akhilpandey95/century_elec_engineering

A Century of Electrical Engineering

This project adapts the core move of A century of physics to electrical engineering using sciscinet-v2.

The paper's key idea is not "count papers in a list of journals." It is:

1. Define a field by how papers cite and are cited.
2. Recover the field's broader interdisciplinary literature outside its obvious venues.
3. Study long-run growth, impact, interdisciplinarity, and internal structure.

For electrical engineering, that is the right framing too. If you only count IEEE-style venues, you will miss a large share of the work that shaped the field: optics, communications, control, hardware, semiconductor-adjacent materials, applied physics, computer systems, signal methods in medicine, and papers published in broad or non-EE venues.

Hugging Face Artifact Bundle

Large derived outputs are published as a Hugging Face dataset:

akhilpandey95/century_elec_engineering

Use that dataset as the one-stop artifact bundle for reproducing or auditing the experiments. It includes the code, manuscript assets, generated figures, run logs, IEEE taxonomy input, phase-2 derived tables, taxonomy-conditioned experiment intermediates, and impact-lifetime outputs. It does not redistribute raw SciSciNet-v2 parquet files; restore those separately if you want to rebuild everything from raw data. See docs/huggingface_dataset_manifest.md for the dataset layout and reproduction commands.

Repository

.
├── docs
│   ├── data_skill_sciscinet_v2.md
│   ├── nphys3494.pdf
│   └── requirements.md
├── README.md
└── src
    └── ee_seed_profile.py

About

Build an "extended electrical engineering" corpus from sciscinet-v2, then use it to answer:

• What counts as electrical engineering over the last century?
• How much of EE lives outside canonical EE journals?
• How has EE grown relative to science overall?
• Which sub-areas are self-contained versus field-bridging?
• Which sub-areas produce the highest long-run impact, novelty, disruption, and translational output?
• How have the boundaries between EE, physics, computer science, materials science, and medicine shifted over time?

Recommended Corpus Definition

Do not start from a single field label. Start from a seed bundle, then expand recursively with citation behavior.

Seed EE bundle

Use these sciscinet_fields level-1 concepts as the first-pass seed:

• C119599485 — Electrical engineering
• C24326235 — Electronic engineering
• C133731056 — Control engineering
• C76155785 — Telecommunications

Optional phase-2 extensions, depending on how broad you want the project:

• C49040817 — Optoelectronics
• C120665830 — Optics
• C24890656 — Acoustics
• C9390403 — Computer hardware
• C113775141 — Computer engineering

The seed should be intentionally conservative. The Barabasi paper used a core corpus first, then expanded outwards. Do the same here.

How This Maps to SciSciNet-v2

Tables you need first

• sciscinet_fields.parquet
• sciscinet_paperfields.parquet
• sciscinet_papers.parquet
• sciscinet_paperrefs.parquet
• sciscinet_papersources.parquet
• sciscinet_sources.parquet

Tables you need for the full paper-style version

• sciscinet_paperdetails.parquet or sciscinet_papertitleabstract.parquet
• normalized_citations_level0.parquet
• normalized_citations_level1.parquet
• hit_papers_level0.parquet
• hit_papers_level1.parquet
• sciscinet_authors.parquet
• sciscinet_authors_paperid.parquet
• sciscinet_paper_author_affiliation.parquet
• sciscinet_affiliations.parquet
• sciscinet_link_patents.parquet
• sciscinet_link_nih.parquet
• sciscinet_link_nsf.parquet
• sciscinet_link_newsfeed.parquet
• sciscinet_link_twitter.parquet

Why sciscinet-v2 is a good fit

The corrected v2 schema already contains most of the metrics you would want for a century-scale field study:

• publication year and date
• citation windows: C3, C5, C10
• disruption
• novelty / conventionality
• sleeping beauty metrics
• team size and institution count
• funding links
• patent / news / clinical trial / social links
• field-normalized citations and hit-paper flags

That means you do not just reproduce the 2015 paper. You can build a stronger EE version.

Build Plan

Phase 1: Define the seed corpus

Use sciscinet_paperfields.parquet to collect all papers assigned to the seed EE bundle.

Recommended practical rule:

• keep papers whose fieldid is in the seed bundle
• for stricter membership, require score_openalex >= 0.3 or use the paper's highest-scoring field if you want a single primary label

This phase gives you:

• annual EE output
• first-pass venue list
• first-pass author and institution coverage
• a reproducible base corpus for iteration

Phase 1.5: Add an IEEE venue taxonomy audit

Use the subject-grouped IEEE title definitions from IEEEabrv.bib or IEEEfull.bib as a curated venue taxonomy. This does not replace citation expansion, but it gives the project a stronger EE-native side channel for:

• auditing the mechanically generated canonical venue proxy,
• finding IEEE venues missed by the core-share threshold,
• building an EE-native internal map for venue-linked papers,
• tracking where recursive expansion turns from field membership into a broad influence sphere.

See docs/ieee_taxonomy_experiment_plan.md and run:

python3 scripts/ee_ieee_taxonomy_experiments.py \
  --ieee-bib /path/to/IEEEabrv.bib \
  --source-core-stats data/derived/ee_phase2_2208/tables/source_core_stats.parquet \
  --canonical-venues data/derived/ee_phase2_2208/tables/canonical_ee_venues.parquet \
  --out-dir data/derived/ee_phase2_2208/tables

Phase 2: Recover the extended EE literature

This is the direct adaptation of A century of physics.

For each candidate paper connected to the seed by citation edges:

1. Compute the share of its references that point into seed EE.
2. Compute the share of citations it receives from seed EE.
3. Compare both to a null expectation.
4. Promote papers whose EE linkage is much higher than chance.
5. Repeat until convergence.

In practice:

• candidates come from sciscinet_paperrefs.parquet
• outgoing linkage uses cited_paperid
• incoming linkage uses citing_paperid
• a simple first null model is year-conditioned random mixing
• a better null model conditions on year, doctype, and optionally broad field

Outputs:

• core_ee
• extended_ee
• ee_boundary_candidates

This is the most important step in the whole project. It is what turns the project from "a trend report on papers tagged electrical engineering" into a true field-mapping study.

Phase 3: Growth and boundary figures

Replicate the paper's logic with EE-specific views:

• annual counts of core_ee and extended_ee
• EE share of all science by year
• fraction of EE-like papers outside canonical EE venues
• citation flow between EE and non-EE fields
• reference age over time
• author growth versus paper growth
• team-size growth over time

Recommended joins:

• sciscinet_papers for year and impact metrics
• sciscinet_papersources + sciscinet_sources for venue analysis
• sciscinet_authors_paperid and sciscinet_authors for authorship growth

Phase 4: Internal map of EE

The physics paper builds a map of subfields and studies self-reference and citation barriers. Do the same for EE.

Candidate EE subdomains:

• power and control
• communications and networking
• electronics and devices
• optics and photonics
• acoustics and signal-adjacent methods
• hardware and systems
• EE-materials boundary
• EE-biomed boundary

Concrete implementation:

1. Define an EE subfield panel from sciscinet_fields.
2. Build a citation matrix between subfields from sciscinet_paperrefs.
3. Normalize observed links against expected links.
4. Build:
   • a positive network for over-citation / affinity
   • a negative network for citation barriers
5. Plot self-reference ratio versus subfield size.

This reproduces the most interesting figure class from the paper.

Phase 5: Impact and translational reach

This is where sciscinet-v2 is better than the original setup.

For each EE subfield and decade, compute:

• raw citations: citation_count, C10
• field-normalized impact
• hit-paper rates
• disruption
• novelty and conventionality
• sleeping beauties
• patent linkage
• NIH / NSF linkage where relevant
• news and social visibility if you want public-facing figures

This lets you distinguish:

• fast-burn subfields versus long-horizon subfields
• basic versus translational EE
• novel versus consolidating sub-areas

Suggested MVP

If you want the fastest useful version, do this first:

1. Build the seed bundle.
2. Count yearly EE papers from 1900 onward.
3. Show EE share of all science over time.
4. Show top venues and top adjacent fields.
5. Build one citation-flow figure between EE, physics, computer science, materials science, and medicine.

That is already enough to validate the project direction.

Suggested Full Deliverables

• a clean data/derived/ layer for core_ee, extended_ee, and subfield assignments
• a notebook or script per figure family
• a small static site or report with 6 to 10 polished figures
• a methods appendix documenting the field-expansion algorithm

Concrete File Plan

Suggested repo structure:

century_elec_engineering/
├── README.md
├── scripts/
│   ├── ee_seed_profile.py
│   ├── build_extended_ee.py
│   ├── build_ee_citation_flows.py
│   └── build_ee_subfield_map.py
├── data/
│   └── derived/
├── notebooks/
│   ├── 01_seed_corpus.ipynb
│   ├── 02_extended_corpus.ipynb
│   ├── 03_growth_and_boundaries.ipynb
│   └── 04_internal_map.ipynb
└── figures/

Local Grounding

Using a local SciSciNet copy on this machine, the conservative 4-field seed

• Electrical engineering
• Electronic engineering
• Control engineering
• Telecommunications

already covers:

• 11,458,660 unique papers in union
• about 0.96% of all science papers in 1900
• about 5.43% to 5.74% of all science papers in 2021 to 2024

That is enough scale for a real century-long project.

It also confirms an important modeling choice: EE is not one label. It is a bundle with strong overlap into adjacent fields and venues.

Important Constraint

The local SciSciNet directory available on this machine looks older and narrower than the corrected data_skill_sciscinet_v2.md schema. It contains:

• sciscinet_papers
• sciscinet_paperfields
• sciscinet_paperrefs
• sciscinet_sources
• sciscinet_papersources
• sciscinet_authors
• sciscinet_link_patents

but not the full v2 table set documented in the markdown file.

So:

• you can build the seed corpus, growth curves, citation flows, venue views, and early patent analyses immediately
• you need the full corrected v2 snapshot to do the stronger impact, normalized citation, author-affiliation, and funding analyses

Reproduction Status Against This Local Copy

If the question is "can this local directory reproduce the core logic of nphys3494.pdf?", the answer is mostly yes.

If the question is "can it reproduce every analysis cleanly from one consistent sciscinet-v2 schema?", the answer is no.

What is available now:

• sciscinet_papers.parquet gives paperid, year, date, doc type, and citation-window counts.
• sciscinet_paperrefs.parquet gives the citation graph plus year and ref_year, which is enough for field-expansion, citation-flow, and reference-age analyses.
• sciscinet_sources.parquet plus sciscinet_papersources.parquet is enough to define a core venue set and study papers inside versus outside canonical venues.
• sciscinet_fields.parquet plus sciscinet_paperfields.parquet is enough to define an EE seed bundle and build an OpenAlex-field-based internal map.
• SciSciNet_Papers.tsv contains older paper-level metrics such as Disruption, Team_Size, Atyp_*, SB_B, SB_T, and external-impact counts.
• paper_level.parquet, sciscinet_disruption_alpha.parquet, papers_SB.parquet, and SciSciNet_Link_NobelLaureates.tsv fill some of the same gaps in parquet or TSV form.

What you can recreate well from the local data:

• define a conservative EE core from sources or fields
• expand to an interdisciplinary EE corpus using citation behavior
• measure growth over time and EE share of all science
• measure reference depth and citation-age shifts over time
• measure citation flows between EE and neighboring domains
• build an analogous internal EE map using OpenAlex field labels
• validate some discoveries with Nobel-linked papers or patent links

What you cannot recreate cleanly from the local data:

• the author-growth and per-author productivity panels from the paper, because there is no local equivalent of sciscinet_authors_paperid.parquet or sciscinet_paper_author_affiliation.parquet
• the exact PACS-based subfield assignment used in Fig. 3; you can build an analogous OpenAlex-field version, but not the same PACS workflow
• title- and abstract-driven inspection from local files, because sciscinet_paperdetails.parquet and sciscinet_papertitleabstract.parquet are not present here
• the clean integrated v2 metric workflow documented in data_skill_sciscinet_v2.md, because the local copy spreads metrics across older TSV files and sidecar parquet files
• field-normalized citation and hit-paper analyses from the corrected v2 tables, because normalized_citations_* and hit_papers_* are not present here
• institution, funding, newsfeed, twitter, and author-affiliation analyses from the corrected v2 schema, because those tables are missing locally

Practical conclusion:

• for reproducing the paper's main conceptual move for EE, this local data is sufficient
• for reproducing every panel or for building the stronger "full v2" version described in the markdown doc, this local data is not sufficient

One technical note: the local directory mixes schemas. The newer parquet files use OpenAlex-style IDs like W2789336 and S4306400194, while older TSV files use numeric IDs like 2789336. These can be joined, but only after explicit ID normalization.

What To Do Next

Immediate next steps:

1. Set a canonical SCISCINET_V2_ROOT.
2. Lock the seed field bundle.
3. Run scripts/ee_seed_profile.py on that root.
4. Build build_extended_ee.py around the recursive citation-expansion step.
5. Pick the first three figures and drive the rest of the repo from them.

The central bet of the project should stay the same as the paper:

Electrical engineering is not just what gets published in electrical-engineering journals. It is the broader body of work that electrical engineers build on and that in turn builds on electrical engineering.