汪秉宏-双向选择模型应用于相亲大会.pptx

人类行为动力学及其应用研究进展双向选择模型应用于相亲大会配对成功率的预测汪秉宏2015-4-20 上海大学 第11届网络科学论坛bhwang@Empirical data of e-mail communication3188 users exchange e-mails to each other for 3 months (with time resolution: 1 second). [A.-L. Barabási, Nature 435, 207 - 211 (2005) ]: time interval between two consecutive emails sent by a userP(t): The probability distribution of the time intervalPower-law exponent = 1The response time distribution obeys the same power law[A.-L. Barabási, Nature 435, 207 - 211 (2005) ]the time taken by the user to reply to a received message.P(t): The distribution. of the time intervalPower-law exponent = 1The correspondence patterns of Darwin and EinsteinIn their life:Darwin sent out at least 7591 letters and received 6530 letters； Einstein sent out 14500 letters and received 16200 letters The correspondence patterns of Darwin and Einstein[Oliveira Barabási, Nature 437, 1251 (2005)]Power-law exponent = 1.5What causes the non-Poisson statistics?Which one should I do first? Give a talkAnalytical solution of Barabasi’s modelThe timing of the tasks follows a power-law tailed distribution a = 3/2 for infinite L, a = 1 for finite (and fixed) L.The human dynamics belong to one of the two universality classes with distinguishable power law exponent, a, of the inter-event time distribution: a = 3/2 (surface mail) a = 1 (e-mail, web browse,…). Is it the really case? Vazquez, et al., Phys. Rev. Lett. 95, 248701 (2005); Phys. Rev. E 73, 036127 (2006)Does all the real human behavior belong to one of the only two universality classes?exponent=1.2, Dezso, et al., Phys. Rev. E 73, 066132 (2006) Only one example containing small samples supports the 1.5 universality A huge data-set –Netflix rating system One of the world-largest on-line movie-sharing systemPeople watch movies and vote on themThe data-set involves more than 20,000 movies and more than one million usersThe total number of records (user, movie, rating, time) are more than 1,000,000,000Perfect for empirical studyThe role