Efficient mining and prediction of user behavior patternsin mobile web systems
Vincent S. Tseng , Kawuu W. Lin
Information and Software Technology 48 (2006) 357–369
69821002 朱玉棠69821016 黃弓凌69821028 張治軍
Outline
Introduction & system architectureMining of sequential mobile access
patterns-SMAPPrediction strategiesExperimental evaluationConclusions & associated thinking
Introduction
What benefits for effectively modeling the behavior patterns of users?
To help the user get desired information in a short time
behavior patterns: a sequence of requests of a user form a location-service stream
Introduction
System architecture
SMAP-MINE:Construction of SMAP-Tree
User ID Access pattern
123456
<(a,1)(b,2)(c,5)(d,8)><(a,1)(b,3)(c,5)(d,8)><(a,3)(b,2)(d,7)><(c,6)(b,2)(d,7)><(c,8)(b,1)><(a,3)(b,6)(c,8)(d,7)>
SMAP-Tree
SR-Tree(service request tree)
SMAP-Mine algorithm
Threshold: δ (30%→6x0.3=2)
SMAP-Mine algorithm
CMAP-Mine
3
c:2
B: A:
8:2
SMAR prediction
Sequential mobile access rulesSMAR-Location SMAR-ServiceSMAR-L&S
Strength = sup * conf
( RHS = LHS * conf )
)(),)...(,)(,( 112211 mmmL lslslslR
)())...(,)(,( 2211 mm SlslslRs
),(),)...(,)(,( 2211& mmmmsL SlslslslR
SMAR prediction
Because the number of generated rules might be huge, we create a corresponding hashing tree to accelerate the access.
LHS決定 hash value RHS is calculated by
multiplying support and confidence
root
…
LHS1LHS2
RHS
SMAR prediction
SMAR-N-gram Ex1: a historical behavior is <(a,1)(b,2)(c,5) >
set n = 2, the last two pair location-services pair plus current location
now at location d, <(b,2)(c,5)(d)> as LHS
Ex2:a historical behavior is <(a,1)(b,2)(c,5) >
set n = 2, the last two pair location-services pair
<(b,2)(c,5)> as LHS
<(b,2)(c,5)(d)>
205
<(b,2)(c,5)> (e,20)(d,5)
Experimental evaluation
Probability of backward movement, Pb = 0.1 Probability of next node movement: Pn = 0.2 Probability of staying in the same node: Ps = 0.3
Experimental evaluation
Experimental evaluation
Experimental evaluation
Experimental evaluation
Conclusions & associated thinking
The proposed data mining method, namely SMAP-Mine
One physical scan on the database is needed
The prediction function : SMAR-N-gram, which is based on the N-gram model
Conclusions & associated thinking
Mining and predicting behaviors of driver for:Drunk driving Car racingetc…