Erster f vortrag_personalized_rec_sys_for_rbl__20110919_ma_v5.0

Erster_F_Vortrag_Personalized_Rec_Sys_for_RBL__20110919_MA_v5.0.ppt

KOM - Multimedia Communications LabProf. Dr.-Ing. Ralf Steinmetz (Director)

Dept. of Electrical Engineering and Information TechnologyDept. of Computer Science (adjunct Professor)

TUD – Technische Universität Darmstadt Rundeturmstr. 10, D-64283 Darmstadt, Germany

Tel.+49 6151 166150, Fax. +49 6151 166152 www.KOM.tu-darmstadt.de

© 2011 author(s) of these slides including research results from the KOM research network and TU Darmstadt. Otherwise it is specified at the respective slide

httc – Hessian Telemedia Technology

Competence-Center e.V - www.httc.de

Dipl. –Inform. Mojisola Anjorin

[email protected] Tel.+49 6151 166160

12. April 2023

Personalized Recommender Systems for Resource-Based Learning

Ranking Learning Resources in Folksonomies

KOM – Multimedia Communications Lab 2

Resource-Based Learning


Challenge: What is relevant to me right now?


Solution: Ranking of Learning Resources


Overview

1. Basics

2. Application Scenario: CROKODIL

3. Related Work Ranking Algorithms in Folksonomies Recommender Systems in E-learning

4. Research Topic Research Question Objectives Research Approach Current Progress

5. Future Work

6. Summary


Folksonomy Systems

A folksonomy is a system of classification derived from the practice of collaboratively creating and managing tags to annotate and categorize content. [Peters, 2009]

a.k.a Social Tagging Systems, Collaborative Tagging Systems


Folksonomy Model

A folksonomy is a quadruple

F:= (U, T, R, Y)

where

U - Users

T - Tags

R - Resources

Y R × T × U - tag assignment ⊆

[Hotho et al, 2006]


Overview

1. Basics




5. Future Work

6. Summary


Application Scenario: CROKODIL

CROKODIL is a platform offering support for resource-based learning in professional education

Semantic Tag Types Activities Learner Groups and Friendships Recommendations

[Anjorin et al, EC-TEL 2011]


CROKODIL Extends the Folksonomy Model …


Semantic Tag Types


Learner Groups and Friendships


Activities


Overview

1. Basics




5. Future Work

6. Summary


Recommendation Techniques

Collaborative Filtering Recommendation Techniques: Nearest Neighbor (cosine, correlation) Clustering Bayesian Networks Neural Networks Probabilistic Models

Graph-Based Techniques


PageRank Algorithm (Page & Brin, 1998)

Wikipedia Commons. An art draw drawn by Felipe Micaroni Lalli


Ranking Algorithms in Folksonomies


Ranking Algorithms in Folksonomies

Ranking Strategy Applicable For Topic-sensitive(adapts to context)

Group-sensitive

FolkRank Users, Tags, Resources Yes No

GFolkRank Users, Tags, Resources Yes Yes

GFolkRank+ Users, Tags, Resources Yes Yes

GRank Resources Yes Yes

SocialPageRank Resources No No

Personalized SocialPageRank

Resources Yes No

[Abel et al, 2008]


Recommender Systems in E-Learning

Recommender Systems Descriptions

ReMashed[Drachsler et al. 2009]

Recommendations for Web 2.0 content User-based collaborative filtering Informal Learning Networks

RACOFI (Rule-Applying Collaborative Filtering)[Anderson et al. 2003; Lemire 2005]

Recommendations of audio Learning ObjectsRule-based and Collaborative filteringUsing domain taxonomies

RPL recommender[Khribi et al. 2009]

Hybrid recommender system Rated recommendations Learning at work for specific tasks

[Manouselis et al, 2011]


Overview

1. Basics




5. Future Work

6. Summary


Research Question

What semantic information in folksonomies can be exploited to rank learning resources in graph-based recommender systems?

How can these be used to provide personalized recommendations in resource-based learning?


Objectives

1. Investigate ranking algorithms and graph-based recommender techniques for folksonomies

2. Design and implement a personalized graph-based recommender system for resource-based learning

1. Identify semantic information to rank learning resources in the application scenario CROKODIL

2. Integrate relevance feedback to personalize ranking of learning resources

3. Integrate explanations for graph-based recommendations


Research Approach


Current Progress: Conceptual Architecture

[Anjorin et al, ISWC Workshop 2011]


Current Progress: 3a & 3b

3a 3d3c3b


Tag weights are determined based on the usage frequency of semantic tag types Tag Types give additional information about the tag and the tag assignment Assuming usage frequency indicates importance of tag type Therefore tag types indicate the importance of tags

Tag Weights based on Semantic Tag Types

[Böhnstedt, 2011]

[Anjorin et al, DeLFI Workshop 2011]

Tag Type Topic Person Goal Event Genre Location

Usage Frequency

30% 22% 20% 6% 5% 3%


Graph-Based Recommendations using Semantic Tag Types


0.52 - 0.20 = 0.32

0.32 - 0.20 = 0.11

0.35 - 0.20 = 0.15

R1.2.1 Weight: 0.52

R1.1 Weight: 0.35

0.30

0.22

0.05

0.30

0.06

0.20

0.20


Graph-Based Recommendations using Semantic Tag Types

Approach: Traverse the links between activities to find relevant resources (3-hop transitive

associations) [Huang et al, 2004]

Weight resources based on semantic tag types Rank resources according to resource weights propagated along the activity

hierarchy

Aim: To generate recommendations for new users (alleviate the cold-start problem) To alleviate the data sparsity problem



Next Steps

Analyze Ranking Algorithms for Folksonomies

Investigate Semantic Information in the application scenario

CROKODIL

Implementation of Concepts

Evaluation of Concepts


Overview

1. Basics




5. Future Work

6. Summary


Future Work: 3c & 3d

3a 3d3c3b


Future Work: Recommendation Feedback Loop

Recommendation Feedback Loop Rank Resources Explain recommendations Relevance Feedback from Learner Re-Rank Resources

[Harrach and Anjorin, ITiCSE 2011]


Future Work: Evaluations 5 & 6


Overview

1. Basics




5. Future Work

6. Summary


Personalized Recommender Systems for Resource-Based Learning

R1.2.1 Weight: 0.52

R1.1 Weight: 0.35

0.52 - 0.20 = 0.32

0.32 - 0.20 = 0.11

0.35 - 0.20 = 0.15

0.30

0.22

0.05

0.30

0.06

0.20

0.20

Motivation Due to the vast amount of resources

available on the Internet, learners require support in identifying and ranking relevant resources for learning purposes.

Challenge Exploit additional semantic

information in folksonomies to improve graph-based recommendations

Identify semantic information in a resource-based learning scenario like CROKODIL, which could be used to rank learning resources

Contributions Approach using activity hierarchies

and semantic tag types to rank learning resources

Conceptual architecture of a personalized recommender system providing explanations and considering relevance feedback from the learner


Publications

[ARS11] Mojisola Anjorin, Christoph Rensing, Ralf Steinmetz: Towards Ranking in Folksonomies for Personalized Recommender Systems in E-Learning (accepted for publication). October 2011.

[ARB+11] Mojisola Anjorin, Christoph Rensing, Kerstin Bischoff, Christian Bogner, Lasse Lehmann, Anna Lenka Reger, Nils Faltin, Achim Steinacker, Andy Lüdemann, Renato Domínguez García: CROKODIL - a Platform for Collaborative Resource-Based Learning (accepted for publication). September 2011.

[RBP+11] Christoph Rensing, Christian Bogner, Thomas Prescher, Renato Domínguez García, Mojisola Anjorin: Aufgabenprototypen zur Unterstützung der Selbststeuerung im Ressourcen-

basierten Lernen. DeLFI 2011, Sept 2011.

[ABR11] Mojisola Anjorin, Doreen Böhnstedt, Christoph Rensing: Towards Graph-Based Recommendations for Resource-Based Learning using Semantic Tag Types. DeLFI 2011, Sept 2011.

[AaaC11] Mojisola Anjorin, Renato Domínguez García, Christoph Rensing: CROKODIL: a platform supporting the collaborative management of web resources for learning purposes. ITiCSE, ACM, June 2011.

[HA11] Sebastian Harrach, Mojisola Anjorin: Optimizing collaborative learning processes by using recommendation systems. ITiCSE, ACM, June 2011.

[Ren11-2] Christoph Rensing,Stephan Tittel, Mojisola Anjorin: Location based Learning Content Authoring and Content Access in the docendo platform. PerCom-WORKSHOPS 2011, March 2011.


Literature

[ Abel et al, 2008] Fabian Abel, Nicola Henze, and Daniel Krause. Analyzing Ranking Algorithms in Folksonomy Systems. Technical

Report, 2008.

[Böhnstedt, 2011] Doreen Böhnstedt. Phd Thesis, Technische Universität Darmstadt, 2011.

[Huang et al, 2004] Zan Huang, Hsinchun Chen, and Daniel Zeng. Applying Associative Retrieval Techniques to Alleviate the Sparsity Problem in Collaborative Filtering. ACM Transactions of Information Systems, 2004.

[Hotho et al, 2006] Andreas Hotho, Robert Jäschke, Christoph Schmitz, and Gerd Stumme. Information Retrieval in Folksonomies: Search and

Ranking. In ESWC, Lecture Notes in Computer Science, 2006.

[Manouselis et al, 2011] Nikos Manouselis, Hendrik Drachsler, Riina Vuorikari, Hans G. K. Hummel, and Rob Koper. Recommender Systems in

Technology Enhanced Learning. In Recommender Systems Handbook. Springer, 2011.

[Peters, 2010] Isabella Peters. Folksonomies. Indexing and Retrieval in Web 2.0. De Gruyter - Saur, Berlin, 2010.


Any Questions ?

Erster f vortrag_personalized_rec_sys_for_rbl__20110919_ma_v5.0

Technology

Transcript of Erster f vortrag_personalized_rec_sys_for_rbl__20110919_ma_v5.0