Module 4 – Data Visualization & Interpretation in Bibliometrics

Focus: Data visualization & interpretation in bibliometrics.
Instructor: Taylor Pape.
Context: Final module in the series (after data collection & cleaning from previous modules).
Core objective: “Bring data to life” so that trends, patterns, and structures hidden in bibliometric spreadsheets become immediately comprehensible.

Enables rapid detection of:
- Publication trends (e.g.
- Which topics are rapidly growing.
- Which authors receive the highest citation counts).
- Collaboration structures (authors, institutions, countries that co-publish).
- Research clusters: groups of papers linked by shared keywords or citations.
- Topic evolution across time.
Practical benefits:
- “Big-picture” view for scholars, librarians, policy analysts.
- Supports strategic decisions—e.g., choosing collaborators or funding priorities.
- Storytelling & communication of findings to non-experts.

Co-authorship networks
- Nodes = authors.
- Edges = joint publications.
- Edge weight = number of shared papers.
Co-citation networks
- Nodes = articles (or books).
- Edges = frequency with which two items are cited together.
- Reveals intellectual lineage & foundational papers.
Keyword co-occurrence maps
- Nodes = keywords.
- Edges = joint appearance in the same article.
- Cluster colors → research themes/hotspots.
- Example clusters (public-health dataset):
- Red = epidemic preparedness.
- Green = disaster response.
- Blue = COVID-19.
Temporal keyword trend analyses
- Line plot of keyword frequency vs. year.
- Tracks rising/falling popularity of concepts.
- Example (intelligence research 2002–2016): terms like “modeling,” “fluid intelligence,” “reasoning” shift in prevalence.

ggplot2
- Flexible grammar of graphics; ideal for time-series & trend plots.
- Supports smoothing (trendlines) + confidence-interval shading.
bibliometrix / Biblioshiny
- networkPlot() for quick co-citation, co-authorship, and keyword networks.
igraph
- Core network-analysis engine; converts adjacency matrices to graph objects.
ggraph
- ggplot2-style plotting layer for igraph objects.
- Common layout used: Fruchterman–Reingold (layout = fr).
VOSviewer (external desktop software)
- Specialized for bibliometric mapping; demonstrated for a sustainability-tourism dataset.

Node size
- Often encodes importance (e.g., high citation count or high keyword frequency).
Cluster color
- Algorithms (e.g., Louvain, VOS clustering) group densely connected nodes → indicates thematic areas.
Timeline (horizontal axis in overlay or longitudinal maps)
- Right-most/upper-most nodes frequently indicate newer or emerging topics.
Edge thickness / weight
- Co-authorship: number of joint papers.
- Co-citation: strength of intellectual connection.
Practical reading strategy:
1. Scan for large nodes (influential items).
2. Examine tightly knit colored clusters (research specialties).
3. Follow thick edges to trace critical collaborations or conceptual links.
4. Compare positions across time when timeline overlay is present.

Co-authorship map (sustainability & tourism)
- Clear clusters suggest institutional or topical subgroups.
- Bridges identify authors who connect otherwise separate clusters.
Keyword co-occurrence map (public health)
- Three dominant clusters: epidemic preparedness, disaster response, COVID-19.
- Term proximity reveals interdisciplinarity.
Temporal keyword line plot
- Example of spiny lobster keyword: stable presence from 1990–2020.
- Highlights sustained versus fad topics.
Co-citation network (niche pyloric-circuit neuroscience)
- Dense, compact graph → small, highly inter-referencing community.
- Thick edges among a handful of seminal papers.

Setup (reuse Module 3 data)
- Install required packages: ggplot2, bibliometrix, igraph, ggraph, etc.
- setwd() to working directory containing cleaned data.
Keyword Trend Plot
- Select top keywords: trendplotdata keywordtrends filter(keyword %in% topkeywords).
- ggplot template:
ggplot(trendplotdata, aes(x = year, y = count, color = keyword)) + geomline(size = 1.2) + labs(title = "Top Keyword Trends Over Time", x = "Publication Year", y = "Number of Articles", color = "Keyword") + thememinimal()
- Interpretation: Spikes show moments of heightened research interest.
Co-authorship Network
- Convert adjacency matrix → igraph:
gauthors graphfromadjacencymatrix(as.matrix(net_author), mode = "undirected", weighted = TRUE, diag = FALSE)
- Plot with ggraph:
ggraph(gauthors, layout = "fr") + geomedgelink(aes(width = weight), alpha = .6) + geomnodepoint(color = "blue", size = 5) + geomnodetext(aes(label = name), repel = TRUE, size = 3) + themevoid() + ggtitle("Co-Authorship Network")
- Reading: Thick edges → frequent collaborations; isolated nodes → lone researchers.
Co-citation Network (Top 50 References)
- Identify top items: top order(rowSums(net_citation), decreasing = TRUE)[1:50].
- Subset matrix; convert to igraph (diag = FALSE, weighted = TRUE).
- Plot with same ggraph scaffold.

Module 1–2: Data collection from Web of Science → exported records.
Module 3: Cleaning, deduplication, creation of adjacency matrices for authors, citations, keywords.
Module 4 rides on those data frames to build visuals.

Transparent visualization assists:
- Funding agencies in spotting impactful or neglected areas.
- Researchers in finding collaborators.
- Policymakers in understanding knowledge landscapes.
Responsibility: Avoid misleading scales or cherry-picked time windows.
Encourage sharing code & figures for reproducibility (cf. Slack channel & office hours support).

Experiment with your own datasets (not limited to examples shown).
Tweak aesthetics (colors, layouts, node sizing) to enhance clarity.
Combine multiple visual layers (e.g., overlay temporal evolution on network maps).
Engage with community via Slack, office hours; submit Google Form to receive tailored guidance.