[论文解读] A Survey of Information Cascade Analysis: Models, Predictions, and Recent Advances

Fan Zhou, Xovee Xu+2 more|arXiv (Cornell University)|May 22, 2020

Complex Network Analysis Techniques参考文献 191被引用 49

一句话总结

This survey comprehensively reviews information cascade popularity prediction methods, categorizing by feature-based, generative, and deep learning approaches across macro-, micro-, and meso-level predictions, and outlines open challenges.

ABSTRACT

The deluge of digital information in our daily life -- from user-generated content, such as microblogs and scientific papers, to online business, such as viral marketing and advertising -- offers unprecedented opportunities to explore and exploit the trajectories and structures of the evolution of information cascades. Abundant research efforts, both academic and industrial, have aimed to reach a better understanding of the mechanisms driving the spread of information and quantifying the outcome of information diffusion. This article presents a comprehensive review and categorization of information popularity prediction methods, from feature engineering and stochastic processes, through graph representation, to deep learning-based approaches. Specifically, we first formally define different types of information cascades and summarize the perspectives of existing studies. We then present a taxonomy that categorizes existing works into the aforementioned three main groups as well as the main subclasses in each group, and we systematically review cutting-edge research work. Finally, we summarize the pros and cons of existing research efforts and outline the open challenges and opportunities in this field.

研究动机与目标

• Define types and problem formulations of information cascades and popularity prediction.
• Provide a taxonomy of prediction methods across feature-based, generative, and deep learning approaches.
• Review macro-, micro-, and meso-level prediction tasks and evaluation protocols.
• Summarize datasets, evaluation metrics, and open challenges in information diffusion research.

提出的方法

• Classify information cascades into prediction as classification or regression.
• Differentiate ex-ante prediction from peeking strategies based on available data.
• Organize methods into feature-based, generative, and deep learning categories with cross-network applicability.
• Discuss evaluation metrics and benchmark datasets used in popularity prediction.
• Highlight the trade-offs, advantages, and limitations of each methodological approach.
• Survey recent literature on graph representation learning and sequential models for diffusion tasks.

实验结果

研究问题

• RQ1 What are the main problem formulations for information cascade popularity prediction (classification vs regression, ex-ante vs peeking, macro/micro/meso levels)?
• RQ2 What are the predominant methodological approaches (feature-based, generative, deep learning) and their trade-offs across different networks and data types?
• RQ3 How do evaluation metrics and datasets shape comparisons of information cascade models?
• RQ4 What are the open challenges and opportunities in modeling information diffusion and predicting cascade popularity?

主要发现

• The paper provides a broad taxonomy: prediction can be classification or regression, before or after publication, and macro-, micro-, or meso-level in scope.
• It covers three method groups: feature-based methods, generative models, and deep learning approaches, including graph representation learning and sequential models.
• It discusses multiple networks and data domains (e.g., social networks, content sharing, and citation networks) and notes deep learning methods have recently become more popular.
• It reviews evaluation metrics (e.g., accuracy, precision, recall, F1, AUC, MAE, RMSE) and discusses issues with highly skewed popularity distributions.
• It emphasizes that many models are not easily generalizable across platforms and datasets, and highlights open challenges and opportunities in the field.