A Predictive Model for the Temporal Dynamics of Information - PowerPoint PPT Presentation

International Workshop on Mining Social Network Dynamics Held in conjunction with the 21st International World Wide Web Conference - Lyon, France - 16th April, 2012 A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks Adrien Guille*, Hakim Hacid** *ERIC Lab, Université Lumière Lyon 2 **Alcatel-Lucent Bell Labs France

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Online social networks are very powerful tools for the spread of information ◦ 2010 Arab spring (Howard et al., 2011) ◦ 2008 U.S. presidential elections (Hughes et al., 2009) ◦ etc. • The case of Twitter ◦ Explicit network ◦ Following & mentioning ties Context 2 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • A model capable of predicting the temporal dynamics Volume of tweets Time • Input: a social network, an information topic Output: the time-serie of the volume of tweets generated by the diffusion Purpose of our proposal 3 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Predictive models for diffusion on Twitter basis network prediction nature object Bakshy et al., passive links topological URL 2011 Yang & Counts, active links topological topic 2010 passive & Galuba et al., topological URL 2010 active links Yang & Leskovec, non-graphical temporal hashtag 2010 passive & temporal topic Our approach active links State-of-the-art 4 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Raw data ◦ Yang & Leskovec: 4,76.10 8 tweets ◦ Kwak et al.: followers graph (1,47.10 9 edges) • Preprocessing ◦ Extraction of the active ties ◦ Identification of sub-networks • Identification of cascades Data 5 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • AsIC principle (Saito et al., 2010) ◦ A network: the graph of followers ◦ Diffusion probabilities & time-delays • Asynchronous Independent Cascades Basis of the proposed model 6 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Assumption ◦ The dynamics of the spreading process at the macroscopic level is explained by interactions at a microscopic level between pairs of users and the topology of their interconnections. • Dimensions social semantic temporal user I H dTR hM mR hK A u 1 0,78 0,7 0 0,65 1 0,125 0,12 0,12 u 2 0,23 0,23 1 0,10 1 0,33 ... = 13 attributes for each set (u 1 ,u 2 ,topic,time) Our approach 7 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Supervised classification task Binary class: diffusion/non-diffusion ◦ Bayesian Logistic Regression (BLR) • Results ◦ Social dimension only precision rate of 79% ◦ Social + temporal + semantic dimensions gain of 7% • Global precision: 85% Estimating diffusion probabilities 8 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Input ◦ a network of users U = {u 1 ,u 2 ,...,u n } user I dTR mR ◦ an information i = {k 1 ,...,k p } u 1 0,78 0,45 0,21 ... u 2 0,23 0,89 0,09 ◦ a subset of initialy informed users u 3 0,98 0,19 0,85 ... ... • extension of AsIC ◦ f u1,u2 (i,t) based on BLR ◦ r u1,u2 = (1-I(u 2 )).10 • t simulates the course of each day Building a prediction engine 9 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Visualization Information about the release date of the new iPhone Time-serie: predicted real 30 70 22,5 52,5 Volume of tweets 15 35 17,5 7,5 0 0 1 3 5 7 9 1 3 5 7 Time (days) Time (days) E1 E2 Evaluation on a prediction task 10 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Gain of our approach over the 1-time lag predictor (Yang & Leskovec, 2010) • Two aspects: ◦ Relative error on predicting volume ◦ Relative error on dynamics E1 E2 E3 E4 ALL Gain over dynamics 25,19 % 39,23 % 29,21 % 3,22 % 24,21 % Gain over volume 42,89 % 47,70 % 34,49 % 40,07 % 41,29 % Overall gain 34,04 % 43,46 % 31,85 % 21,65 % 32,75 % Quantitative results 11 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Initial assumption confirmed by the experiments i.e. predicting diffusion from local properties • Observation of a particular pattern • Improve volume estimation 80 Time-series: 60 Volume of tweets adjusted prediction (c=1,6) 40 predicted real 20 0 1 3 5 7 9 Time (days) Conclusion 12 / 12

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • P. N. Howard, A. Duffy, D. Freelon, M. Hussain, W. Marai, and M. Mazaid. Opening closed regimes, what was the role of social media during the arab spring?, 2011. • A. Hughes and L. Palen. Twitter adoption and use in mass convergence and emergency events. International Journal of Emergency Management, 6(3), 2009. • E. Bakshy, J. M. Hofman, W. A. Mason, and D. J. Watts. Everyone’s an influencer: quantifying influence on twitter. WSDM’11, 2011. • J. Yang and S. Counts. Predicting the speed, scale, and range of information diffusion in twitter. ICWSM’10, 2010. • J. Yang and J. Leskovec. Modeling information diffusion in implicit networks. ICDM’10, 2010. • W. Galuba, K. Aberer, D. Chakraborty, Z. Despotovic, and W. Kellerer. Outtweeting the twitterers - predicting information cascades in microblogs. In Proceedings of the 3rd conference on Online social networks, WOSN’10, 2010. • K. Saito, M. Kimura, K. Ohara, and H. Motoda. Selecting information diffusion models over social networks for behavioral analysis. PKDD’10, 2010. • H. Kwak, C. Lee, H. Park, and S. Moon. What is Twitter, a social network or a news media?, WWW’10, 2010. References

A. GUILLE and H. HACID - A Predictive Model for the Temporal Dynamics of Information Diffusion in Online Social Networks - MSND/WWW’12 • Visualization Information about the launch of Googlewave Time-serie: predicted real 40 40 30 30 Volume of tweets 20 20 10 10 0 0 1 3 5 7 9 1 3 5 7 9 Time (days) Time (days) First network Second network Annexes