No icon

Learning Multiple Factors-Aware Diffusion Models in Social Networks

Learning Multiple Factors-Aware Diffusion Models in Social Networks

Abstract:

Information diffusion is a natural phenomenon occurring in social networks. The adoption behavior of a node toward an information piece in a social network can be affected by different factors, e.g. freshness and hotness. Previously, many diffusion models are proposed to consider one or several fixed factors. In fact, the factors affecting adoption decision of a node are different from one to another and may not be seen before. For a different scenario of diffusion with new factors, previous diffusion models may not model the diffusion well, or are not applicable at all. Moreover, uncertainty of information exposure intrinsically exists between two connected nodes, which causes modeling diffusion more challenge in social networks. In this work, our aim is to design a diffusion model in which factors considered are flexible to be extended and changed and the uncertainly of information exposure is explicitly tackled. Therefore, with different factors, our diffusion model can be adapted to more scenarios of diffusion without requiring the modification of the learning framework. We conduct comprehensive experiments to show that our diffusion model is effective on two important tasks of information diffusion, namely activation prediction and spread estimation.

Existing System:

In addition to many factors that affect the adoption decision, uncertainty of information exposure essentially exists in the real world and is noteworthy to be considered for modeling diffusion in social networks. For instance, to assume that a user will read every information from her neighbors in Online Social Networks (OSNs) will overestimate the spread size for influence maximization easily.

 

 

 

 

Proposed System:

Our MFAD model is flexible to extend and change factors since the proposed learning frameworks are independent of factors considered, achieved by employing a classification approach to predicting the adoption behavior of a node. Moreover, the frameworks are not limited to a specific classification algorithm.

Due to the limitation of observation on diffusion in the real world, to predict adoption behaviors is hard to reach good results. We explicitly tackle this issue by learning nodes’ classifiers for adoption decision with only positive and unlabeled instances in the first scenario and by an Expectation-Maximization approach in the second scenario.

With further explicitly modeling uncertainty of information exposure between nodes, experimental results show that MFAD is effective on activation prediction and spread estimation. Especially on spread estimation, MFAD provides better quality. The quality is to measure how many nodes in the estimated spread of a diffusion model truly adopt an item after propagation. Previous studies mainly consider the quantity aspect, i.e. spread size, while ignoring the quality aspect, which will cause the estimated results of a solution to influence maximization far from the true real-world diffusion results.

Comment As:

Comment (0)