Group Formation in eLearning-enabled Online Social Networks Steffen Brauer, Thomas C. Schmidt steffen.brauer@haw-hamburg.de, t.schmidt@ieee.org iNET RG, Department of Computer Science Hamburg University of Applied Sciences September 26, 2012
Group Formation Outline 1 Motivation 2 eLearning-enabled OSN 3 Group Formation Approach 4 Evaluation 5 Conclusion Steffen Brauer HAW Hamburg 2
Group Formation Motivation Motivation Classic eLearning environments Intra-group communication in Online social networks (OSN) predefined classrooms Socialize with friends Managed by instructor Groups are user-triggered Creates groups Ubiquitous use Analyses course results Tracks learning progress How to provide a platform for self-paced learning on topics of personal interest? jdflksajfdlkjsalkfjlksaä löksaölfdköskdaf ölsaökdsaf ölksölfdkösa Steffen Brauer HAW Hamburg 3
Group Formation Motivation Motivation Objectives & Challenges Our work focuses on integrating an OSN and an eLearning environment by removing the instructor Removal of instructor leads to challenges 1 How to stimulate a team building process that is effective for learners? 2 How to provide access to the relevant content for a learning group? 3 How to facilitate a consistent learning progress, include feedback and corrective actions? Steffen Brauer HAW Hamburg 4
Group Formation Motivation Motivation Objectives & Challenges Our work focuses on integrating an OSN and an eLearning environment by removing the instructor Removal of instructor leads to challenges 1 How to stimulate a team building process that is effective for learners? 2 How to provide access to the relevant content for a learning group? 3 How to facilitate a consistent learning progress, include feedback and corrective actions? Steffen Brauer HAW Hamburg 4
Group Formation eLearning-enabled OSN eLearning-enabled OSN Base Structure Extend commercial OSN by adding learning related features Communication is handled by commercial OSN via APIs All relevant objects are represented in the OSN Classical representation of an OSN user1 user2 user3 Steffen Brauer HAW Hamburg 5
Group Formation eLearning-enabled OSN eLearning-enabled OSN Base Structure Extend commercial OSN by adding learning related features Communication is handled by commercial OSN via APIs All relevant objects are represented in the OSN Representation using the unified approach user1 content1 member related to group1 topic1 friends edits studies member edits user2 user3 Steffen Brauer HAW Hamburg 6
Group Formation eLearning-enabled OSN eLearning-enabled OSN User Model Availability Motivation of an user to start collaboration Steffen Brauer HAW Hamburg 7
Group Formation eLearning-enabled OSN eLearning-enabled OSN User Model Availability Motivation of an user to start collaboration Learning style (Felder & Silverman Theory) Active or Reflective (Processing) Visual or Verbal (Input) Sensing or Intuitive (Perception) Sequential or Global (Understanding) Steffen Brauer HAW Hamburg 7
Group Formation eLearning-enabled OSN eLearning-enabled OSN User Model Availability Motivation of an user to start collaboration Learning style (Felder & Silverman Theory) Active or Reflective (Processing) Visual or Verbal (Input) Sensing or Intuitive (Perception) Sequential or Global (Understanding) Knowledge Represented by tags Each topic defines required tags with weights Users also hold tags with an activity index Knowledge Rank is calculated by product of weights and activity index Steffen Brauer HAW Hamburg 7
Group Formation Group Formation Approach Group Formation Overview 1 User initiate group building by selecting a topic, which requires collaboration Steffen Brauer HAW Hamburg 8
Group Formation Group Formation Approach Group Formation Overview 1 User initiate group building by selecting a topic, which requires collaboration 2 Starting at the initiator, the social network is searched for candidates Steffen Brauer HAW Hamburg 8
Group Formation Group Formation Approach Group Formation Overview 1 User initiate group building by selecting a topic, which requires collaboration 2 Starting at the initiator, the social network is searched for candidates 3 If a number of candidates is found, the group formation tries to find the best constellation Steffen Brauer HAW Hamburg 8
Group Formation Group Formation Approach Group Formation Overview 1 User initiate group building by selecting a topic, which requires collaboration 2 Starting at the initiator, the social network is searched for candidates 3 If a number of candidates is found, the group formation tries to find the best constellation 4 Selected users are invited and learning experience starts Steffen Brauer HAW Hamburg 8
Group Formation Group Formation Approach Candidate Selection Input : social network, number of candidates, threshold Vertex is added to candidate set, if distance to initiator and topic is lower than threshold Distance formula includes learning style and knowledge rank (scale: 0 - 1) Implemented search algorithms: Breath First Search(BFS) Random Walk Search(RWS) Best Connected Search(BCS) Output : candidate set Steffen Brauer HAW Hamburg 9
Group Formation Group Formation Approach Group Formation Input : candidate set Group fitness defined by: common learning style high knowledge rank low distance in social network Implemented by genetic algorithms to reduce complexity Group constellations are treated as chromosomes in a population In each generation cross-over and mutation operations are performed Only constellations with a high fitness are selected for next generation Output : best group constellations Steffen Brauer HAW Hamburg 10
Group Formation Evaluation Evaluation Open questions 1 How are the user attributes distributed? 2 What is the impact of search algorithms? 3 Does the threshold influence the search complexity? 4 Does the candidate count influence the group fitness? Steffen Brauer HAW Hamburg 11
Group Formation Evaluation Evaluation Generating test data No implementation exists and no appropriate test data Evaluation on synthetic data Simplification: Only user objects in the social network and all users are available Forest fire model was used to generate a social network with 1000 vertices and 31522 edges Challenge: How to distribute the user attributes? Learning style: empirical data from Felder & Spurlin Knowledge: 20 tags are power-law distributed over all vertices with random activity index Steffen Brauer HAW Hamburg 12
Group Formation Evaluation Evaluation User Model How are the user attributes distributed? Knowledge rank Distance in learning style 0.06 0.04 Frequency 0.20 Frequency 0.02 0.10 0.00 0.00 0 0.13 0.3 0.45 0.6 0.75 0.9 0 0.25 0.5 0.75 1 Knowledge Rank Distance 0 = 0.27 Normal distribution Very low average knowledge Low average distance rank Steffen Brauer HAW Hamburg 13
Group Formation Evaluation Evaluation Candidate Selection What is the impact of the search algorithms? Group Distance Learning Style 0.8 Group Knowledge Rank 0.30 No significant 0.6 differences in distance 0.4 0.15 of learning style and 0.2 knowledge rank 0.00 0.0 BFS RWS BCS BFS RWS BCS 0.30 BFS and BCS produce Group Density 0.20 nearly equal results 0.10 RWS produce low 0.00 group density BFS RWS BCS Steffen Brauer HAW Hamburg 14
Group Formation Evaluation Evaluation Candidate Selection Does the threshold influence the search complexity? 80 Breath First Random Walk Best Connected Visited Vertices 60 RWS performs best if threshold < 0.7 40 BFS and BCS convert at 0.9 20 0 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 Threshold Steffen Brauer HAW Hamburg 15
Group Formation Evaluation Evaluation Group Formation Does the candidate count influence the group fitness? 3.0 BFS was used to find Group fitness candidates 2.0 Threshold = 0.8 No significant change 1.0 in group fitness by increasing candidate 0.0 count 10 15 20 25 30 35 40 Candidate count Steffen Brauer HAW Hamburg 16
Group Formation Conclusion Conclusion & Outlook Problem : How to simulate a team building process that is effective for learners? User model includes availability, learning style and knowledge Approach divided in two parts: Candidate selection Group formation Evaluation based on synthetic data Future research Improve data base by empirical data Include tie strength to take full advantage of unified approach Steffen Brauer HAW Hamburg 17
Recommend
More recommend
