Social Tagging Networks Structure Dynamics Applications ZiKe ZHANG

Social Tagging Networks: Structure, Dynamics & Applications Zi-Ke ZHANG (张子柯) Collaborators: Chuang LIU (刘闯) , Yi-Cheng ZHANG (张翼成) Tao ZHOU (周涛) Department of Physics, University of Fribourg, Switzerland 2020/9/26 Dep. phys. , Univ. Fribourg 1

Outline n Structure and Dynamics of Social Tagging Networks ¨ What is SNT? ¨ Hypergraph Strutures ¨ Dynamics and emergent properties n Applications in Personalized Recommendation (PR) ¨ Why Recommendation? ¨ How Tags benefit PR? n Conclusions & Discussion 2020/9/26 Dep. phys. , Univ. Fribourg 2

What is social tagging networks? 2020/9/26 Dep. phys. , Univ. Fribourg 3

What is social tagging networks? 2020/9/26 Dep. phys. , Univ. Fribourg 4

From Graph to Hypergraph Unipartite Network Bipartite Network [EPL, 90 (2010) 48006] Hyper Network [from wikipedia. org] 2020/9/26 Dep. phys. , Univ. Fribourg 5

Hypergraph structures in STN Hyperedge: basic unit Hyper Network Zhang and Liu, J. Stat. Mech. (2010) P 10005 2020/9/26 Dep. phys. , Univ. Fribourg 6

Two roles of social tags n Roles ¨ Role 1: an accessorial tool helping users organize resources: Fig. (a) ¨ Role 2: a bridge that connects users and resources: Fig. (b)Zhang and Liu, J. Stat. Mech. (2010) P 10005 2020/9/26 Dep. phys. , Univ. Fribourg 7

Dynamics and evolution of social tagging networks (1/3) n At each time step, a random user can either: n Choose an item(resource), and annotate it with a relevant or random tag with probability p (Role 1) n or choose a tag, and find a relevant or random item with probability 1 -p (Role 2) 2020/9/26 Dep. phys. , Univ. Fribourg 8

Dynamics of social tagging networks (2/3) n Hyper. Degree Distribution: n Clustering Coefficient: 2020/9/26 Dep. phys. , Univ. Fribourg 9

Dynamics of social tagging networks (3/3) n Average Distance: 2020/9/26 Dep. phys. , Univ. Fribourg 10

Applications in Personalized Recommendation (PR) n Why Recommendation? n n Information overload! How to be personalized? ¨ Social influence ¨ Content-based recommendation ¨ Network-based recommendation Zhang et al. Physica A 389 (2010) 179 2020/9/26 Dep. phys. , Univ. Fribourg 11

Method I&II: Tripartite Hybrid(Role 1) n n n Item-user: Item-tag: Linear Hybrid: [Method I, PRE 76 (2007) 046115] [Method II] Zhang et al. Physica A 389 (2010) 179 2020/9/26 Dep. phys. , Univ. Fribourg 12

Method III: Tag-driven(Role 2) n Method III: Zhang et al. Accepted by EPL 2020/9/26 Dep. phys. , Univ. Fribourg 13

Algorithm Performance (1/3) 2020/9/26 Dep. phys. , Univ. Fribourg 14

Algorithm Performance (2/3) 2020/9/26 Dep. phys. , Univ. Fribourg 15

Algorithm Performance (3/3) 2020/9/26 Dep. phys. , Univ. Fribourg 16

Conclusions and Discussion n Conclusions ¨ Structure and Dynamics ¨ The roles of social tags ¨ Application in Personalized recommendation n Discussion ¨ Recommendation with full hypergraph structure ¨ Multi-scale recommendations (semantic-based) ¨ Recommendation with community structures 2020/9/26 Dep. phys. , Univ. Fribourg 17

Thank You! Zi-Ke ZHANG zhangzike@gmail. com 2020/9/26 Dep. phys. , Univ. Fribourg 18