2211 North First Street
San Jose, CA 95131, USA

Ph: 408 967 4151

E-mail: sodas@ebay.com

Eduardo M. Segura (Co-Chair)

vrlSoft, Inc.

2065 Martin Ave., Suite 103

Santa Clara, CA 95050-2707

Phone/Fax: (408) 654-8972

E-mail: esegura@vrlsoft.com, eduardo.segura@sjsu.edu

http://www.vrlsoft.com

Rami Bahsoon, Aston University in Birmingham, United Kingdom

Rogerio Atem de Carvalho,  Federal Center for Technological Education of Campos, Brazil

Chia-Chu Chiang, University of Arkansas, Little Rock, USA

Issam Wajih Damaj, Dhofar University, Salalah, Sultanate of Oman
Somenath Das, eBay, Inc., USA

Dilip Patel, London South Bank University, United Kingdom

Jurgen Dix, Clausthal University of Technology, Germany

M.E. Fayad,  San Jose State University and vrlSoft, Inc, Silicon Valley, USA

Jaafar Gaber, Université de Technologie de Belfort-Montbéliard, France

Rosario Girardi,  Federal University of Maranhão, São Luís, Brasil

Dr. Tarek Helmy,  King Fahd University of Petroleum and Minerals,  Dhahran, Saudi Arabia

Hoda Hosny, The American University in Cairo, Egypt

A. Kannammal, Coimbatore Institute of Technology, TamilNadu, India

Mohamed-Khireddine Kholladi, University of Constantine, France
Dae-Kyoo Kim, Oakland University, USA

Roger (Buzz) King,  University of Colorado, Boulder CO, USA

Jianzhi Li, De Montfort University, United Kingdom

Nashat Mansour, Lebanese American University, Lebanon

Tokuro Matsuo,  Yamagata University, Japan

Srini Ramaswamy,  University of Arkansas, Little Rock, USA

Miguel Garre Rubio, Universidad de Alcalá, Madrid, Spain

Eduardo M. Segura,  San Jose State University and vrlSoft, Inc, Silicon Valley, USA

Jaroslav Zendulka,  Brno University of Technology, Czech Republic