NetMiner

19/05/2026

β coefficients reveal the average. SHAP reveals what the average hides.

Regression estimates linear relationships across the full dataset. When patterns are non-linear, or when you need to explain why a specific group or case looks different, it has limits.

SHAP fills that gap. It shows how much each variable contributed to a given prediction, captures non-linear effects automatically, and requires no statistical assumptions.

A few examples from recent research:

- 243 survey questions. SHAP identified the 27 that actually explain life satisfaction, with nearly identical accuracy (Khan et al., 2024)
- In a credit card dataset, being female lowered predicted VIP approval odds even when all other variables were identical (Moon, 2026)
- Spam tweet spread was predicted better by profile photo facial expression than by tweet content (Dhar & Bose, 2025)

SHAP is not a replacement for regression. It is what you reach for when the average is not enough.

👉 Read how SHAP complements regression in research: https://medium.com//what-%CE%B2-coefficients-dont-tell-you-explaining-models-with-shap-d477cb4bcb62

1. Where Regression Analysis Falls Short

15/05/2026

Can a straight line explain everything in your data?

Regression struggles to explain non-linear patterns or why a specific group turns out differently from the rest.

XAI (Explainable AI) is now available in NetMiner. See each variable's actual contribution at the case level, including non-linear effects. Two-mode Network Centrality is also included in this update.

👉 See the update: https://stib.ee/ECeN

11/05/2026

When SNA provides AI with structured context, reasoning accuracy changes.

95% of enterprise generative AI pilots are failing — MIT NANDA points to a lack of structured context as the root cause.
- GraphRAG with SNA: up to 35% improvement in answer accuracy over vector-only retrieval
- SNA-based anomaly detection: AUC improved by 20–30% on average

See how SNA's topological and structural context gives AI genuine situational awareness.

👉 https://netminer.medium.com/when-sna-becomes-the-context-for-ai-de6c9672a29d

Key Takeaways

08/05/2026

Formally assigned mentoring is not what drives teacher growth. The informal conversations that happen without any institutional design are.

Six recent SNA studies in education reveal what conventional statistics miss.

- The undesigned network was stronger than the designed one
- Teacher collaboration shows no direct path to student achievement. Only when instructional quality mediates the relationship does it become significant
- High participation rates in online courses often mask a star-shaped network with little peer-to-peer exchange

A roundup of how SNA is being applied in educational research.

👉 Read the full roundup: https://medium.com//social-network-analysis-sna-in-educational-studies-ebcb32983308

Education is a discipline that explores the complex social ecosystem in which human development and learning unfold. The interactions among…

06/05/2026

UPDATE! Chinese text analysis is now built into NetMiner.

No external preprocessing needed. From morphological analysis to co-occurrence network generation, it's handled in one place.

- Dedicated Simplified Chinese morphological analyzer, newly added
- Fully compatible with existing text mining features (word extraction, POS filtering)
- Co-occurrence networks generated automatically from analysis results

If you work with Chinese-language research data, this update is ready to use.

👉 https://stib.ee/ylUN

NetMiner provides graph/network analysis and graph machine learning algorithms to help you discover the extraordinary value in your graphs.

30/04/2026

Map a prefabricated building supply chain through Social Network Analysis and risk propagation paths become visible.

Zhu & Li (2025) analyzed ICT-based information sharing risks using NetMiner and Social Network Analysis.
Traditional risk checklists couldn't reveal what network structure made clear.

- "Lack of information management specification (S1R27)" recorded the highest out-status centrality at 3.090007 among 49 risk nodes
- Mitigating critical nodes reduced network density by 41% and cohesion by 24% — data-driven evidence for where to act first

The numbers make the case for governance frameworks before ICT investment.

👉 Read more: https://netminer.medium.com/more-information-more-risk-the-hidden-downside-of-smart-construction-ab9c3cb5c1fd

Smart construction is accelerating. IoT sensors monitor job sites in real time, BIM models every component in 3D, and big data coordinates…

27/04/2026

Mapping a digital classroom through Social Network Analysis reveals the actual structure of a learning community.

Exarchou et al. (2017) analyzed the online collaborative activities of 16 students using NetMiner and Social Network Analysis.
What traditional participation counts couldn't show, network structure made clear.

- Network density of 0.488 — nearly half of all possible student connections were realized
- St19 and St21 recorded outdegree 14, serving as dominant knowledge hubs — actionable insight for teachers and instructional designers

Visualizing who connects with whom helps educators spot peripheral learners before disengagement sets in.

👉 https://medium.com//31e67ca066eb

Digital collaboration tools have transformed the way students learn. Social bookmarking platforms, shared maps, and online discussion…

23/04/2026

Analyzing national R&D projects as a co-occurrence network reveals gaps in how funding is distributed.

Yang & Cho (2023) analyzed 4,739 government-funded R&D projects in South Korea using co-occurrence network analysis.
What paper counts and patent tallies couldn't show came into focus.

- Some technology fields with low network centrality were receiving the largest budgets including mechanical/electronic component materials and high-strength heat-resistant materials.
- Meanwhile, hub technologies critical to the broader ecosystem were underfunded thin-film technology, nano-composites, and others.

See how network analysis can help diagnose R&D budget allocation at the system level.

👉 Read more: https://netminer.medium.com/050b1cd042e7

&DPolicy

South Korea invested 4.55% of its GDP in R&D in 2017 — nearly double the OECD average of 2.37%. With that scale of public funding, how the…

23/04/2026

Struggling with a network so dense you can't make sense of it?
Not sure where to even start with preprocessing?
Check out what's new in this update.

- Core network filtering (Backbone): Automatically extract only statistically significant connections
- Subgroup-level metrics: Compare structural differences across communities with quantitative indicators
- And isolated node removal, new network visualization layout, k-NN regression, and normalization options

Try it free today.
👉 https://www.netminer.com/en/support/news.php?bgu=view&idx=523

21/04/2026

30 years of nursing medication safety research, mapped as a network — and it looks different than you'd expect.

Park & Han (2023) analyzed 2,085 papers using NetMiner and text network analysis.
Reading papers one by one doesn't show you this.

- The most connected keyword across 30 years wasn't 'medication errors' — it was 'patient'
- Medication error research has the largest share (32%), but its proportion has been declining over time

See where the field is heading:

👉 Read the full case study: https://medium.com//94d3501f8473

Every year, millions of medication errors occur worldwide. The WHO identifies them as one of the most common types of preventable harm in…