Revealing Mechanisms in Online Learning Networks Moshe Mazuz Prof. Reuven Aviv
Online Learning Networks Online communication. Learning community. Collaboration by responsiveness. Creation of shared knowledge.
Motivation & Objective How do actors choose their response partners in Online Learning community? Discovering which mechanism is most descriptive of the networks. Too many similar models.
Database 500 open university courses networks. Filtering too small networks. Selecting 35 random networks.
Models 1. Directed Random Graph ( RR ). 2. Static preferential response ( PR ). 3. Dynamic preferential response ( DPR ). 4. Small World ( SW ). 5. Dynamic Copying ( DC ).
Methodology Checking existence of responses. Systematic creation of attributes. Training high precision pair-wise classifiers. Robustness checking. Voting. Checking results confidence level.
Results Classifiers have very good precision
Results Very high robustness
Result Voting Results Classifier/ Classifier Votes RRPRDPRDCSW RR2/4-100%-71.43%100%91.43% PR4/4100% DPR3/471.43%-100%100%80% DC0/4-100% SW1/ %-100%-80%85.71%
Results Accuracy as function of confidence level
Results Preferential response Mechanism. Responding to partners with a-priori response attraction power. Attraction power spans over large range. Identification of Key players
Limitation & Future plans Limitation & Future plans Few models Considering response weights.