Content Recommendation Through Linked Data
2 likes271 views
This document summarizes Iacopo Vagliano's PhD thesis on using linked data for content recommendation. The thesis proposes a linked data-based recommendation framework called Allied, a recommendation algorithm called ReDyAl that generates recommendations by traversing and hierarchical generation over linked data relationships, and an ontology called RSCtx for representing user context to enable context-aware recommendations. It also discusses visualizing linked data-based recommendations and concludes with contributions, limitations, and perspectives for future work.
Content Recommendation Through Linked Data
- 1. Content Recommendation Through Linked Data 1 Iacopo VAGLIANO Supervisor: Prof. Maurizio MORISIO Ph.D. Thesis Defense – Torino, 5 May 2017
- 2. 2 Information Overload 3 million likes, 240 thousand new photos, 3 million posts shared 56 thousand pictures 430 thousand tweets 300 hours of new videos 120 new accounts http://www.go-globe.com/blog/60-seconds/
- 3. 3 Recommender Systems “Which movie can I watch?”
- 4. 4 Linked Data Enormous repository of multi-domain knowledge Standard access Semantic relationships between structured and interlinked data
- 5. 5 Goals Recommender System Linked Data User Context Visualization Contextual preferences Exploitation of Linked Data relationships User and context representation Effective visualization
- 6. 6 Outline A Linked Data based Recommendation Framework A Linked Data based Recommendation Algorithm Ontologies for Context-Aware Recommendations Visualizing Linked Data based Recommendations Conclusions and Perspectives
- 7. 7 Outline A Linked Data based Recommendation Framework A Linked Data based Recommendation Algorithm Ontologies for Context-Aware Recommendations Visualizing Linked Data based Recommendations Conclusions and Perspectives
- 8. dct:subject Valentino Park Valentino Park Visitor attractions in Turin Visitor attractions in Turin Museums in Turin Museums in Turin National Museum of Cinema National Museum of Cinema 8 Linked Data Relationships Egyptian Museum Egyptian Museum Mole Antonelliana Mole Antonelliana TurinTurin dbo:location-city skos:broader dbo:location-city dbo:location Category-Category Resource-Resource Resource-Category
- 9. Allied: a Linked Data RS Framework Generation Ranking Classification Hierarchical Transversal Dynamic Hierarchical Transversal Hybrid Dynamic Category Tree KB 9 Categorizer Tree Graph Mobile REST Interface ReDyAl Resources
- 10. 10 Outline A Linked Data based Recommendation Framework A Linked Data based Recommendation Algorithm Ontologies for Context-Aware Recommendations Visualizing Linked Data based Recommendations Conclusions and Perspectives
- 11. 11 Linked Data-only Recommendations Algorithm Cross- dataset Cross- domain Transversal links Hierarchical links dbrec HyProximity ReDyAl
- 14. 14 Traversal Generation National Museum of Cinema National Museum of Cinema Mole Antonelliana Mole Antonelliana TurinTurin dbo:location-city dbo:location
- 16. 16 Hierarchical Generation Museums in Turin Museums in Turin National Museum of Cinema National Museum of Cinema Egyptian Museum Egyptian Museum
- 17. 17 Hierarchical Generation dct:subject Valentino Park Valentino Park Visitor attractions in Turin Visitor attractions in Turin Museums in Turin Museums in Turin National Museum of Cinema National Museum of Cinema Egyptian Museum Egyptian Museum Mole Antonelliana Mole Antonelliana skos:broader
- 18. 18 ReDyAl Ranking Quotient of a weighting function and a distance (maxDistance) Links between 2 resources (incoming/outgoing, direct/indirect) Parameters to privilege transversal or hierarchical ranking
- 20. 20 Outline A Linked Data based Recommendation Framework A Linked Data based Recommendation Algorithm Ontologies for Context-Aware Recommendations Visualizing Linked Data based Recommendations Conclusions and Perspectives
- 21. 21 Context-Aware Recommendations “Which movie can I watch?” Saturday Cinema Partner Tuesday Home Friends Which context?
- 22. 22 The RSCtx Ontology Address contextual data sparsity General model of contextual information that could be reused in multiple domains and applications Represent context parameters on different levels of granularity Reuse existing ontologies – Extension of PRISSMA
- 23. 23 Context Ontologies Domain independence Coverage Variable context granularity Extensibility Core ontology approach Lightweight ontology Reuse of existing terms Availability on the web SOUPA CoOL CONON CoDaMoS Koripää Hervás DCO RSCtx PRISSMA
- 24. 24 Time Representation in RSCtx
- 25. 25 Context-Aware Approach with RSCtx Generalize context Find preferences Enough Run recommendation algorithm N RSCtx COUP
- 30. 30 Evaluation of Our Approach Algorithm MAE with RSCtx MAE without RSCtx Random Guess 0.2315 2.0998 User Average 0.2312 0.3026 Item kNN 0.2312 0.3976 SVD++ 0.2514 0.3511 Time SVD++ NA 0.2693
- 31. 31 Outline A Linked Data based Recommendation Framework A Linked Data based Recommendation Algorithm Ontologies for Context-Aware Recommendations Visualizing Linked Data based Recommendations Conclusions and Perspectives
- 32. 32 Linked Data Visualization Tools Work Lay users Mobile Cross- domain Cross- dataset DBpedia mobile DBpedia viewer Pubby Payola DBpedia ME
- 33. 33 Use cases Browse whole DBpedia Browse a specific domain in DBpedia User Build applications Developer Only movies Only POIs Recommend movies or tourist attractions
- 34. 34 Architecture Code generation module SPARQL / RDF End user application
- 35. 35 Approach View categories of a resource SELECT View a resource Browse a category SELECT DESCRIBE
- 36. 36
- 37. 37
- 38. 38
- 39. 39
- 40. 40 Outline A Linked Data based Recommendation Framework A Linked Data based Recommendation Algorithm Ontologies for Context-Aware Recommendations Visualizing Linked Data based Recommendations Conclusions and Perspectives
- 41. 41 Conclusions Recommender System Linked Data User Context Visualization Contextual preferences Exploitation of Linked Data relationships User and context representation Effective visualization
- 42. 42 Main PhD Thesis Contributions A systematic literature review of Linked Data based RS A new recommendation algorithm Context representation for recommendation algorithms Participation in development of a framework to execute Linked Data based recommendation algorithms Application of results in several use cases of Telecom Italia
- 43. 43 Limitations and Open Issues ● Test in multiple domains ● Test using other datasets ● More resource in input ● Context-awareness ● UI Evaluation ● Test in multiple domains ● Test using other datasets ● Test in multiple domains ● Test with other context dimensions A Linked Data based Recommendation Framework A Linked Data based Recommendation Algorithm Ontologies for Context-Aware Recommendations Visualizing Linked Data based Recommendations
- 44. 44 Perspectives Diversity of recommendations Mining micro-blogs and user reviews Explaining recommendations Serendipity Exploratory search
- 45. 45 Thank You Karpus, A. Vagliano, I. Goczyła, K. Morisio, M. An Ontology-based Contextual Pre-filtering Technique for Recommender Systems. FedCSIS 2016. Vagliano, I. Figueroa, C. Rodríguez Rocha, O. Torchiano, M. Faron-Zucker, C. Morisio, M. ReDyAl: A Dynamic Recommendation Algorithm based on Linked Data. In: CBRecSys, ACM RecSys 2016. Vagliano, I. Marengo, M. Morisio, M. DBpedia Mobile Explorer. In: IEEE RTSI 2015. Figueroa, C. Vagliano, I. Rodríguez Rocha, O. Torchiano, M. Faron-Zucker, C. Corrales, J. C. Morisio, M. Allied: A Framework for Executing Linked Data- based Recommendation Algorithms. IJSWIS (In Press). RSCtx softeng.polito.it/rsctx ReDyAl natasha.polito.it/LDRecommenderWeb Allied natasha.polito.it/AllliedWI http://softeng.polito.it/vagliano/