Final Recomputation & Visualization Regeneration

Current Status

✅ Phase Complete: Enrichment & Validation

• Seed list (200 papers) verified: 185/200 in corpus (92.5%)
• Missing papers identified & flagged: 15 papers (pending verification)
• Author merging applied: 17 expert name merges (35,797 → 35,641 authors)
• Paper roles refined: 10 granular categories with data-driven thresholds
• Key experts identified: Career arcs extracted for top 50 researchers

📊 Corpus Ready for Final Recomputation

• Size: 7,503 papers (with author merges)
• Authors: 35,641 (after merging)
• Citation edges: 85,589 (directed)
• Co-authorship edges: 511,670 (undirected)

Final Recomputation Pass: Tasks

1. Recompute All Metrics (on full corpus)

• PageRank (citation network, all 7,503 papers)
• HITS hub/authority scores
• Betweenness centrality
• K-core decomposition
• Community detection (Louvain)
• Composite scores (updated formula)

Output: paper_rankings_all_final.json (with fresh metrics)

2. Recompute Author Rankings

• Paper count (using merged author names)
• Co-author network metrics
• PageRank on co-authorship network
• Specialization scores
• Career timeline statistics

Output: author_rankings_final.json (35,641 authors)

3. Regenerate Visualizations

• field_map_full.html — Citation network (7,503 nodes, 85K edges)
  • Force-directed layout with zoom/pan
  • Ranked list (all papers, sortable)
  • Toggle citation/co-authorship networks
• coauthor_map_full.html — Co-authorship network (35,641 nodes)
  • Force-directed layout
  • Ranked author list
• career_arcs_plot.html — Interactive visualization (NEW)
  • Timeline: publications per year × expert
  • Color by paper role/importance
  • Hover: citation impact trajectory
• journal_club_threshold_*.html (4 versions)
  • Refresh with updated metrics
  • Threshold 10, 15, 20, 30
• kcore_map.html — K-core shells
  • Updated k-values and coloring
• evolution_graph_full.html — Field timeline
  • Verify data freshness
• index.html — Dashboard (updated)
  • Link to all visualizations
  • Updated statistics

4. Journal Club Refresh (all thresholds)

• Compute new importance scores
• Apply thresholds: 0.10, 0.15, 0.20, 0.30
• Assign tier labels (Platinum/Gold/Silver/Bronze)
• Calculate in-degree/out-degree metrics
• Generate markdown + JSON for each threshold

Output: 4 sets of files (threshold_10, 15, 20, 30)

5. Comprehensive Documentation Update

• BIBLIOGRAPHY_ANALYSIS_DOCS.md
  • Verify corpus size: 7,503 papers
  • Verify author count: 35,641 (after merges)
  • Update metrics explanations
  • Document career arc visualization
• METHODOLOGY_AND_PIPELINE.md (NEW)
  • Document full pipeline flow
  • Explain metric computations
  • Discuss thresholds and choices
  • Flag limitations & biases
  • Prepare for critical review

Quality Assurance Checklist

Before considering this “ready for review”:

• Verify all 7,503 papers have metrics
• Check author merges applied correctly (Gray-Roncal, Jefferis, etc.)
• Spot-check: top 10 papers ranked correctly
• Spot-check: Lichtman papers present and ranked appropriately
• Journal club at threshold 0.20 has ~22 papers ✓
• Visualizations load without errors
• Dashboard links all work
• Career arcs show expected trajectories
• Seed list (200 papers) confirms 185/200 in corpus

Execution Plan

Recommended sequence:

1. Re-run PageRank/HITS/betweenness (can parallelize)
2. Re-run k-core decomposition
3. Re-detect communities
4. Recompute author rankings (depends on corpus)
5. Regenerate ALL visualizations in parallel
6. Update documentation
7. QA spot checks
8. Lock for methodology review

Estimated scope: 6-8 discrete computational tasks, some parallelizable

For Future: Methodology Review

Once this recomputation is complete and verified, separate work session to:

1. Critical review of thresholds
   • In-degree/out-degree cutoffs (10 for each direction — too strict? too loose?)
   • Composite score formula (80% PageRank + 20% k-core — still optimal?)
   • Journal club thresholds (0.10, 0.15, 0.20, 0.30 — right choices?)
2. Methods paper weighting
   • Current: out-degree ≥37 (top 5%)
   • Consider: boost formula for infrastructure papers?
   • Consider: temporal dynamics (older methods vs. newer)?
3. Community detection robustness
   • Louvain algorithm choices (randomness, resolution)
   • Alternative: other clustering methods?
   • Validation: compare to hand-labeled communities
4. Author merging completeness
   • Are there other Gray-Roncal-like merges we missed?
   • Automated name similarity scoring?
5. Graph properties
   • Citation network: power-law? small-world?
   • Co-authorship: assortative? community structure?

Next Session: “Methodology Review & Refinement”

Once locked and verified, can tackle:

• Refined composite scoring
• Dynamic importance (temporal decay)
• Cross-validation against external rankings
• Potential precomputed indices for faster querying

Ready to execute final recomputation pass?