The Four Dimensions of Social Network Analysis: An Overview of Research Methods, Applications, and Software Tools

Motivation

• The exponential growth and vast data volumes of online social networks have created major challenges for traditional analysis methods and motivated new computational approaches.OSNs had around 3.484 billion active social media users, up 9% year-on-year, while data mining, statistics, and machine learning methods face scalability challenges.Source: §1 · p. 2

Research scope

• A scientometric study identifies active SNA research areas, application domains, and emerging topics, guiding the paper’s literature review.The study covers graph theory and network analytics, community detection, information diffusion, user profiling, topic extraction, sentiment analysis, Healthcare, Marketing, Tourism & Hospitality, Cybersecurity, Politics, fake news, and Multimedia.Source: §1 · p. 2

SNA dimensions

• The paper defines four SNA dimensions—Pattern & Knowledge discovery, Information Fusion & Integration, Scalability, and Visualization—and uses them to construct metrics called degrees.The dimensions address what can be discovered, scalability limits, the kinds of data that can be integrated, and what can be shown.Source: §1 · p. 3

Evaluation framework

• A global Capability metric named CSNA combines the defined metrics to rank SNA technologies, frameworks, and tools by capability and maturity.The assessment is intended to reveal strengths and weaknesses and provide guidance for SNA researchers.Source: §1 · pp. 2–3

Tool assessment

• The study evaluates 20 relevant SNA frameworks and tools using the dimensions and CSNA metric.This assessment supports comparison of available technologies and helps identify their main strengths, weaknesses, challenges, and trends.Source: §1 · pp. 2–3

Historical evolution

• RPYS identifies pronounced SNA citation peaks in 1967, 1973, 1977, 1979, 1988, 2011, and 2012, with a relatively stable period lacking standout works from 1994 to 2010.Standard RPYS detects deviations from five-year median citation counts, while Multi RPYS distinguishes lasting influence from short-term impact.Source: §2 · pp. 4–5, Figure 2 · p. 5

Research evolution

• SNA research shifted from defining foundational principles and structural-centrality measures in the 1970s and 1980s toward domain-specific methods, algorithms, applications, and community detection since 2000.Highly cited work moved from Social Sciences toward Physics, Mathematics, and Computer Science [34] [36] [37].Source: §2 · p. 5

Topics and applications

• Keyword topic analysis identifies Pattern & Knowledge Discovery, Information Fusion, Scalability, and Visualization, alongside applications in healthcare, education, governance, business, culture, and games.LDA extracts the top 20 topics from article keywords and visualizes frequent terms in a word cloud; representative terms include classification, prediction, Facebook, performance, perspective, and flow.Source: §2 · pp. 5–6, Figure 4 · p. 6

Current and emerging areas

• Current hot areas include Data Science and Big Data, network analysis, social media, sentiment analysis, text mining, and information diffusion, while leading applications include health, marketing and business, and tourism.The analysis also detects increasing activity in fake news and misinformation and cybersecurity, which are treated as emerging areas for further review.Source: §2 · pp. 6–7

3 Techniques and algorithms

• Social network analysis uses data-science and big-data algorithms to process, extract, and visualize knowledge from structured and unstructured data at large scale.In computer science and related fields, these methods address datasets too large or complex for traditional data-processing algorithms.Source: §3 · p. 7
• Structural-based analysis examines network connections, interactions, and topology, whereas content-based analysis focuses on information shared by users.The section presents these as complementary approaches to analyzing online social networks.Source: §3 · p. 7
• Analyzing only structural or content data provides an incomplete view of an online social network and may miss underlying patterns and knowledge.A complete analysis therefore considers both data types.Source: §3 · p. 7

3.1.1 Graph Theory and Network Analytics

• Graph theory represents social entities as nodes and their relationships as edges, enabling mathematical and visualization-based analysis of network structure through measures such as centrality, connectivity, cohesion, transitivity, and clustering.Centrality measures identify structurally important or influential nodes using degree, eigenvector, PageRank, betweenness, closeness, and group-based variants.Source: §3.1.1 · pp. 7, 9
• Network models reproduce real-network characteristics using degree distribution, clustering coefficient, and average path length, while social networks commonly exhibit community structure, high clustering, short paths, and sometimes scale-free degree distributions.Scale-free networks follow P(k) ∼ k^-γ, typically with 2 < γ < 3, and contain a small number of high-degree hubs; social networks also differ from other real-world networks through stronger community division and positive degree correlation.Source: §3.1.1 · pp. 8–9

3.1.2 Community Detection

• Community detection partitions graphs into densely interconnected clusters with sparse external links, using methods such as edge betweenness, modularity, random walks, spectral clustering, and statistical approaches.Non-overlapping methods assign each node to one community, whereas overlapping methods allow nodes to belong to several; performance varies with overlapping density, and detection remains unresolved in highly overlapping networks.Source: §3.1.2 · pp. 10–11
• Community-finding taxonomies distinguish static from dynamic analysis: static methods use a single network snapshot, whereas dynamic methods model community evolution, persistence, splitting, growth, and life cycles over time.Static methods are easier to apply but may be unrepresentative for continuously changing OSNs; dynamic models include aggregate, snapshot, and temporal-network representations of interactions.Source: §3.1.2 · pp. 10–12
• Dynamic community detection uses Snapshot or Temporal Network models, but granularity, community-event identification, stochastic instability, and long-term coherence remain key challenges.Snapshot methods require relating communities across adjacent snapshots, whereas Temporal Network methods continuously update communities and may diverge from static structures over time.Source: §3.1.2 · p. 13
• Snapshot methods track, detect, combine, or segment community structures, while Temporal Network methods update, track, or identify persistent communities across network evolution.Snapshot Community Tracking represents life cycles through Growth, Contraction, Merging, Splitting, Birth, Death, and Resurgence [43].Source: §3.1.2 · pp. 13–14

3.1.3 Information Diffusion Models

• Information diffusion models address viral propagation used in marketing and politics, while also examining misuse such as propaganda, recruitment, and fake-news spread.Models are grouped into explanatory epidemic-based models and predictive models; explanatory models include SIS [169], SIR, and SIRS, whereas predictive models include ICM and LTM.Source: §3.1.3 · pp. 14–15
• The Independent Cascade Model assigns one independent infection attempt per active node, whereas the Linear Threshold Model activates users when active-neighbor influence exceeds their thresholds.In LTM, each individual has a threshold φ ∈[0, 1], and diffusion ends when the active population stabilizes.Source: §3.1.3 · p. 15

3.2 Content-based Analysis

• Content-based SNA analyzes interaction content and node information to extract topics, opinions, and behavioral profiles using clustering, behavioral analysis, content analysis, and face detection.Profiles vary by application and purpose, while NLP, Data Science, and Big Data methods support content collection, preprocessing, mining, and pattern extraction.Source: §3.2.1 · p. 15, §3.2 · p. 15

3.2.1 User Profiling

• User profiling models behavioral patterns, correlations, activities, communication, and social ties to identify roles, interests, influential bloggers, and semantic user clusters.Methods include keyword-to-recipient mapping, group prediction, role labeling, influence and productivity ranking, similarity and propagation analysis, and SVD over socio-interaction matrices.Source: §3.2.1 · pp. 15–16

3.2.2 Topic Extraction

• Topic extraction discovers abstract document themes using mixture models, latent semantic indexing, LDA, and knowledge-driven methods, often combined with SNA to infer interests and communities.Related models integrate authors, recipients, roles, communities, text, links, retweet relations, categories, topical friends, and topic matching to predict interests or links.Source: §3.2.2 · p. 16

3.2.3 Sentiment Analysis

• Sentiment analysis is a complex NLP problem spanning personality detection, domain adaptation, multitask learning, and multiple granularity and modality choices.Applications include healthcare, political forecasting, tourism, misinformation detection, dialogue systems, trend discovery, profiling, influencer detection, polarization analysis, and group decision making; combining sentiment analysis with social structure improved product-classification rules over a no-SNA baseline [214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225].Source: §3.2.3 · pp. 16–17

Domain selection

• A scientometric study narrowed the extremely large set of OSN application domains to a selected subset for analysis.The selected domains are organized in Sections 4.1–4.5.Source: §4 · p. 17

Established domains

• The principal application domains are Healthcare, Marketing, Tourism and Hospitality, and Cyber Security.These domains are covered in Sections 4.1–4.4.Source: §4 · p. 17

Emerging areas

• Politics and detection of fake news and misinformation are identified as emerging areas of social network analysis.They are described in Section 4.5.Source: §4 · p. 17

4.1 Healthcare

• OSNs enable patients to share experiences and provide health knowledge, while social networks also facilitate the spread and analysis of obesity, smoking, alcohol use, and drug use.Research on OSNs and health has grown substantially, extending interaction beyond the traditional doctor-to-patient paradigm.Source: §4.1 · p. 17
• Obesity may spread through OSNs in a quantifiable pattern shaped by social-tie type, sex, smoking, and geographic distance.Christakis and Fowler analyzed 12,067 people repeatedly measured over 32 years in the Framingham Heart Study.Source: §4.1 · p. 17
• Online exposure to friends’ partying or drinking pictures was significantly associated with adolescent smoking and alcohol use, whereas OSN frequency and friend count were not.Among 1,563 10th-grade students, drinking friends increased drinking risk, while adolescents without drinking friends were more affected by risky online pictures; smoking also spread through close and distant ties.Source: §4.1 · p. 18
• 0.78 was the best correlation between flu-related Twitter messages and CDC statistics, while an SVM with polynomial kernels achieved the highest influenza-classification F-Measure of 0.756.OSN posts, relationships, and location tags support public-health surveillance, including outbreak tracking, disease assessment, mood analysis, and vaccination-community detection.Source: §4.1 · p. 18

Marketing

• Social media serves as a central platform for promotional activities, allowing companies to promote products and assess customer opinions while customers exchange views about brands and products.Source: §4.2 · p. 18
• Customer responses to social media advertising affect promotional effectiveness, while hedonic features that provide pleasure can generate more positive customer attitudes.Source: §4.2 · p. 18
• 19% of analyzed tweets mentioned a brand, and 20% of those tweets expressed brand sentiment; Twitter also reflected customer perceptions and influenced purchasing decisions.Source: §4.2 · pp. 18–19
• A Twitter model based on tweet-creation rates outperformed market-based predictors when forecasting movie box-office revenues.Source: §4.2 · p. 19
• Social media supports customer experience and relationship management, but its effects differ by platform: Facebook users follow brands yet are less likely to maintain contact or reshare content, whereas Twitter brand retweeters show trust, identification, and community commitment.Source: §4.2 · p. 19

• Social media supports hotel and destination marketing, while half of tourists change travel plans after researching trips on social media.Source: §4.3 · p. 19
• Tourism studies analyze online text, photos, or both, with the data source determining applicable algorithms and the patterns or knowledge extracted.Source: §4.3 · p. 19
• Text-based research mainly investigates customer opinions and experiences, using LDA, Sentic LDA, sentiment analysis, and clustering to study satisfaction, product attitudes, and geotagged trips.Sentic LDA uses SenticNet semantics to infer polarity for tourism aspects including accommodation, entertainment, food, and transportation.Source: §4.3 · p. 19
• Photo-based research typically analyzes metadata such as geolocation, timestamps, identifiers, titles, descriptions, and tags to study behavior, cluster photos, and recommend travel paths.Researchers use density-based spatial clustering to address centroid-based limitations, while Markov models predict tourists’ next destinations from current locations.Source: §4.3 · p. 19

Cyber-crime detection

• SNA extracts criminal and terrorist patterns from rapidly propagating online social-network content to detect crimes and relationships among offenders.Its applications support law-enforcement strategies against cyber-crime and cyber-terrorism across platforms such as Twitter and Facebook.Source: §4.4 · p. 20

Crime analysis

• Decision-support systems combine NLP, similarity measures, and classification to filter citizen crime reports, link related incidents, analyze trends, and improve prevention.Chih-Hao and Gondy designed such a system to facilitate communication between citizens and government agencies.Source: §4.4 · p. 20

Geographic crime prediction

• KDE was the most successful compared mapping techniques for identifying crime hotspots, while graph-based geographic analysis discovers regions with similar crime distributions.Phillips et al. modeled aggregated heterogeneous crime data as graphs, and Chainey et al. compared point, thematic, ellipse, grid, and KDE mappings; Gerber also used KDE for geotagged Twitter topics.Source: §4.4 · p. 20

Counter-terrorism in OSNs

• Terrorist organizations use public social networks for ideological promotion and recruitment before moving conversations to private messages, prompting efforts to identify and block associated accounts.Governments, organizations, and platforms search for accounts linked to terrorist groups to disrupt radicalization channels.Source: §4.4 · p. 20
• Text-analysis methods distinguish pro-ISIS from random Twitter accounts through linguistic markers, including plural pronouns, death-, certainty-, anger-related words, negative language, and radical vocabulary.These analyses target leakage, fixation, and identification warning behaviors and can precede studying how identified users influence others.Source: §4.4 · p. 20

Network-based counter-terrorism

• Betweenness centrality identifies critical terrorist-network nodes, but its O(nm) computational cost motivates heuristic and bio-inspired alternatives such as Lozano et al.’s artificial-bee-colony approach.These methods model social networks as graphs to identify accounts or nodes requiring intervention.Source: §4.4 · p. 21

4.5 Emerging Areas

• Emerging SNA research increasingly addresses political analysis, fake news and misinformation, multimedia integration, and future Social Internet of Things applications enabled by 5G, IoT, Machine Learning, and Artificial Intelligence.These topics are described as socially demanding and likely to become high-activity niches for SNA, Data Science, and Big Data research.Source: §4.5 · p. 21

4.5.1 Politics

• Political SNA applies text, network structure, and diffusion analysis to hate speech, polarization, communities, information exchange, and electoral behavior, with reported accuracies of 84% for right-wing hate speech and 86% for political-leaning classification.Studies also report positive links between affective language and retweet rates, politically segregated information exchange, complementary content and connectivity signals for community detection, and major quantitative but not ideological shifts after Egypt’s 2013 military takeover.Source: §4.5.1 · pp. 21–22

4.5.2 Detection of Fake News & Misinformation

• Fake-news detection comprises linguistic approaches using lexical and syntactic features and network approaches using stance, co-occurrence, or friendship graphs, with network metrics and SVD extracting diffusion or latent-node information [330] [332].The section defines fake news around verifiable falsity and intent to mislead, emphasizing the importance of detection because false content can spread rapidly through online social networks.Source: §4.5.2 · pp. 22–23

4.5.3 Multimedia

• Multimedia SNA is expanding as image, audio, and video sharing grows, with image methods supporting sentiment analysis, annotation, clustering, geolocation, popularity prediction, event discovery, and user-characteristic inference.Image analysis uses crowdsourcing, handmade features, or deep CNNs; audio remains largely untapped for SNA, while video is least studied and most complex, with existing work mainly transcribing content before sentiment analysis [349, 350, 351, 352, 353, 354, 355, 356, 357, 358, 359, 364, 365, 366, 367, 368, 369, 370, 371, 372].Source: §4.5.3 · pp. 23–24

Research challenges

• Billions of users generate massive textual, video, and multimedia streams, making even partial OSN processing exceed classical algorithms and requiring multidisciplinary collaboration.The needed collaboration spans fields from social sciences to science and engineering, including sociologists and data scientists.Source: §5 · pp. 24–25
• Four dimensions—knowledge discovery, scalability, information fusion, and visualization—address the core OSN questions of what to learn, process, integrate, and show.They correspond to structural/content analysis, exponentially growing data, heterogeneous formats and sources, and information visualization challenges.Source: §5 · p. 25

Technology assessment

• The proposed dimensions support quantitative maturity assessment of SNA tools and frameworks through spider-graph representations and the global CSNA metric.These dimensions are operationalized as metrics called degrees∗.Source: §5 · p. 25

Future directions

• The framework also identifies future challenges and development opportunities across Data Science, Big Data, information fusion, and visualization, with potential impact on science, industry, and society.It is intended to detect spaces for improvement and emerging technologies where new developments could have high impact.Source: §5 · p. 25

6 The Four Dimensions of Social Network Analysis

• The framework quantifies SNA technologies through dimensions corresponding to knowledge discovery, scalability, information integration, and visualization, clarifying their current status and future challenges.This part develops Pattern & Knowledge Discovery, Scalability, and Information Fusion & Integration; Visualization is identified but not detailed here.Source: §6 · p. 25

6.1 Pattern & Knowledge Discovery (D1)

• Pattern & Knowledge Discovery evaluates the capacity of SNA techniques and tools to discover non-trivial knowledge from OSN, including graph-theoretic analysis and cohesive-subgroup identification.Examples include centrality and transitivity measures, community identification, topic detection, and sentiment analysis.Source: §6.1 · pp. 25–27

6.3 Information Fusion & Integration (D3)

• Information Fusion & Integration measures the variety of data usable for SNA through multichannel data types, multimodal OSNs, and multi-representation models.This dimension parallels Big Data’s concept of Variety and asks what kinds of data can be integrated.Source: §6.3 · p. 30
• Information Fusion/Integration measures data variety with Multichannel, Multimodality, and Multi-representation, each ranging from 1/3 to 1, and assesses normalized dVariety on [0,1].Multichannel counts data types, Multimodality counts OSNs, and Multi-representation reflects representation-model complexity.Source: §6.3 · pp. 31–32

6.4 Visualization (D4)

• Visualization evaluates tools by the number of visual variables they support and their interaction capabilities, rather than ranking diagram types.The framework includes position, size, shape, orientation, color, saturation, and texture, alongside zoom, filter, highlight, grouping, and multiview interactions.Source: §6.4 · pp. 34–35
• The visualization degree combines visual-variable and interaction counts through Equation 8, with equal weights for the two characteristics and their constituent features.The proposed metric uses FV_isual and FInter with α=β=0.5 and γ=θ=1.0.Source: §6.4 · p. 35

6.5 Summary on Social Network Analysis Dimensions

• The four research questions are mapped to quantitative SNAdegrees for Pattern & Knowledge discovery, Scalability, Information Fusion & Integration, and Visualization.Table 1 summarizes the dimensions and their defined quantitative measures, while radar diagrams display each technology’s strengths and weaknesses.Source: §6.5 · p. 35, Table 1 · p. 35
• CSNA ranks SNA technologies by computing the area of the irregular polygon formed by the four normalized dimension values, with C4 ∈[0,2].If two dimensions are zero, C4 uses a two-dimensional projection; if three are zero, the corresponding single SNAdegree evaluates capability, and the metric is used to rank analyzed tools and frameworks.Source: §6.5 · pp. 35–36

Tool selection

• The study selects 20 representative tools from 70 candidates, including computing libraries, web, distributed, and desktop applications under open-source and proprietary licenses.Selection considered license type, documentation quality, and current community impact.Source: §7 · p. 37

Evaluation method

• Each tool is scored top-down with global C4 followed by the four SNAdegrees, using a publicly available rubric grounded in documentation, official websites, and published technical studies.The authors use this approach because installing and experimentally evaluating every software package was not feasible.Source: §7 · p. 38

Overall rankings

• Graphistry achieves the best global C4 score, leading Information Fusion, Scalability, and Visualization, while ranking 13th in Knowledge Discovery; Neo4j is the best open-source tool.No single tool dominates every dimension, and 15 of the 20 analyzed tools appear in at least one top-five ranking.Source: §7 · pp. 38–39, Table 3 · p. 38, Figure 9 · p. 39

Dimension rankings

• ORA-LITE/PRO leads Knowledge Discovery and Graphistry leads Information Fusion, while Graphistry, AllegroGraph, and Neo4j tie for top Scalability and ORA-LITE/PRO, Graphistry, and Neo4j lead Visualization.ORA-LITE/PRO exceeds SNAP in Knowledge Discovery, and Graphistry exceeds NetMiner in Information Fusion.Source: Figure 9 · p. 39, §7 · p. 40, Table 4 · p. 40

Feature-level analysis

• Feature coverage is uneven: Measures Topology, Link analysis, and Functional features score highly across tools, whereas Opinion Mining, Homophily, and Multi-Modality are addressed by few tools.ORA-LITE/PRO is the only tool scoring above zero on every Knowledge Discovery feature, while Graphistry reaches the maximum on every Information Fusion feature.Source: §7 · pp. 40–41, Figure 11 · p. 40

Cross-dimension relationships

• Visualization is positively related to Knowledge Discovery and Information Fusion, Information Fusion is not positively related to Knowledge Discovery, and Scalability is inversely related to both Knowledge Discovery and Information Fusion.The inverse relationship is stronger for Scalability versus Knowledge Discovery; Graphistry strongly dominates the Scalability–Information Fusion pairing, highlighting a gap between its commercial capabilities and open-source tools.Source: Figure 12 · p. 41, §7 · p. 42

Scope and contributions

• The study narrows a rapidly expanding SNA literature through scientometric analysis of major research areas and application domains, then defines dimensions and degrees to assess 20 popular tools.The reviewed methods span structural, content-based, and multimedia analysis, while the dimensions target Pattern & Knowledge discovery, Information Fusion & Integration, Scalability, and Visualization.Source: §8 · pp. 42–43

Conclusions by research question

• Pattern & Knowledge discovery is common for topology, link analysis, and static communities, but rare for dynamic communities, opinion mining, virality, and homophily, indicating that OSN content remains underexploited.The authors link this gap to a lack of proportionality between Information Fusion and Knowledge Discovery.Source: §8 · p. 43
• Most tools handle graphs of around 100,000 nodes, offer customizable code and APIs, but few support BigData, and those that do show low or medium Knowledge Discovery capabilities.Future tools are expected to scale toward millions or hundreds of millions of nodes and vertices while improving fusion and integration across sources and tools.Source: §8 · p. 43
• Most tools support complex graph representations and two or three data types, but generally extract data from only one OSN, motivating research on cross-format and cross-network information fusion.The paper expects substantial growth in integrating different data formats and, where possible, multiple OSNs.Source: §8 · p. 43
• Visualization capabilities are more evenly distributed than other dimensions, yet tools combining high Scalability and Visualization remain scarce, making visualization a major future priority.Suggested directions include dynamic community finding, data analytics, and pattern finding.Source: §8 · p. 43

Limitations and future use

• The four dimensions and degrees are an initial, incomplete framework whose scores reflect the current technology and evaluation rubric rather than fixed limits on future improvement.The authors present the metrics as a potential decision tool for future SNA researchers and practitioners.Source: §8 · pp. 43–44

Open access and collaboration

• The website gives researchers, analysts, developers, and engineers access to the article’s data and supports future collaboration on network analysis.It is designed to broaden participation across social sciences, science, engineering, and SNA practice.Source: Appendix: Open Access to Social… · p. 44

Shared resources

• The repository includes new software, documents, open papers, technical data, the evaluation rubric, completed tool assessments, collaborative README guidance, and a paper summary.These resources support evaluating SNA tools, frameworks, and algorithms.Source: Appendix: Open Access to Social… · p. 44

Community participation

• The appendix encourages the community to submit evaluations of assessed and additional SNA tools to promote technology use and new developments.The authors state that continued progress depends on community cooperation.Source: Appendix: Open Access to Social… · p. 44