■
NetSci雑感.
若い人の主役の会議.年上の大御所は参加しているけれど,オブザーバー的な感じで直接発表していない.(Kertesz, Stanley, Mantegna etc.) ニューヨーク州立大の佐山さんによると,世界中に散らばったバラバシの門下生が年に一度集まる同窓会みたいなもん,らしい.
こちらから見ると,普段読んでる論文の著者やエディターがうじゃうじゃいる訳で,結構目移りしながらの聴講.FujiRockとかSummerSonicとかもこんな感じなのかも.行ったこと無いけど.さらに良いことに,すぐ近くにそんな有名人が座っていたり,質問したり,されたり,なんでもできる感じで楽しい.最新の研究とかはちょっと「それで?」みたいなのも多くて,ピンとくる話は少ないが,分野全体の傾向を空気で感じられるのが貴重.
リンクが現れたり消えたりする,Temporal networkが流行中.物理学者が7割,コンピュータ科学者3割くらいな印象.物理学者もオリジンが物理で,今やっているのは計算機科学に近い人が多数で,属性が私に似ている.
データーの貴重さ等もあり,オープンか否かが1つのキーワード.bring home messageをキッチリ言う人が多いのも特徴的.
私自身の発表は成功と言えるが,質疑応答の質がNG.何を言っているかはほぼ,わかる(5/6)のだけれど,答えがone phraseで終わる.他の人はここぞとばかりに,付加情報をしたり,もっと知ってるんだぜ,やってるんだぜアピールをするのに,全然生かせない.この問題は英語というよりも,研究自体に関する問題なのでもっと,詰める研究をしてないと.物理学会での質疑応答もしかり.
あとはよくわからんが,googleのauthor検索みたいなものをみんな発表者に対して行っていた.私もやってみる.
以下はランダムなメモ書き.
1. Havlin social network 1 How created Gallos et al. PRX(2012) Motifs occur significantly 2 How indentify spreaders Kitsak et al. Nature Phys (2020) K-shell better than degree or betweenes 3 Formed? Shao PRL(2009), Qian et al. J. Stat.Phys (2013) Agent's own opinion should be considered 4 extreme opinion Makase et al. arXiv Agents surrounding by less extreme people are more stubborn(頑固) 1 Interaction: 5 basic mechanism change, balance, distant, collective attention, structure hole -> distant most big 2 k-shell decomposition analysis can identify the spreader 3 J. Shao's idea -> majority rule + own opinion = Non concensus state (NCO) 4 extreme shows nonlinear increase with normal NCO model can reproduce 2. Tumminello statistically validated network Bipartise network 1. suspect - crime 2. market members - transactions crimes per suspect power law types per crime power law Bonferroni test -> restrict network happy aging? 4. Adamic Cascade two types of photo Obama Victory vs Million Like Me age, gender, country, political difference spreading is different [exogenous effect exists, TV or web news ? vs exogenous] 5. Pedreschi Orange challenge -> Ivory court mobile phone data 6. Toole Mobile data and employment Do layoff affect social networks? Call data and layoff data 7 Yasseri Revert network 10 Lamboitte Online game network : structural balance myPersonality -> facebook application : psychological mesaurment 5 dimentions OCEAN 30 -1000 facebook friends = typical person (historical study is limited, biased but massive data ) Kertez: online <- -> offline = some similarity 11 Tredan SOUK project Tag + human -> position ----------- 0. Leskovic Collaborate with twitter : follow - follower network very temporal Visited Tokyo for Kitsunegawa on March 15? students 1. Lamboitte M. McPherson, L. Smith-Lawin J.M. Cook (2001) Amm.Rev.Social 27.415 Need for new algorithms, New theoretical questions 2. Masuda SocioPatterns Project burst : CV = stad(tau)/<tau> CV=1 for poisson, CV=0 for periodic Softky & Neurosc (i1993) classic or Goh & Barabasi B Correlated inter event time by Takaguchi et al., PRX -> Model : Queue model by OR = Exponential + Priority queue model by Barabasi =Power-law task has priority i.i.d. random value reproduce power-law inter-event time distribution with exponent 1 but some result shows 1.5. but not every task has priority -> Model Two-state point process idea is well known in neuroscience up/down state + Cascading Process process by Malgren, Amaral et al., PNAS 2008 circadian pattern and self-exciting nature ( Neuroscience seems to be useful...) + Two-state model by Karsai Sci. Rep. (2012) normal <--> excited Temporal walk Synchronized triangles Randomization distribution keep + order destroy (1)Activity-driven model for temporal network Perra et al., Sci. Rep. 2012 -> most stable model for the moment Holme Data-driven vs hypothesis-driven Leskovic badge change behavior? -> around thsrehold, yes 2-dimention model direction = preference ---------- Raj World citation How about population. Talking about Gravity raw, I naturally think that compare mass as population. You put number of citation as mass term. You can compare these two. Is there any threshold, number of authors or countries ? How about Japan cost efficiency? I think Kweit is spend money is for other foregin countries. Budgets can be normalized by DGP or currency rate or publication fee. Global language network Wikipedia, Twitter, Translation -> Japanese is not central but second largest population in twitter. but wikipedia is well connected LINKS 2013 July 22-23 2013 Cambridge ---------- Leskovec Community -> based on weak ties i.e., centrality, betweenness Overlapping communities : confirm based on true data Communities s "tiles" = Simmel(1955), ot Granovetter(1973) Overlapping, Non-overlapping, nested core = most overlapping area Moon snowball sampling: サンプルしたら,次のサンプルの人を紹介してもらう Sano Self-avoiding effect ? by Sayama Why shuffled document. Marton Activity driven model with memory Reinforced network Temporal effect + memory Rumor spreading Daley & Kendall (1964) http://arxiv.org/pdf/1303.5966v1.pdf Hon Contextual analysis framework Burst + Context Context = friends
Barabasi, Watts, Strogatz, Takayasu(の本)が呼び戻してくれた.自由で刺激的な世界.ありがたい.この世界でどうか生き続けられますように.