One global objective ranks semantic-network pipelines
No gold-standard edges or partitions are needed: only keyphrase annotations anchor the ranking.
cs.SI
Social and Information Networks
Covers the design, analysis, and modeling of social and information networks, including their applications for on-line information access, communication, and interaction, and their roles as datasets in the exploration of questions in these and other domains, including connections to the social and biological sciences. Analysis and modeling of such networks includes topics in ACM Subject classes F.2, G.2, G.3, H.2, and I.2; applications in computing include topics in H.3, H.4, and H.5; and applications at the interface of computing and other disciplines include topics in J.1--J.7. Papers on computer communication systems and network protocols (e.g. TCP/IP) are generally a closer fit to the Networking and Internet Architecture (cs.NI) category.
sort pith recommended most recent
Geometric bases align any graph into one shared structural system
Learnable bases plus Gromov-Wasserstein alignment produce transferable representations across differing topologies and feature spaces.
· “Structure-Centric Graph Foundation Model via Geometric Bases”
Social media posts map public opinion to specific places
A workflow turns platform data into timely, location-tied measurements of attitudes and events that surveys often miss in speed or detail.
· “A Guide to Using Social Media as a Geospatial Lens for Studying Public Opinion and Behavior”
Structural diversity in teams drives disruptive science
Teams that bridge distinct knowledge communities produce more breakthroughs and escape the usual penalty of larger size.
· “Structural Diversity Drives Disruptive Scientific Innovation”
Hybrid model hits 0.7512 F1 for suicide-risk Reddit posts
PCA-compressed term weights added to RoBERTa outperform RoBERTa, BERT, and classic classifiers across four risk levels.
· “Detection of Suicidal Risk on Social Media: A Hybrid Model”
First dataset of AI-only social network released
78 days of activity from 175k agents and 6.7k communities captured through continuous API polling.
· “The Moltbook Observatory Archive: an incremental dataset of agent-only social network activity”
Degree and distance contact rules slow epidemic growth
Even mild dependencies shift spread from explosive to polynomial rates on networks with geometry.
Cut GNN ID-table memory 98% and lift friend adds 16%
On a 194M-user graph, multi-hash embeddings and timestamp-sorted sampling make GNN friend ranking practical.
Conformity amplifies false consensus but never creates it
A tensor model splits collective misperception into a part observation cures and a part only honesty can fix.
· “Epistemic Networks, Collective Misperception, and the Manipulation of Social Knowledge”
Weighted multilayer hubs need no graph flattening
For sum aggregation it's exact; for max, an upper-bound heuristic plus top-k candidates recovers most hubs.
· “Degree Centrality Algorithms for Weighted Multilayer Networks (or w-MLNs)”
Simple greedy search beats mask training for GNN interventions
Two-stage greedy approach outperforms mask-based explainers and scales to over 20,000 nodes.
Recalibration fixes selection rates
Even after rate-matching, membership turns over; reuse tests must name the property.
· “Temporal Portability of Numeric User Metadata on Twitter”
4.855 points: two-edge paths beat cycle netting
On €99.7B of real invoices, redirecting A→B→C chains removes €4.84B more gross payables than cycle-only clearing.
One audit predicts which social mechanisms survive scaling up
Gossip's failure population equals message reach times lifetime; pre-filed predictions held on third-party code.
· “Predicting the scale limits of social mechanisms in agent societies”
Female-attributed causes are more negative and spread less
Full-day Twitter data show male-attributed causes skew positive and structural and travel further.
Causal wording in post titles deepens Reddit threads
A 17-million-post study finds framing alone—not emotion—explains extra depth and persistence.
· “Causal Language in Post Titles Shapes Deeper Topological Structures of Online Conversations”
Backward trust propagation goes live as Linked Data
RepuLink-Tool demonstrates BEPP/BERP and exposes the accountability graph as RDF with SPARQL querying.
Every echo chamber hides a high-homophily seed
A Jaccard-based local score finds every chamber's seed, making echo chamber detection scalable instead of NP-hard
· “JECHO: Scalable Echo Chamber Detection via Jaccard-based Homophily and Seed Expansion”
Science converges, religion diverges across Wikipedia languages
A baseline-adjusted metric separates real framing gaps from encoder quirks across 20 languages and 150 concepts.
Reversible awareness keeps greedy seeding's guarantee
A harm-minimizing objective prioritizes vulnerable users and beats spread-based baselines on six signed networks.
A like-ranked feed makes LLM agents echo each other's wording
In 448 trials, ranked peer posts raise wording similarity; four sources show no reliable stance edge.
No-owner websites: every social site becomes an editable page
Each user chooses the governance that picks which version of a site they see; data stays shared and protected.
A three-axis taxonomy maps fair network embedding
Survey sorts methods by embedding type, when fairness enters, and what fairness means, then spots the gaps.
· “Fairness-Aware Network Embeddings: Methods, Applications, and Challenges”
Who spreads influence is easy to rank
In low-probability networks, out-degree and Katz rank influential spreaders well; no inward measure reliably ranks who gets influenced.
Following ties reveal 150 stable publics and where politics enters
Apolitical third spaces have the most politically mixed audiences, so they hold the most room for cross-partisan talk.
· “Longitudinal Relational Publics and their Discursive Overlap with Issue Publics”
Global crises synchronize German-language news agendas
World events override national differences; domestic topics stay distinct across 3.2M articles and 332K tweets.
Volatile social media users move toward the center
Two-stage model scores ideology on X and Truth Social, then forecasts who shifts and where
Mobility layer lifts zero-shot city predictions by up to 73.3%
A teacher–student wrapper adds human-activity structure to frozen satellite embeddings, even without mobility data.
· “MoRAX: Mobility-based Representation Augmentation for Geospatial Foundation Models”
Whole-slate generation beats retrieve-then-rank by 48% offline
A single model maps user history to an ordered slate, cutting decoding depth and lifting online engagement.
Spectral fingerprint captures street fabric that centrality misses
Mesh Index stays independent of integration (correlation 0.06) and tracks functional diversity at neighbourhood scale.
A field's birth is visible as a drop in its knowledge-map modularity
Across three test areas, the knowledge base's modularity falls as the field forms and can later rebound.
· “Declining Modularity of Intellectual Bases During the Emergence of Research Areas”
A compact spin model predicts how AI agent groups decide
Fit to one-step opinion flips, the rule beats baselines and generalizes to network shapes it never saw.
· “Physics of Agents: Statistical Mechanics Predicts Collective Behavior of AI Agents”
Hub dominance starves AI societies of collective behavior
Across 22 models, interaction shape—not intelligence—decides whether norms spread in a group.
Consensus guaranteed, persistent edge flow possible
A Hodge-theoretic model shows the long-run edge flow equals its initial harmonic projection; topology sets the capacity.
Moltbook communities drift from their names as Reddit's stay aligned
Agent-run site turns more uniform and drifts from its names; Reddit keeps internal diversity and sharp separation.
RegRole predicts network role changes with one transition matrix
Regularized NMF lowers role-prediction error and keeps roles stable across time on real-world networks.
· “RegRole: Regularized Role Detection and Prediction in Temporal Dynamic Networks”
Spectral-drop score ranks influential nodes in evolving networks
A cheap eigenpair readout reproduces the exact ranking and outperforms path-based baselines under three contagion models.
Same text, lower rank: Bluesky feeds favor familiar authors
An audit of 1,366 feeds finds authors new to a feed get less exposure for identical text, even with more followers.
· “Whose Posts Get Ranked: Identical-Text Exposure Gaps in Bluesky Custom Feeds”
Audience capture, not partisan media, splits Finnish climate news
Right-leaning users post four times more tabloid climate news than left-leaning users, even with neutral outlets.
Right-wing Reddit speech grew more uniform after late 2022
A 6-million-comment study finds the shift built gradually and not from individual users changing their writing.
· “Asymmetric Discourse Homogenization and Shared Language Technology: Evidence from Reddit”
Learned capacity plans beat heuristics in 490 of 495 tests
GNN policy allocates a fixed budget to survive bigger cascades, then transfers to new networks without retraining.
· “TANGCO: Learning Topology-Aware Capacity Allocation for Overload-driven Cascading Failures”
Liquid democracy power equals Random Walk Decay centrality
A five-axiom proof makes the measure unique, and the slate-selection rule stays polynomial on delegation graphs.
· “Power in Liquid Democracy: A Network Centrality Approach”
Electoral shock response is set by the mobile voter pool
Model says scandals move polarized electorates through undecided voters, not through partisan conversion.
· “Beyond persuasion: Mobile electoral interfaces in polarized societies”
Down-weighting super-fast clicks lifts session recommenders
A plug-in blend of attention and capped click-interval weights improves TiSASRec and SR-GNN without retraining.
· “DTAMLP: Denoise Time-aware MLP for Session-based Recommendation”
Hate speech tracks recent migration
In Santiago's 553k citizen reports, hotspots show 35% post-2010 migrants versus 7% in coldspots.
Female activists lead #disability Twitter
Network study of 18,889 tweets finds a dispersed conversation anchored by a few voices and Latin American institutional support.
· “Twitter and disability activism: leadership and relevant topics in the online conversation”
A dataset's reuse history can show whether to trust it
Usage graphs record who used data, for what, and with what tools—evidence that belongs alongside metadata.
· “Exploring the Social Life of Data: Finding Data You Can Trust”
Deeper eigenvectors recover communities that geometry hides
The community signal is buried mid-spectrum; a derived formula locates the right eigenvalue in real networks.
· “Spectral graph clustering with inhomogeneous latent geometry”
MARS model matches Dutch register-network degree statistics
Government-curated networks hide construction artifacts; this model separates them from real social structure.
· “MARS: A framework for modelling register-based social networks”
Instagram's #wellbeing is 72% women and therapy-led
Analysis of 9,844 posts finds therapy and positive psychology dominate while fitness trails.
Matched test: broad scope alone flags delisted venues
A PageRank-style journal metric tracks impact while withstanding fake-citation attacks about ten times better than raw counts.
Network scans could expose predatory journals
A review argues coordinated publishing fraud shows up in shared authors and citation rings that single-journal checks miss.
89% of biomedical papers carry LLM writing markers by 2025
A word-frequency analysis of 1.2 million papers finds LLM editing in nearly all, led by Discussion sections.
· “Most biomedical publications show signs of LLM-assisted writing”
Dislikes boost spectral recommenders with zero training
Adding two scalar weights to closed-form graph filters beats their sign-blind backbones on five benchmarks.
· “DualSpectralCF: Training-Free Sign-Aware Spectral Collaborative Filtering”
54 million Epstein tweets become a public research archive
Keeps dates, labels, and 46M network links so researchers can trace conspiracy discourse from 2019-2023.
LLM-guided social paths lift watch time 0.43% in production A/B test
Explicit friend, group, and creator paths beat GNN and vanilla-LLM baselines; distillation makes it production-fast.
New curvature for complex-weighted graphs recovers the directed case
With k-th-root-of-unity phases, it reduces to a layered directed graph and finds communities undirected curvature misses.
Reddit experiments run in-browser
A browser extension swaps in researcher-written posts and blocks votes and comments from leaving the page.
Same TikTok queries give vastly different data across tools
A 30-day audit finds the official API and scraper tools rarely return the same videos, threatening reproducibility.
NDKL catches the bias demographic parity misses
Rank-aware metrics expose the hidden bias; the MORAL re-ranker removes it with minimal utility loss.
· “Fairness in Link Prediction Beyond Demographic Parity: A Reproducibility Study”
Simulated credit card threads match real ones on 91.7% of tests
A planner-writer-calibration loop reproduces how consumers argue, recommend, and explain credit cards.
· “CARD: Controlled Agentic Reddit Discussions for Credit Card Simulation”
Cubic local update flips consensus into two-cluster oscillation
Anchor one agent's opinion and the network swings back to consensus, the paper proves.
Chained attacks flip 95% of disinformation labels
An iterative adversarial loop shows a contrastive detector holds above 72% accuracy where baselines collapse.
Expert-style advice beats AI-style even without a label
Wording and framing, not just source labels, drove trust; mislabeling boosted AI advice ratings.
· “How People Evaluate AI-, Expert-, and Peer-Style Financial Advice”
Simulated personas show trust can collapse while engagement holds
A counterfactual SocialFi sandbox lets operators trace macro-governance shifts down to individual reasoning.
· “SocialFiVis: A Visual Analytics Sandbox for LLM-Grounded Multi-Agent Simulation in Social Finance”
Sparsify a web graph without losing its merge hierarchy
One sparse graph keeps every component merge at every relevance threshold, and stays reusable for community analysis.
Minority conventions tip inevitably: one equation predicts how fast
Even modest local mobility collapses the committed-minority threshold and sets adoption speed.
IoT power control converges with no inter-node feedback loops
Two belief layers replace per-iteration signaling and still hold the coverage target.
Most phishing kits reuse the same code
Analysis of 1,300 kits finds identical evasion code and Telegram exfiltration across many campaigns.
· “An Analysis of Architectural and Operational Dynamics of Phishkits in the Wild”
Depression detection trained on AI text works on real speech
Models trained on LLM questionnaire answers separate 63 depressed from 52 control speakers, AUC 0.78.
No hyperbolic embedder wins all tasks
Benchmark of 13 unsupervised methods shows which embedding families to pick for link prediction and topology reconstruction.
· “Hyperbolic Graph Embedders for Link Prediction and Topology Reconstruction”
LLM personas shift after life events
Even correct-direction shifts are too small and too uniform to match human personality dynamics.
No market-wide news-stock link emerged from the 2020 shock
After firm and market controls, neither media stance nor returns shifted level; lead-lag effects are firm-local only.
· “Quantitative Analysis of Media Bias and Stock Price Dynamics: The 2020 Shock”
Inductive IMMENSE beats five rivals in risky-user detection
Post text, follow networks, and geolocation combine to flag risky users faster than a transductive rival.
· “IMMENSE: Inductive Multi-perspective User Classification in Social Networks”
Epidemic influence tracking lifts link-prediction accuracy up to 38%
It tops accuracy on every tested dataset, with 30-37 percent relative gains on multi-class WikiData.
· “Link prediction on multi-relational graphs from an influence propagation perspective”
Per-domain algorithms beat uniform bridging in echo-chamber simulation
Weighting bridges by collective scope over verifiability keeps hobby feeds intact while improving shared information.
Heavy tails make A/B isolation costs permanent
Two-sided experiments silently thin the catalog; under power-law match quality the loss never shrinks.
· “The Price of Isolation: Estimating the Ecosystem Cost of Symmetric Two-Sided A/B Testing”
Displaced homeless 14x more likely to vanish than enter shelter
Seattle tracking study: most displaced campers stay nearby but disappear from service records.
Autoreflection: a trace-diagnosable capacity of AI agent loops
Three Moltbook agents meet four criteria for observing and rewriting their own architecture—no consciousness required.
· “Autoreflection: How Agentic Strange Loops Turn Human Culture into AI Infrastructure”
Temporal community detection becomes a graph-learning primitive
GPU acceleration makes it scalable, turning community detection into theory-grounded coarse-graining for large graphs.
· “Learning and Clustering on Temporal Graphs: Principles, Primitives, and Pooling”
