network data mining: methods and techniques for discovering deep linkage between attributes from :...

22
NETWORK DATA MINING: METHODS AND TECHNIQUES FOR DISCOVERING DEEP LINKAGE BETWEEN ATTRIBUTES FROM ACM SIGKDD YEAR 2007 組組 組組組組 組組組組 組組組 :、、

Upload: joella-beasley

Post on 13-Dec-2015

225 views

Category:

Documents


3 download

TRANSCRIPT

NETWORK DATA MINING: METHODS AND TECHNIQUES FOR DISCOVERING DEEP LINKAGE BETWEEN ATTRIBUTESFROM : ACM SIGKDD YEAR : 2007

組員:黃于珊、李界寬、程尚文

Outline

Introduction Network Data Mining Step1:Translate biz problem into DM problem Step2:Select appropriate data Step3:Get to know the data Step4: Create a model set Step5: Fix problems with the data Step6: Transform data to bring information to the surface Step7:Build models Step8:Assess models Step9:Deploy models Step10:Assess results Critics to the work

Introduction

This paper present a human- centred network data mining methodology Addresses the issues of depicting implicit

relationships between data attributes and/or specific values of these attributes.

Case: 利用 Network Data Mining 來偵測與發現可疑交易情形,協助發現是否有詐騙的情形產生,並且進而預防詐騙事件的發生。

Network Data Mining(1/2)

Definition: the process of discovering emergent network patterns and models in large and complex data sets mining network models out of data sets mining network data

Network Data Mining(2/2)

Network data mining as a human-centered knowledge discovery process

Step1:Translate biz problem into DM problem 利用 Network Data Mining 來偵測與發現可

疑交易情形,協助發現是否有詐騙的情形產生,並且進而預防詐騙事件的發生。

Step2: Select appropriate data A typical social network analysis

research scenario involves data collection through

questionnaires or tables individuals describe their interactions with

other individuals in the same social setting 案例資料( Data )來源自為一間位於澳洲的

保險公司,抓取其資料庫中之五千筆的機車責任險資料作為 Pilot Project (先導專案)。

Step3: Get to know the data

Collected data is then used to infer a social network model nodes represent individuals edges represent the interactions between these

individuals. Classical social network analysis studies deal

with relatively small data sets and look at the structure of individuals in the network, measured by such indices as centrality Connectivity

The data from the company was essentially open-ended and without specification

Step4: Create a model setStep5: Fix problems with the data A set of linkages :從其資料庫中抓取責任險

中的五個資料欄位的資料作為 Data set 個人 地址 保單號碼 電話號碼 銀行帳戶

Step6: Transform data to bring information to the surface Looked at three fields of data and the

linkages identified between them : 個人 地址 保單號碼

Step7:Build models(1/2)

Tool : NetMaphttp://www.netmap.com.au/

Slogan of NetMap"We reveal the invisible truths hidden in

your data“ Power of NetMap

Fraud detection Monitoring data integrity Auditing of processes and systems Mapping how an organisation works

Step7:Build models(2/2)

Demo http://www.netmap.com.au/demo.html

Name

Claim

Address

Step8:Assess models

Simons JL!

Step9:Deploy models(1/1)

以 Simons JL為出發點

Portion of Data

Step9:Deploy models(2/2)

發現之規則: Simons → Wesson → Verman

Step9:Deploy models(3/3)

Enrich:銀行帳戶、電話號碼Enrich:銀行帳戶、電話號碼

以 Verman為出發點

Step10:Assess results

由圖可知, Verman 的電話號碼與 Wesson有關聯性,因此可以認為 Verman 為詐欺犯。

Conclusion(1/2)

• Describe the concept of network data mining and present a case study

• Focuses upon knowledge discovery• Main steps(Fayyad et al. 1996):

– 1. Define scenarios in terms of query specifications and exception rules

– 2. Process the data– 3. Interpret or initiate action– 0. Discover patterns and qualify them as

scenarios• Discovery and exception detection are inter-

related, but discovery usually takes place first

Conclusion(2/2)

Verman’s case illustrated the cornerstone of the network data mining approach

Successful discovery occurred through Train of thought Discovery tool A complementary usage of traditional data

mining. Ex: the same family name

Critics

Depends on the intuitive power of human needs the aid of the specialists wrong judgment in exceptional conditions

Ex: wrong data Needs much time and efforts of human

Thanks for your listening!!