The document discusses a data-driven study aimed at enhancing recommendation accuracy in social learning platforms through various collaborative filtering techniques. It evaluates the effectiveness of different algorithms, including classical approaches and graph-based methods, across multiple datasets such as Open Discovery Space (ODS) and MovieLens. Key research questions focus on improving recommendations by analyzing user interactions and employing graph walking algorithms to identify users with similar interests.

1. 250"
200"
150"
MovieLens"
degree%
OpenScout"
100"
MACE"
Travel"well"
50"
0"
u1"
u2"
u3"
u4"
u5"
u6"
u7"
u8"
u9"
u10"
Top)10%central%users%
Degree
centrality
of
first
10
central
users
EU
FP7
Open
Discovery
Space
(ODS)
DATA-­‐DRIVEN
STUDY:
AUGMENTING
PREDICTION
ACCURACY
OF
RECOMMENDATIONS
IN
SOCIAL
LEARNING
PLATFORMS
MACE%
0.1"
Tanimoto4Jaccard"(CF1)"
Loglikelihood"(CF2)"
0.08"
Tanimoto3Jaccard"(CF1)"
0.06"
Loglikelihood"(CF2)"
Euclidean"(CF3)"
0.04"
Graph4based"(CF4)"
0.02"
Euclidean"(CF3)"
Graph3based"(CF4)"
0"
5"
7"
3"
10"
5"
0.2"
0.06"
Tanimoto3Jaccard"(CF1)"
0.04"
Loglikelihood"(CF2)"
Euclidean"(CF3)"
Graph3based"(CF4)"
3"
5"
7"
10"
size%of%neighborhood%(n)%
•
Tanimoto0Jaccard"(CF1)"
0.15"
Loglikelihood"(CF2)"
0.1"
Euclidean"(CF3)"
0.05"
Graph0based"(CF4)"
0"
0"
Classical
collaboraHve
filtering
based
on
nearest
neighbors,
beside
a
graph-­‐
based
approach
Three
datasets
similar
to
the
ODS
data
plus
MovieLens
F1@10%
F1@10%
0.25"
0.02"
•
10"
MovieLens%
Travel%well%
0.1"
0.08"
Offline
data
study
7"
size%of%neighborhood%(n)%
size%of%neighborhood%(n)%
1-­‐
Conceptual
model
2-­‐
Offline
data
study
✓
3-­‐
User
study
4-­‐
Go
online
OPEN
UNIVERSITEIT
NEDERLAND
FIRSTNAME.LASTNAME@OU.NL
0.12"
3"
Research
methodology
SOUDE
FAZELI
HENDRIK
DRACHSLER
PETER
SLOEP
OpenScout%
0.14"
F1@10%
F1@10%
RQ1:
How
to
generate
more
accurate
recommendaHons
by
taking
into
account
user
interacHons
in
social
learning
pla_orms?
RQ2:
Can
the
use
of
the
graph
walking
algorithms
improve
the
process
of
finding
like-­‐
minded
users
within
a
social
learning
pla_orm?
0.1"
0.09"
0.08"
0.07"
0.06"
0.05"
0.04"
0.03"
0.02"
0.01"
0"
3"
5"
7"
10"
size%of%neighborhood%(n)%
F1
score
of
different
user-­‐
based
collaboraHve
filtering
algorithms