personalisation and fuzzy bayesian nets
DESCRIPTION
TRANSCRIPT
RajendraAkerkar
1R. Akerkar
Personalisation - user profile and prototypesp p ypto provide: What is received is what is required
EXAMPLESWeb page retrieval to satisfy customerComputer InterfaceNews Services AdvertisingNews Services, AdvertisingRegulation of messages -SMS. E-MailPersonalised Data Mining
AI Machine Learning, Inferenceand Linguistics provides the intelligent agents for this personalisation service
2R. Akerkar
message
P1 P2 P3 P4 P5 prototypes
s1 s2 s3 s4 s5 supports for interest
Personalised Fusion
to users
send ?Final Support3R. Akerkar
Fuzzy Bayesian Netsy y
Fuzzy Decision TreesCan we construct a theory
Fril programsCan we construct a theoryof prototypes ?
Evidential Logic Rules
Neural netsCan be have a theory offusion ?
Fuzzy Conceptual graphs
Cluster over persons to get clusters for prototypes4R. Akerkar
Data Base
WEB
answer fuzzy decisiontree
search agentuser query
5R. Akerkar
Fril Evidential logic ruleFril Evidential logic rule
x1 is g1 with weight w1x2 is g2 with weight w2l i f IFF
throughfilts g w we g w
. . . .xn is gn with weight wn
class is f IFF filterh
For input {xi = fi}Let i = Pr(gi | fi)
f, gi, fi, h are fuzzy sets
(g | )class is f with probability where = (w11 + w22 + + wnn) = (w11 + w22 + … + wnn)h
6R. Akerkar
Random VariablesOffer Type
RateLoanRate
companysize company
age
Random Variableswith conditionalindependencies
OfferRateSuit
CompS it
MedicalReqSuit
Su t Suit Req
Loan InterestCompany Size : {l di ll}Loan
TimeInterest
Offer : {mortgage, personal loan, car loan, credit card
{large, medium, small}
car insurance, holiday insurance, payment protection,charge card, home insurance}
7R. Akerkar
Offer : {mortgage personal loan car loan credit cardOffer : {mortgage, personal loan, car loan, credit cardcar insurance, holiday insurance, payment protection,charge card, home insurance}Type Rate : {fixed, variable}LoanRate : {good, fair, bad}Company Size : {large, medium, small}p y { g , , }Company Age : {old, middle, young}Loan Period : {long, medium, short}Rate Suitable : {very average little}Rate Suitable : {very, average, little}Company Suitable : {very, average, little}Medical req : {yes, no}Offer Suit : {good, av, bad}Interest : {high, medium, low}
8R. Akerkar
X1 X2 X3
X4X5
We update this priorjoint distribution with
t t iX4 5 respect to any givenvariable instantiations
X6
P (X X ) P (X | P t (X ))n
iPr(X1, ..., Xn ) Pr(Xi | Parents(Xi ))i1 Prior
9R. Akerkar
C0C0 = {Offer*, LoanPeriod*, OfferSuit*}C1 = {OfferSuit,Offer, RateSuit}C2 = {RateSuit*, LoanRate*, Offer, TypeRate*}C3 {CompSuit MedReq * OfferSuit RateSuit
C1C3 = {CompSuit, MedReq,* OfferSuit, RateSuit,
Interest*}C4 = {CompSize*, AgeComp*, CompSuit*}
C2 C3
C4
Compile Clique tree : equivalent to forming prior joint p q q g p jdistribution using an efficient message passing algorithm
10R. Akerkar
Message :4% fixed rate long term mortgages availablefrom 40 year old fairly large Company FRIL
conceptual graph
graph instantiations
Offer = mortgage, Type rate = fixed, loan time = longCompSize = dist, CompAge = dist.
Use message passing algorithm again to update clique distributionswith graph instantiations.
Result will be distribution over interest variable - defuzzify toobtain degree of interest
11R. Akerkar
w1 w2 w3 w4 w5 w6 w7 X replaced with {w }w1 w2 w3 w4 w5 w6 w7 X replaced with {wi}1
We can also
X0
We can alsofuzzify discretesets
X
Mass Assignment Theory provides:{Pr(wi | x)} where x is point value interval or fuzzy set{Pr(wi | x)} where x is point value, interval or fuzzy seti.e we translate value on X to distribution over {wi}
Mass Assignment Theory provides:Mass Assignment Theory provides:Defuzzification: if {i} is distribution over {wi} thenx i i
i where i is expected value of wi 12R. Akerkar
medium
0 1 2 3 4 5 6
small large
0 1 2 3 4 5 6DICE SCORE
Random Set of Voters %
≤ 3 4 5 6x% accept x as large % 0 20 80 100
large = 4 / 0.2 + 5 / 0.8 + 6 / 1
1 2 3 4 5 6 7 8 9 101 2 3 4 5 6 7 8 9 106 6 6 6 6 6 6 6 6 65 5 5 5 5 5 5 5 4 4
MAlarge ={6} : 0.2, {5, 6} : 0.6, {4, 5, 6} : 0.2
4 4
constant threshold : if voter accepts x and y>x then he must accept y13R. Akerkar
weighted dice is small where small = 1 / 1 + 2 / 0 7 + 3 / 0 3
MA = {1} : 0.3, {1, 2} : 0.4, {1, 2, 3} : 0.3
weighted dice is small where small 1 / 1 + 2 / 0.7 + 3 / 0.3prior for dice : 1:0.1, 2:0.1, 3:0.5, 4 :0.1, 5:0.1, 6:0.1
MA {1} : 0.3, {1, 2} : 0.4, {1, 2, 3} : 0.3small
Pr(dice is x | small) = probability that a randomly chosen voter h l f di f b i ld di l i llchooses x as value of dice after being told dice value is small
Least Prejudiced Probability Distribution for dice value =1 : 0 3 + 0 2 + 0 3(1/7) = 0 54291 : 0.3 + 0.2 + 0.3(1/7) = 0.54292 : 0.2 + 0.3(1/7) = 0.24293 : 0.3(5/7) = 0.2143
Pr(x | small)
For continuous fuzzy set we can derive a least prejudiced density function whose expected value can be used for defuzzification
14R. Akerkar
What is Pr(dice is about 2 | dice is small) where( _ | )about 2 = 1 / 0.5 + 2 / 1 + 3 / 0.5
small0.3 0 4 0 3Prior1 : 0.1
about_2 {1} {1, 2} {1, 2, 3}
{2}0 5
0.3 0.4 0.3
01/2(0.2)
0 11/7(0.15)
= 0 0214 2 : 0.13 : 0.54 : 0.1
{2}
{1, 2, 3}
0.5
0.5
= 0.1 = 0.0214
0.15 0.2 0.15
5 : 0.16 : 0.1
{ , , }
Pr(about_2 | small) = 0.6214
Point Semantic Unification used to determine Pr(f | g)Point Semantic Unification used to determine Pr(f | g) where f is fuzzy set, g is point, interval or fuzzy set
15R. Akerkar
output_sm output_me output_laprediction Means of
output_sm, output me
t ll l i f t OUTPUT
output_me, output_laare1, 2, 3mutually exclusive fuzzy sets on OUTPUT
modelinput output {distribution over words} output_sm : 1
output me : 2
1, 2, 3
model output_me : 2output_la : 3
Prediction = 11 + 22 + 33 Defuzzification
16R. Akerkar
V i bl X lVariable X - value xx is point or fuzzy set defined on R
{fi}
Bayes NodeInstantiate :{fi P (fi | )}
Prior Updated{ }{fi : Pr(fi | x)}Distribution Distribution
NET
Classical probability theory does not allowfor this distribution updatefor this distribution update.
17R. Akerkar
P’
UpdatedDistribution
onDistribution Constrainton Xk - D P
oX = (X1…Xn)
Minimum Relative Entropy
Pr’(X) ( )Pr’(X)MINPriorDistribution
onX = (X1…Xn)
Pr’(X) Ln ( )Pr(X)MIN
P'(X)D satisfied
Prior Update Updatevar 1 distribution var 2 distribution
until convergence18R. Akerkar
Bayes Net Compiled to give prior probability distributions.
Update Compiled Net with input variables instantiatedto probability distributions usingto probability distributions usingrelative entropy updating
Bayes Net message passing algorithms forClique tree updating modified to be equivalent to this relative entropy updatingbe equivalent to this relative entropy updating
19R. Akerkar
Fuzzify :A B C
A : {a1, a2, a3}B : [1 10]
Fuzzify :A : same as A aboveB : {small, medium, large}C : {low, av, high}
D
B : [1, 10]C : {1, 2, 3, 4, 5, 6}D : {d1, d2, d3}
D : same as above
small medium large
A B C Da1 x y {d1, d2} 1 10
low = 1 / 1 + 2 / 0.8 + 3 / 0.4av = 2 / 0.2 + 3 / 0.6 + 4 / 1 + 5 / 0.2high = 5 / 0 8 + 6 / 1
= Pr(medium | x) = Pr(large | x) = Pr(low | y) = Pr(av | y)Pr(d1) = 0.5 Pr(d2) = 0.5 high 5 / 0.8 + 6 / 1Pr(small | x) = Pr(high | y) = 0Notex and y can be point values, intervals or fuzzy sets
20R. Akerkar
Red ced Data Base Joint
A B C DReduced Data Base
a1 medium low d1 5
ProbabilityDistribution
a1 medium low d1a1 medium low d2a1 large low d1a1 large low d2
.5
.5
.5
.5a1 medium av d1a1 medium av d2a1 large av d1a1 large av d2
.5
.5
.5
.5a1 large av d2 .5
Repeat for all lines of database
Calculate Pr(D | A, B, C)
21R. Akerkar
1 Search and Scoring based algorithms1. Search and Scoring based algorithms
2. Dependency Analysis algorithms Complexity
A. Node OrderingB Without node ordering
(a) n2
(b) n4
B. Without node ordering
QueryingQueryingMarkov Cover :Parents of query node + children of queryParents of query node + children of querynode + parents of these children
22R. Akerkar
((prototype p1 (offer Mortgage) ) (evlog disjunctive ((AgeOfCompany old_age) 0.1(CompanySize large_size) 0.15(LoanRate small rate) 0.4(LoanRate small_rate) 0.4(TypeRate quite_fixed) 0.1(LoanPeriod long_period) 0.2(MedicalRequired bias_to_yes) 0.05))) : ((1 1) (0 0))
((prototype p1 (offer PersonalLoan) ) (evlog disjunctive ((AgeOfCompany old_age) 0.1(CompanySize large size) 0.15
Rules can alsobe used forfusion( p y g _ )
(LoanRate medium_rate) 0.4(TypeRate quite_variable) 0.1(LoanPeriod short_period) 0.2(M di lR i d bi t ) 0 05))) ((1 1) (0 0))
fusion
(MedicalRequired bias_to_no) 0.05))) : ((1 1) (0 0))
ETC - rules for other offers and other prototypes23R. Akerkar