EU FP7 Open Discovery Space (ODS)
Research methodology 1-‐ Conceptual model 2-‐ Offline data study ✓ 3-‐ User study 4-‐ Go online
Offline data study • Classical collaboraHve
filtering based on nearest neighbors, beside a graph-‐based approach
• Three datasets similar to the ODS data plus MovieLens
DATA-‐DRIVEN STUDY: AUGMENTING PREDICTION ACCURACY OF
RECOMMENDATIONS IN SOCIAL LEARNING PLATFORMS
Degree centrality of first 10 central users
0"
50"
100"
150"
200"
250"
u1" u2" u3" u4" u5" u6" u7" u8" u9" u10"
degree%
Top)10%central%users%
MovieLens"
OpenScout"
MACE"
Travel"well"
F1 score of different user-‐based collaboraHve filtering
algorithms
SOUDE FAZELI HENDRIK DRACHSLER PETER SLOEP OPEN UNIVERSITEIT NEDERLAND [email protected]
0"0.01"0.02"0.03"0.04"0.05"0.06"0.07"0.08"0.09"0.1"
3" 5" 7" 10"
F1@10%
size%of%neighborhood%(n)%
MACE%
Tanimoto4Jaccard"(CF1)"
Loglikelihood"(CF2)"
Euclidean"(CF3)"
Graph4based"(CF4)"
0"
0.02"
0.04"
0.06"
0.08"
0.1"
0.12"
0.14"
3" 5" 7" 10"
F1@10%
size%of%neighborhood%(n)%
OpenScout%
Tanimoto3Jaccard"(CF1)"
Loglikelihood"(CF2)"
Euclidean"(CF3)"
Graph3based"(CF4)"
0"
0.02"
0.04"
0.06"
0.08"
0.1"
3" 5" 7" 10"
F1@10%
size%of%neighborhood%(n)%
Travel%well%
Tanimoto3Jaccard"(CF1)"
Loglikelihood"(CF2)"
Euclidean"(CF3)"
Graph3based"(CF4)"0"
0.05"
0.1"
0.15"
0.2"
0.25"
3" 5" 7" 10"
F1@10%
size%of%neighborhood%(n)%
MovieLens%
Tanimoto0Jaccard"(CF1)"
Loglikelihood"(CF2)"
Euclidean"(CF3)"
Graph0based"(CF4)"
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?
Recommended
