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  FIRSTNAME.LASTNAME@OU.NL  

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?