– Monograph summarizing research results on modeling, optimization, and learning of information and influence diffusions: “Information and Influence Propagation in Social Networks, Morgan and Claypool, 2013“.
– Multi-round influence maximization [KDD’18]: We study the multi-round influence model where propagation from potentially different seeds occurs independently in different rounds, with the goal to collectively active nodes in social networks. This corresponds to multiple stages in a marketing campaign. We develop scalable algorithm for both non-adaptive and adaptive settings.
– Influence-based centrality [WWW’17]: We propose the study of influence-based network centrality and provide a comparative study on two centrality measures: single node influence (SNI) centrality and Shapley centrality. We provide axiomatic characterizations for both centralities as well as scalable centrality computation algorithms based on the reverse influence sampling approach.
– Robust influence maximization [KDD’16, arXiv:1601.06551]: We address the issue of robust optimization when parameters for the influence maximization are not accurate. We propose a new algorithm to achieve robust influence maximization with a guarantee, and investigate sampling methods to effectively improve parameter accuracy for the robust influence maximization task.
– Influence diffusion model covering the spectrum from competition to complementarity [VLDB’16, arXiv:1507.00317]: We propose an extension to the classical independent cascade model to cover the entire spectrum from competition to complementarity for two propagating items in a social network. We study the properties of the model and two variants of optimization problems for the mutually complementary regime. The paper also proposes a general sandwich approximation and a general condition for applying reverse reachable set approach to any diffusion model.
– Amphibious influence maximization [EC’15, arXiv:1507.03328]: we propose the model of combining traditional media marketing with viral marketing by selecting seed content providers to initiate cascades and selecting seed consumers to propagate the cascades in the social network. We prove hard-to-approximate result of the problem for general graphs, and propose an approximation algorithm for a certain class of restricted bipartite graph between content providers and consumers.
– Seed minimization with probabilistic coverage guarantee [KDD’14, arXiv: 1402.5516]: We look into the optimization problem of minimizing the seed set to ensure a probabilistic influence coverage guarantee. The main difficulty is that the objective function is not submodular. We provide theoretical analysis on the approximation guarantee of this problem and experimental study on the effectiveness of our algorithm.
– Active friending in social networks [KDD’13, arXiv: 1302.7025]: Active friending provides a new perspective on using social influence in social networks. Instead of finding seed nodes to maximize influence, it needs to find influence pathways to maximize influence. We provide efficient algorithms for finding intermediate nodes so as to increase the probability of a target node accepting the friending request.
– Scalable influence maximization: We study new algorithms and heuristics to improve the speed of finding influential nodes in a social network for viral marketing.
– Influence diffusion modeling and maximization for complex social interactions: We model complex social interactions including negative opinions due to product defects, competitive influence diffusion, and friend/foe relationships, and study influence maximization problem in these contexts.
- IC-N model and MIA-N algorithm [SDM’11, MSR-TR-2010-137]: We extend the independent cascade model to incorporate the emergence and propagation of negative opinions due to product defects, and study the influence maximization problem in this new model.
- Competitive influence diffusion CLT model and influence blocking maximization algorithm CLDAG in [SDM’12, arXiv:1110.4723]: We study the competitive linear threshold model, and propose efficient heuristic algorithm CLDAG for selecting the positive seed set that most effectively block the diffusion of negative influence.
- Time-critical influence maximization with time-delayed diffusion process [AAAI’12, arXiv:1204.3074]: We study influence maximization in which diffusion on each step may be delayed, and the objective is to maximize influence spread within a certain deadline. Both IC and LT models are extended, and efficient algorithms are proposed and evaluated.
- Influence diffusion dynamics and maximization in networks with friend and foe relationships [WSDM’13, arXiv:1111.4729]: We extend the voter model to signed networks representing both friend and foe relationships, and study the influence diffusion dynamics and influence maximization problem in this context.
– Participation maximization based on influence in online discussion forums [ICWSM’11, MSR-TR-2010-142]: We propose and study the participation maximization problem in online discussion forums, in which forum participation is determined by influence propagation through forum users.