Universitatea Politehnica Bucureti Facultatea de Automatic i Calculatoare
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Towards a Peer-to-Peer Recommender System Based on Collaborative Filtering Techniques Authors: Sâmbotin Ana-Delia, Mugurel Andreica E-mail: ana. sambotin@cti. pub. ro, mugurel. andreica@cs. pub. ro 17. 12. 2021 1
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Outline • Introduction • Problems • Design • Architecture • Conclusion 17. 12. 2021 2
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Introduction Peer-to-peer Recommender system • One-to-one communication model • Automate the recommendation of a song when large sets of data are involved • Each node is at the same time both supplier and consumer • File sharing, resource sharing • Content based approach – uses the product’s description • Collaborative filtering – uses user's social environment 17. 12. 2021 3
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Problems • Problems – There isn’t any recommender system that is based on an infrastructure specific to its needs – Usually, these applications force the users to express their preferences • Application purpose – To improve the main properties of a recommender system(the speed of a file transfer, the stability of the network) – To indicate which node is the closest one with similar preferences 17. 12. 2021 4
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Design • Used properties: – User profile analyzer – the files shared by a user – Network manager – round time trip between 2 nodes – Recommendation system – the search queries 17. 12. 2021 5
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Design • Different strategies will be adopted in dynamic way: – Improving the stability of the network – Improving the file transfer – Improving the semantic distance – Improving the number of messages through the network 17. 12. 2021 6
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Architecture • Main modules – Network • Stability • Localization • Similarity – Profile analyzer • Recommender system – File search 17. 12. 2021 7
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Architecture • Two main roles: Bootstrap and normal node • The nodes will be first helped by the bootstrap and after organizing themselves • Different types of messages depending on the adopted strategy 17. 12. 2021 8
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Strategies • A network of “supernodes”, that hide a group of peers • A “supernode” can be consider to be a “proxy” node • Distance strategy– relative geographical coordinates and the distance between peers • File searching strategy – peer’s interests and similarity coefficient 17. 12. 2021 9
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Architecture - formulas • Distance metric • Similarity coefficient 17. 12. 2021 10
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare Conclusions • Future work: – Validate the proposed model – Test with large sets of data – Optimize the metrics • An infrastructure for an recommender system which: – Should indicate who is the closest peer with similar preferences – Should improve the speed transfer – Should reduce the number of messages exchanged through the network – Should improves the search query time 17. 12. 2021 11
Universitatea Politehnica Bucureşti - Facultatea de Automatică şi Calculatoare The end • Thank you! • Questions? 17. 12. 2021 12
- Slides: 12