[Paper Review] Generic Multilayer Network Data Analysis with the Fusion of Content and Structure

Xuan-Son Vu, Abhishek Santra+2 more|arXiv (Cornell University)|May 21, 2019

Complex Network Analysis Techniques28 references4 citations

TL;DR

This paper proposes a multilayer network (MLN) framework that fuses content-based features (e.g., political views from text) with structural network data for flexible, efficient analysis of large social networks. By modeling each feature as a separate layer and using node intersection of precomputed communities, the approach achieves up to 87% reduction in computation time compared to traditional graph-composition methods, especially scaling efficiently with complex, multi-feature queries on real-world Facebook data.

ABSTRACT

Multi-feature data analysis (e.g., on Facebook, LinkedIn) is challenging especially if one wants to do it efficiently and retain the flexibility by choosing features of interest for analysis. Features (e.g., age, gender, relationship, political view etc.) can be explicitly given from datasets, but also can be derived from content (e.g., political view based on Facebook posts). Analysis from multiple perspectives is needed to understand the datasets (or subsets of it) and to infer meaningful knowledge. For example, the influence of age, location, and marital status on political views may need to be inferred separately (or in combination). In this paper, we adapt multilayer network (MLN) analysis, a nontraditional approach, to model the Facebook datasets, integrate content analysis, and conduct analysis, which is driven by a list of desired application based queries. Our experimental analysis shows the flexibility and efficiency of the proposed approach when modeling and analyzing datasets with multiple features.

Motivation & Objective

• Address the challenge of analyzing large, complex social network datasets with multiple heterogeneous features (e.g., age, political views, relationships).
• Overcome limitations of traditional single-graph modeling, which becomes computationally infeasible and inefficient when handling multiple features and relationships.
• Enable flexible, user-driven analysis by allowing arbitrary combinations of features through a decoupled, layered network model.
• Integrate content analysis (e.g., deriving political views from text posts) seamlessly into structural network analysis for richer insights.
• Demonstrate scalability, efficiency, and accuracy of the MLN approach on a real-world, large-scale Facebook dataset.

Proposed method

• Model each feature (e.g., age, political view, locale) as a distinct single-layer graph, forming a multilayer network (MLN) with M layers.
• Precompute communities for each layer using the Infomap algorithm, storing both community lists and internal clustering coefficients to assess self-preserving properties.
• For each query requiring K-layer AND-composition, compute the result by intersecting communities from K layers based on shared nodes, avoiding full graph composition.
• Leverage parallelism to process individual layers simultaneously, significantly reducing overall computation time.
• Use a principled composition method to combine layers without constructing intermediate composed graphs, improving storage and time efficiency.
• Validate results through comparison with traditional graph-composition approaches (SLG), which generate full AND-composed graphs for each query.

Experimental results

Research questions

• RQ1 Can multilayer network modeling enable more flexible and efficient analysis of large, multi-feature social network datasets compared to traditional single-graph approaches?
• RQ2 How does the integration of content-based features (e.g., political views derived from text) with structural network features enhance analytical insight?
• RQ3 What is the computational advantage of using MLN with community intersection over traditional graph-composition methods as the number of queries and layers increases?
• RQ4 To what extent does parallel processing of layer-wise community detection improve scalability and performance on real-world datasets?
• RQ5 Can the MLN approach maintain result accuracy while significantly reducing storage and computation costs in complex, multi-feature analysis scenarios?

Key findings

• The MLN approach reduced total computation time by 36% (from 78.520 to 50.354 seconds) compared to the traditional approach without parallelism.
• With parallel processing, the MLN approach achieved a 87% reduction in total computation time (down to 9.987 seconds), demonstrating strong scalability.
• The most computationally expensive query (Q2b) required only 0.039 seconds for composition, while the least dense layer took 1.61 seconds—highlighting a two-order-of-magnitude difference favoring MLN.
• The MLN approach outperforms traditional methods in both time and storage, especially as the number of queries (N) and layers per query (K) increase, due to avoiding exponential AND-composition operations.
• The method achieved a 80.4% reduction in processing time for the densest layer (L5: Locale) when using parallelism, showing strong performance gains on high-complexity components.
• The results validate that MLN with precomputed communities and node intersection provides accurate, efficient, and scalable analysis for complex, multi-feature social network queries.