[Paper Review] Temporal Network Motifs: Models, Limitations, Evaluation

Penghang Liu, Valerio Guarrasi+1 more|arXiv (Cornell University)|May 24, 2020

Complex Network Analysis Techniques44 references4 citations

TL;DR

This paper provides a comprehensive comparative analysis of four leading temporal motif models, evaluating their strengths and limitations in capturing temporal inducedness and timing constraints. It introduces event pairs as a novel lens to uncover hidden temporal correlations, revealing asymmetrical interaction patterns—especially in message networks—while demonstrating that combining timing constraints yields better motif diversity and reduced bias.

ABSTRACT

Investigating the frequency and distribution of small subgraphs with a few nodes/edges, i.e., motifs, is an effective analysis method for static networks. Motif-driven analysis is also useful for temporal networks where the spectrum of motifs is significantly larger due to the additional temporal information on edges. This variety makes it challenging to design a temporal motif model that can consider all aspects of temporality. In the literature, previous works have introduced various models that handle different characteristics. In this work, we compare the existing temporal motif models and evaluate the facets of temporal networks that are overlooked in the literature. We first survey four temporal motif models and highlight their differences. Then, we evaluate the advantages and limitations of these models with respect to the temporal inducedness and timing constraints. In addition, we suggest a new lens, event pairs, to investigate temporal correlations. We believe that our comparative survey and extensive evaluation will catalyze the research on temporal network motif models.

Motivation & Objective

• To compare and evaluate four major temporal motif models—Kovanen et al., Song et al., Hulovatyy et al., and Paranjape et al.—in terms of their treatment of temporality.
• To identify and analyze overlooked aspects of temporal networks, such as intermediate events and temporal correlations, in existing motif models.
• To propose a new analytical lens—event pairs—to investigate sequence-level temporal correlations and interaction dynamics in motifs.
• To assess the impact of model assumptions on motif frequency and spectrum, particularly in real-world datasets like email, call, and messaging networks.
• To guide future research by highlighting trade-offs in timing constraints and model biases, promoting more robust and application-aware temporal motif analysis.

Proposed method

• Surveyed and contrasted four prominent temporal motif models based on their handling of temporal inducedness and timing constraints.
• Evaluated models using two key metrics: temporal inducedness (consecutive events and constrained dynamic graphlets) and timing constraints (ΔC and ΔW).
• Introduced event pairs as a new analytical lens to represent ordered sequences of events in motifs, enabling analysis of temporal correlations.
• Conducted experiments on real-world datasets (SMS-A, SMS-Copenhagen, Calls-Copenhagen, Email) to compare motif frequencies and sequence patterns.
• Used heat maps to visualize event pair sequences, color-coded by motif density, to reveal dominant interaction patterns across datasets.
• Analyzed the interplay between different event pair types (e.g., repetitions, ping-pongs, in-bursts, out-bursts) and their asymmetrical trends in different network types.

Experimental results

Research questions

• RQ1 How do the four existing temporal motif models differ in their treatment of temporal inducedness and timing constraints?
• RQ2 What are the effects of temporal inducedness restrictions and timing constraints on motif frequency and spectrum across diverse datasets?
• RQ3 How do event pair sequences reveal hidden temporal correlations and interaction dynamics in real-world temporal networks?
• RQ4 What are the implications of model-specific biases for real-world applications such as social media, communication, and neural networks?
• RQ5 Can combining timing constraints (ΔC and ΔW) with a ratio in (1/(m−1), 1) reduce bias and improve motif diversity in temporal motif detection?

Key findings

• Temporal inducedness restrictions—such as requiring consecutive events or constrained dynamic graphlets—introduce consistent bias toward specific motif types across most datasets.
• Timing constraints ΔC and ΔW exhibit complementary strengths: ΔC fails to bound timespans, while ΔW introduces bias for intermediate events, suggesting that combining them via a ratio in (1/(m−1), 1) offers a balanced trade-off.
• Event pair sequences reveal asymmetrical interaction trends: in-bursts and out-bursts are rarely co-occurring, while conveys are often followed by out-bursts but not by in-bursts.
• Message networks show strong preference for sequences involving repetitions and ping-pongs, indicating localized, one-to-one interactions, whereas call and email networks favor out-bursts due to longer event durations.
• Weakly-connected event pairs are rare across all datasets, as they represent non-transitive event orders and are likely to be functionally irrelevant.
• The event pair lens uncovers previously overlooked temporal correlations, particularly in the asymmetry of information propagation patterns, such as conveys distributing information after in-bursts and preceding out-bursts.