ddc2011 - association
TRANSCRIPT
InceptionHow to guide users where they want to go
DATA MINING
User-Intended Guide Search
Web Cloud
Robot::Search docs having ‘ranking’
Robot:: Display documentsInformation Retrieval
Web Cloud
Search Engine
Robot (DAUMOA)
Keyword
N
F
Crawling
Indexing
Ranking
Web Cloud• Crawling• Indexing• Ranking
Search Engine
Keyword
Make users search correctly
How to Search?
Search by Typing
Users’ Intention
Search by Clicking
Provider’s Intention
Guide QueryUser Query
Search by Clicking?
In response to user action
Guide QueryUser Query
User-Intended Guide Query
Why?Correctness Ease to use Business
Suggest Speller Association
Anatomy of Association
101Introduction to Association
Abstraction
Apple
Clustering & Diversifying
Plausible (Fishable?) Options
AssociationAssociated words with the queryAnswers to the queryAdditional information for the queryQuery expansion or contractionQuery correction/reformulationQuery patternRecent issues related to the query
201Construction of Associations
Link. Sink. Rank.
L
S
R
• Keywords in sequential search• Click keywords of same doc.• Query keywords that display same doc.• Keywords from same documents• Contents & rule-based keywords
• Taboo keywords• Incorrect/mis-typed keywords• Morphologically-identical keywords• Representative keywords (+)
• More connections get more relevance• Click-through rate• Business-intensified keywords• Human-intervention
Sequential Keywords
Asso
ciated
Not As
socia
ted
{���} → {���}
Click Keywords
Click Keywords• � ��• � �� ��• �� ���• � �� ���• �� ���• �� �"��• ...
{� ��, � �� ��}{� ��, �� ���}
{� ��, � �� ���}…
{� �� ���, �� �"��}{ �� ���, �� �"��}
{� ��, �� ���}{� ��, �� ���}
{� ��, �� ���}{� ��, �� ���}
{� ��} → { �� ���}
Query Keywords
{�!� ���} → {�!� ���}
SK : CK : QK = 70% : 10% : 20%
FilteringAdult keywordsCopyright keywordsPrivacy keywords / Personal informationIncorrect/mis-typed keywords (with Speller)
Morphologically-identical keywordsSame keystrokes (i.e., Korean ⬌ English)Guide/Operation keyword pairsBanned: User requests (C/S)
Collective Intelligence
More is Better
301Advanced Topics I: Extension
ExtensionsProperty Description
Symmetric A → B then B → A
Transitive A → B → C then A → C
Triangular (A → C) & (B → C) then A → C(A → B) & (A → C) then B → C
Inclusive A ⊃ B → C then A → CA ≈ B → C then A → C
Me
C1 C2 C3 Cn
P
S1 S2 Sn
G
U1 U2 Un
Contents & Properties(keep working)
401Advanced Topics II: System & Service
In Service
DB
Operation: Daum ServiceAnalytics: SAS System
Index
Daily update 24h MNT
25M
Real-time Adaptive Systemwith MapReduce
CARTS
CoverageAccuracyRobustnessTimelinessSerendipity
4M