Download - Real-life Rating Algorithm
Copyright 2007 Digital Enterprise Research Institute. All rights reserved.
Digital Enterprise Research Institute www.deri.ie
Ranking System based on Social Network of Domain Experts
Mateusz MarmołowskiMay 28, 2008
2 of 20
Digital Enterprise Research Institute www.deri.ie
Motivation
Rating / Ranking / Voting Insufficient rating algorithms Real-life situations, human behaviours
Frienship factor Domain experts
Masters Thesis ;-)
3 of 20
Digital Enterprise Research Institute www.deri.ie
Our goals
New, real-life rating algorithm Fully automated and efficient approach Friendship consideration – social network Expertise collection REST SOA for simple access
4 of 20
Digital Enterprise Research Institute www.deri.ie
Rating Algorithm - agenda
Rating Calculation Friendship influence Domain knowledge Multiple domains User Weight Weighted Rate Bayesian Weighted Rate
Expertise Calculation Weighted Expertise Value Activity factor Final expertise calculation Expertise propagation process
5 of 20
Digital Enterprise Research Institute www.deri.ie
Rating Algorithm
6 of 20
Digital Enterprise Research Institute www.deri.ie
Rating Calculation
7 of 20
Digital Enterprise Research Institute www.deri.ie
RC – Friendship influence
2
),(
),(%200
),( yx
yx
yx
UUDISTANCE
UUFRIENDSHIP
UUFW
Range: [50 – 100%]
8 of 20
Digital Enterprise Research Institute www.deri.ie
RC – Domain knowledge
Domain expertise Parameter α – for amateurs Inaccurate domain adjustment
Generalization and particularization Expertise propagation
9 of 20
Digital Enterprise Research Institute www.deri.ie
RC – Multiple domains
Resource with multiple domains assigned Solution - geometric series
MD factor:
d
q
q
dMD
d
1
1
)( 5.0q
d MD(d)
1 100%
2 75%
3 58%
4 47%
5 39%
q
qaS
n
n
1
11
10 of 20
Digital Enterprise Research Institute www.deri.ie
RC – User Weight
Each rate is being weighted with UW Frienship with the author Expertise in resource’s domains
User Weight for a specific resource:
)(*
)(*),(),(),,( 1
dMDd
dMDDUEXPUUFWRUUUW
d
iix
yxjyx
11 of 20
Digital Enterprise Research Institute www.deri.ie
RC – Weighted Rate
Weighted Rate for resource Rate given by user User Weight The author of The number of rates for a resource
)(
1
)(
1
),,(
),,(*),()(
j
j
RRN
ijji
RRN
ijauthoriji
j
RUUUW
RUUUWRUURRWR
)( jRWR
),( ji RUUR
),,( jauthori RUUUW
authorU
)( jRRN
12 of 20
Digital Enterprise Research Institute www.deri.ie
RC – Bayesian Weighted Rate
votesnumthisvotesnumavg
ratingthisvotesnumthisratingavgvotesnumavgBR
____
_*___*__
votesnumavg __
ratingavg _
votesnumthis __
ratingthis _
- The average number of votes of all resources
- The average rating of all resources’ ratings- The number of votes for this
resource
- The rating of this resource
)(_
)(*)(_*_)(
j
jjj RRNnumavg
RWRRRNvalueneutralnumavgRBWR
Final formula:
13 of 20
Digital Enterprise Research Institute www.deri.ie
Expertise Calculation
14 of 20
Digital Enterprise Research Institute www.deri.ie
EC – Weighted Expertise Value
Separately calculated for each domain and user Based on:
BWR for each resource Number of rates for that resource Mutiple Domain factor
Rating scale normalization
scaleratingdMDRRN
RBWRdMDRRN
DUWEV
j
r
jj
r
jjjj
ky _
%100*
)(*)(
)(*)(*)(
),(
1
1
15 of 20
Digital Enterprise Research Institute www.deri.ie
EC – Final expertise calculation
EXP value is used for User Weight calculation EXP strongly depends on WEV WEV recalculation in case of new rates
Parameter α for amateurs (no published resources)
0),(,),(
0),(,),(
kyky
ky
ky DUWEVforDUWEV
DUWEVforDUEXP
scalerating _
%100
16 of 20
Digital Enterprise Research Institute www.deri.ie
Expertise propagation
17 of 20
Digital Enterprise Research Institute www.deri.ie
Expertise propagation – step 1
18 of 20
Digital Enterprise Research Institute www.deri.ie
Expertise propagation – step 2
19 of 20
Digital Enterprise Research Institute www.deri.ie
Expertise propagation – step 3
20 of 20
Digital Enterprise Research Institute www.deri.ie
Summary
Implementation in progress Integration with MarcOnt Portal Masters Thesis Publication
DERILion Deliverable D1.6.3 - Real-life Rating Algorithm
Thank you.