The End Of Knowledge - Vault 6/2 - CVPR - 2015 - Social Interaction in Global Networks

graph LR classDef main fill:#f9d4d4, font-weight:bold, font-size:14px classDef concepts fill:#d4f9d4, font-weight:bold, font-size:14px classDef analysis fill:#d4d4f9, font-weight:bold, font-size:14px classDef models fill:#f9f9d4, font-weight:bold, font-size:14px classDef applications fill:#f9d4f9, font-weight:bold, font-size:14px classDef challenges fill:#d4f9f9, font-weight:bold, font-size:14px Main[Social Interaction in
Global Networks] Main --> A[Online Social Systems Concepts] A --> A1[Online systems: library to
crowd metaphor 1] A --> A2[Social systems: hybrid human-design
principles 2] A --> A3[Online world: social, geographic,
network traces 3] A --> A4[Large networks measure human
social phenomena 4] A --> A5[Network neighborhoods: localized large
network view 5] A --> A6[Sociology principles inform social
media design 6] Main --> B[Network Analysis] B --> B1[Billion-node graph visualization challenging,
lacks scales 7] B --> B2[Dense neighborhoods more tractable
than full graph 8] B --> B3[3D cube reveals interesting
neighborhood structure 9] B --> B4[Empty cube parts: math constraints,
social impossibilities 10] B --> B5[Subgraph frequency boundaries: open
graph problems 11] B --> B6[Neighborhoods and complements cover
more space 12] Main --> C[Network Models] C --> C1[Continuous-time model captures triadic
closure 13] C --> C2[Model explains triangle-square scarcity
in networks 14] C --> C3[High-complexity structures in network
neighborhoods 22] C --> C4[Models with high-complexity nodes
match reality 29] Main --> D[Applications and Measures] D --> D1[Important nodes useful for
various applications 15] D --> D2[Embeddedness: standard tie strength
measure, limited 16] D --> D3[Dispersion outperforms embeddedness identifying
partners 17] D --> D4[Dispersion: mutual friend distribution
measure 18] D --> D5[Dispersion identifies partners better
than alternatives 19] D --> D6[Identification accuracy varies by
demographics 20] Main --> E[Challenges and Future Directions] E --> E1[Failed identification predicts relationship
end 21] E --> E2[Latent neighborhood information largely
unutilized 23] E --> E3[Social networks visible, evoking
mixed reactions 24] E --> E4[Understanding requires multidisciplinary collaboration 25] E --> E5[Data stockpiling dangers raise
important questions 26] E --> E6[Online social tie maintenance
unexplored 27] class Main main class A,A1,A2,A3,A4,A5,A6 concepts class B,B1,B2,B3,B4,B5,B6 analysis class C,C1,C2,C3,C4 models class D,D1,D2,D3,D4,D5,D6 applications class E,E1,E2,E3,E4,E5,E6 challenges

Resume:

1.- Online systems evolved from a library metaphor to a crowd metaphor, with social media enabling direct interactions between people.

2.- Online social systems are a hybrid of organic human behavior and designed features, following their own social principles.

3.- The online world has traces of social phenomena, geography, and network structure, with graphs becoming a foundational representation.

4.- Large online social networks provide unprecedented measurement of human social phenomena that were previously difficult to quantify.

5.- Network neighborhoods, the subgraphs induced on a node's neighbors, provide a localized view to reason about large networks.

6.- Sociology principles like homophily, triadic closure, and the small-world phenomenon have informed the design of social media systems.

7.- Visualizing a billion-node Facebook graph is challenging; intermediate scales are lacking due to the small-world phenomenon.

8.- Analyzing the collection of dense network neighborhoods is more tractable than the full billion-node graph.

9.- Plotting Facebook network neighborhoods in a 3D cube based on subgraph frequencies reveals a serpentine curve with interesting structure.

10.- Parts of the 3D cube of network neighborhoods are empty due to mathematical constraints or because certain structures don't arise socially.

11.- Boundaries of the feasible region for subgraph frequencies are difficult to characterize and are related to open problems in graph theory.

12.- Facebook network neighborhoods and their complements together cover much more of the space of possible subgraph frequencies.

13.- Models with explicit triadic closure are challenging to analyze; a continuous-time graph evolution model based on Poisson processes is proposed.

14.- The proposed continuous-time model naturally captures the scarcity of induced squares compared to triangles in real social networks.

15.- Identifying important nodes in a user's network neighborhood is useful for news feed ranking, information sharing, and understanding social ties.

16.- Embeddedness, the number of mutual friends, is a standard measure of tie strength from sociology, but has limitations in practice.

17.- Dispersion, a new measure quantifying how mutual friends are distributed, outperforms embeddedness at identifying spouses/romantic partners from network structure alone.

18.- Dispersion is computed by measuring the graph distances between mutual friends after removing the two endpoint nodes.

19.- On 1.3M Facebook users, dispersion identified spouses/partners in the top spot over 50% of the time, outperforming embeddedness and activity metrics.

20.- Spouse/partner identification accuracy was higher for married couples, males, and relationships reported longer ago, reaching 70% using combined features.

21.- If the algorithm fails to identify a user's partner, there is a 50% higher chance the relationship ends within two months.

22.- Network neighborhoods contain a few high-complexity structures, like nodes linking multiple clusters, not captured by models focused on triangles alone.

23.- There is much latent information in network neighborhoods that we are only beginning to utilize through structural measures and modeling.

24.- Social media has made visible the anatomy of social networks in unprecedented detail, evoking reactions of unease and fascination.

25.- Understanding online social systems requires collaboration between social sciences, computer science, applied math, and machine learning.

26.- The dangers of stockpiling personal data that fuels social media systems raise important questions that technologists should engage with.

27.- Optimal policies for maintaining social ties online, accounting for interaction frequency and human constraints, are relatively unexplored.

28.- Identifying same-sex relationships from network structure is harder due to less data and possible reporting biases on Facebook.

29.- Models with a few high-complexity nodes disrupting a clustered structure could better match real social networks than pure triangle-closing models.

30.- Siblings and relationship partners create similar high-complexity structures in a person's network neighborhood despite arising through very different processes.

Knowledge Vault built byDavid Vivancos 2024