Citation network

Citation Network & Technology Genealogy Skill

You are building a patent citation network to trace the evolution of a technology and identify the most influential patents in a field. This uses both backward citations (what a patent cites) and forward citations (what cites a patent) to reveal the structure of innovation.

Workflow Overview

Seed patent(s)
  --> Backward citations (prior art references)
    --> Forward citations (who cites this patent)
      --> Identify hub patents (most cited)
        --> Expand hubs (their citations in both directions)
          --> Read abstracts of key nodes
            --> Map technology evolution timeline
              --> Citation network report

Concepts

• Backward citations: Patents and literature cited IN a patent (its prior art references). Retrieved via get_patent_citations.
• Forward citations: Patents that CITE a given patent. Retrieved via search_patents with CQL ct="EP1000000".
• Hub patent / seminal patent: A patent with an unusually high number of forward citations — indicates foundational importance.
• Citation depth: How many generations of citations you trace (depth 1 = direct citations only, depth 2 = citations of citations).

Step 0: Identify Patent Family Context

Before any citation analysis, run get_patent_family on the seed patent to determine if it is a WO (PCT), EP divisional, US continuation, or national phase entry. This is critical because:

• Forward citations concentrate on the parent WO/PCT publication, not divisionals or national phase entries
• EP3401400 (a divisional) may have only 40 forward citations while parent WO2013176772 has 1,771
• If the seed is a divisional/national phase, redirect forward citation analysis to the parent WO for meaningful results

get_patent_family(document_number="EP3401400")

Identify the WO publication in the family, and use that as the primary seed for forward citation counts.

Step 1: Select Seed Patent(s)

The analysis starts from one or more seed patents. These can come from:

• A patent the user provides directly
• Top results from a prior art search
• A known foundational patent in the field

If the user describes a technology area rather than specific patents, run a targeted search first:

search_patents(query='ta="CRISPR" AND ta="Cas9" AND ta="gene editing"', detail_level="compact")

Pick 1-3 highly relevant patents as seeds. Prefer patents that are:

• From the core time period of interest
• From major applicants in the field
• Frequently discussed in the technology area

Step 2: Retrieve Backward Citations

For each seed patent, get its backward citations (prior art):

get_patent_citations(document_number="EP3401400")

This returns:

• Patent citations: Publication numbers and titles of cited patents
• NPL citations: Non-patent literature (journal articles, conference papers)
• Citation categories: Some citations are marked as "X" (particularly relevant), "Y" (relevant in combination), "A" (general background) by the examiner

Note the total count. Typical ranges:

• 5-20 citations: Normal for a focused patent
• 50-100+ citations: Common for US patents (broader duty of disclosure)
• Very few citations: May be an early/pioneering patent in the field

Step 3: Retrieve Forward Citations

For each seed patent, find patents that cite it:

search_patents(query='ct="EP3401400"', count_only=true)

Then retrieve the results:

search_patents(query='ct="EP3401400"', detail_level="compact", auto_paginate=true, max_results=200)

The forward citation count is a key measure of influence:

• 0-5 forward citations: Limited influence (niche patent or very recent)
• 10-50: Significant patent in the field
• 100+: Seminal / foundational patent
• 500+: Field-defining (rare)

Step 4: Identify Hub Patents

From both backward and forward citation data, identify hub patents:

1. Backward hubs: Patents that appear in multiple seed patents' backward citations — the shared prior art foundation
2. Forward hubs: Patents with the highest forward citation counts — the most influential

To check if a backward citation is itself highly cited:

search_patents(query='ct="US5580859"', count_only=true)

Rank patents by combined importance:

• Forward citation count (primary signal)
• Appearance in multiple citation lists (network centrality)
• Applicant/inventor prominence
• Priority date (earlier = more foundational)

Step 5: Expand Key Hubs (Depth 2)

For the top 3-5 hub patents, retrieve their citations in both directions:

get_patent_citations(document_number="US5580859")
search_patents(query='ct="US5580859"', detail_level="compact", auto_paginate=true, max_results=100)

This reveals:

• Predecessors of predecessors: The deepest roots of the technology
• Second-generation derivatives: How the technology branched and evolved

Limit depth to 2 for practical analysis. Depth 3+ grows exponentially and adds diminishing value.

Step 6: Read Key Abstracts and Biblio

For the most important nodes in the network (hubs, seeds, and notable outliers), retrieve abstracts:

get_patent_details(document_number="US5580859")

This gives you:

• Title and abstract (for understanding the technical contribution)
• Applicant (for mapping the competitive landscape)
• Filing/priority dates (for the timeline)
• Classifications (for technology clustering)

Step 7: Build the Technology Timeline

Organize the citation network chronologically:

1. Sort all identified patents by priority date (earliest filing)
2. Group into technology generations/eras
3. Map citation relationships to show how each generation built on the previous
4. Identify branching points where the technology diverged into sub-fields

Example timeline structure:

1990-1995: Foundational work
  US5580859 (1990) — First demonstration of DNA vaccination
  ↓ cited by 847 patents

1996-2005: Core development
  US6110898 (1997) — mRNA optimization for in vivo delivery
  WO0011140 (1999) — Lipid nanoparticle delivery
  ↓ ↓

2006-2015: Platform maturation
  US8278036 (2008) — Modified nucleosides in mRNA
  EP2459231 (2010) — Self-amplifying RNA vaccines
  ↓ ↓ ↓

2016-present: Therapeutic applications
  WO2017070626 (2016) — mRNA cancer vaccine
  EP3718565 (2020) — SARS-CoV-2 mRNA vaccine

Step 8: Compile the Citation Network Report

Report Structure

1. Network Summary

   • Seed patent(s) analyzed
   • Total unique patents in network (backward + forward)
   • Depth of analysis
   • Date range covered

2. Seminal Patents (ranked by influence) For each hub patent:

   • Publication number, title, applicant, priority date
   • Forward citation count
   • Technical contribution (from abstract)
   • Position in the technology genealogy

3. Technology Evolution Timeline

   • Chronological mapping as described in Step 7
   • Key branching points and sub-fields
   • Recent trends (what's being built on now)

4. Applicant / Competitive Landscape

   • Top citing applicants (who is building on this technology)
   • Top cited applicants (who laid the foundation)
   • Collaboration patterns (co-citation clusters)

5. Network Visualization (text-based)

   • Citation flow diagram showing key patents and their relationships
   • Use arrows (→) to show citation direction
   • Group by technology era or sub-field

6. Observations and Insights

   • Technology maturity assessment (growing, plateau, declining based on citation trends)
   • White spaces (areas with few citations — potential opportunity)
   • Key inventor lineages (research group evolution)

Important Notes

• Forward citations via search: The ct= CQL operator is the only way to get forward citations. It returns patents that cite the specified document. This is a search_patents query, not a dedicated endpoint.
• Citation counts grow over time: A 2023 patent will have fewer forward citations than a 2010 patent simply due to age. Normalize by publication year when comparing influence.
• Examiner vs applicant citations: European patents distinguish between citations added by the examiner (more relevant) and applicant (broader). US patents include all applicant-disclosed prior art, inflating backward citation counts.
• NPL citations matter: Non-patent literature citations (journal articles) often point to foundational scientific work. They can't be followed via patent tools but should be noted in the report.
• Patent families: Multiple publications of the same invention (EP, US, WO versions) are different documents but represent ONE invention. Use get_patent_family to deduplicate when counting unique inventions in the network.
• Rate limits: Each forward citation search is a separate API call. For networks with many nodes, prioritize the most-cited patents rather than exhaustively expanding every node.
• NEVER characterize a patent's subject matter without retrieving its biblio. Do not describe backward citations based on training knowledge — always call get_patent_details first. This is the #1 source of errors in citation network analysis. Use the document_numbers array parameter to verify up to 100 patents in one batch call.
• NEVER describe companies from training knowledge. Do not add phrases like "a pioneer in mRNA technology" or "a leading gene therapy company" — these are hallucination-prone. Only state what the patent data shows (e.g., "applicant on 5 patents in the network").
• Distinguish family members from forward citations. Before including a patent in the timeline, verify it is not in the same family as an already-listed patent by checking shared priority claims. EP divisionals, US continuations, and national phase entries of the same PCT are family members, not independent citations.