051202 Cooper WP3

Knowledge sharing & recommendation services: 

Knowledge sharing andamp; recommendation services

Goal of WP3: 

Goal of WP3 ‘The main task is to investigate, design and implement social recommendation and knowledge sharing mechanisms suitable for ... project-centred scenarios’ These [collaborative recommendation] approaches are complemented by latent semantic analysis, ... based on the content of repositories

Knowledge sharing of peers through ad hoc transient communities: 

Knowledge sharing of peers through ad hoc transient communities Example of lsa based recommendation Current project in OUNL Focus on peer support is example of recommendation and knowledge sharing helps solve teacher bandwidth problem increases competencies of both tutor and tutee

Essence of the approach: 

Essence of the approach analyse ‘tutee’ question using latent semantic analysis find and select ‘suitable’ peer-tutors suitable means competent, available, eligible set up wiki and create ad-hoc, transient community of tutee and a number of tutors around it seed wiki with proto-answers, discovered by lsa log results in for instance FAQ, user portfolio, community database, for future use

Slide5: 

Slide6: 

Slide7: 

Slide8: 

The positioning problem: 

The positioning problem Imagine some student, not necessarily involved in a formal programme, with a learning target (s)he has several prior (certified or not) competencies, (s)he enters a ‘learning network’ How to map the target on the learning network? map his or her prior competencies?

Learner positioning: 

Learner positioning

LSA and positioning: our current goals: 

LSA and positioning: our current goals Develop specs for the use of Latent Semantic Analysis for positioning Specify, develop and test a prototypical ‘positioner’ With respect to their validity and reliability, compare human and LSA-based practices of honoring prior competencies

What is LSA?: 

What is LSA? element in user portfolio assetA assetB assetC

How does it work?: 

How does it work? documents terms

How does it work?: 

How does it work? start with term X document matrix many zeros, low average cell freq -andgt; sparse matrix use stop rules use stemming? apply other ‘rules of thumb’ use vector representation per document through SVD (cf. factor analysis)

Compare and contrastan additional goal: 

Compare and contrast an additional goal explicit semantics (metadata approach), requires effort upfront, during development of tasks and portfolios; requires ontology with inference rules; improves over time (semantic web) latent semantics, works during runtime, should be fast enough, no explicit semantics, hence no improvement over time (GTP)

Goal of WP3: 

Goal of WP3 ‘The main task is to investigate, design and implement social recommendation and knowledge sharing mechanisms suitable for ... project-centred scenarios’ These [collaborative recommendation] approaches are complemented by latent semantic analysis, ... based on the content of repositories