Understanding Social Causalities Behind Human Action Sequences.

Area of Science:

• Social Network Analysis
• Causal Inference
• Behavioral Science

Background:

• Understanding human behavior on social networks is crucial.
• Accurate causal discovery is challenged by indirect causalities and confounding factors like homophily and simultaneity.
• Existing methods struggle with latent confounders in observational action sequences.

Purpose of the Study:

• To develop a systematic approach for detecting asymmetric causal relations in human action sequences.
• To address challenges posed by latent confounders, homophily, and simultaneity.
• To eliminate redundant indirect causalities for accurate causal structure learning.

Main Methods:

• Conditional independence testing to infer causal directions between action sequences.
• Asymmetric independence tests to identify pairwise causal relationships.
• Causal structure learning to output refined pairwise causalities.
• Distinguishing relationship types like homophily and simultaneity.

Main Results:

• The proposed approach effectively detects asymmetric relations in human action sequences.
• Empirical evaluations on simulated data demonstrate effectiveness and scalability.
• Four causal relation patterns were identified on Sina Weibo data, including two novel patterns.

Conclusions:

• The systematic approach successfully identifies causal structures in social network actions.
• The method is effective in handling latent confounders and complex relationship dynamics.
• Findings offer new insights into social causality and human behavior patterns online.