Dissertations

University of Chicago — Dissertation Explorer

1893–2025

Visualizing 130+ years of UChicago Dissertation Work

Data Visualization Fellowship 2025–2026  ·  Jasmine Lu  ·  Andrew McGallian  ·  Khushi Desai

The University of Chicago has kept a dissertation archive going back to 1893. This project uses that archive to explore three questions: how have dissertation topics and departmental activity shifted over time; how have the geographic research interests of UChicago scholars moved across the world; and what do advisor and committee relationships reveal about how knowledge and academic lineage are structured here. The approach combines spatial analysis at the campus and global scales, NLP on keywords and subject terms, and network analysis on departments, advisors, and dissertation topics.

The goal is to translate a large institutional dataset into an interactive portrait of academic knowledge production—one that can support institutional reflection, serve as a reference for prospective faculty and students, and offer a lens on how the definition of "PhD-worthy" research has shifted in response to national priorities, technological change, and the emergence of new fields.

Data sources

ProQuest TDM Studio is the foundation with a full export of all 32,013 UChicago dissertation records, 1893–2025, flagged for PhD dissertations. Each record includes title, date, authors, advisors, committee members, subject terms, and paper keywords. The department field is only populated for records filed after 2009. Earlier records also suffer from missing fields, inconsistent use of keywords and descriptions, duplicate or citation-only entries, and occasional mistagging (records filed under the wrong institution or degree type).

UChicago Convocation Programs (1932–2009) fill most of the department gap for pre-2009 records. The programs list PhD graduates by school/division and, most of the time, department. Compiling 22,993 records from fall, winter, spring, and summer convocations across 77 years required manual review and cross-referencing dissertation titles and author names. Limitations include typos, inconsistent department naming conventions across eras, incomplete or ill-preserved program images, and cases where a dissertation does not appear in any convocation record.

HathiTrust Digitized Catalog (1893–1931) covers the earliest era that convocation programs do not reach. The source is a set of OCR-processed catalog documents organized by division and department, parsed with a custom script. Limitations include OCR errors from the digitization, inconsistent document formatting across years, and name changes (marriage name changes in particular) that complicate matching to ProQuest records.

UChicago Heritage Map provides the campus building geometry with polygon footprints for university-owned buildings with historical metadata including year built. This is the source for all campus-level spatial analysis. The main limitation is that the map does not record which departments occupied which buildings over time; that routing was reconstructed manually from cited department histories, facilities records, and library exhibits.

Cleaning and normalization

The department field alone required substantial normalization work. Across all three sources, more than 300 unique raw department strings appear — historical names, abbreviations, all-caps convocation format, committee names, division-level labels, and renames that accumulated over 130 years. All of these are mapped to 44 canonical modern departments in an explicit lookup table, with historical names preserved in the department-history cards. Records that cannot be reliably assigned a canonical department are marked UNKNOWN rather than forced into a misleading category.

Matching ProQuest records to convocation programs uses title + year as the primary key, with a 40-character prefix fallback and author surname as a last resort. Only unambiguous single matches are accepted. Once a canonical department is assigned, each dissertation is routed to the building that department actually occupied at the time of completion, using a year-keyed building history for each of the 44 departments.

Geographic research focus is extracted by running spaCy NER (en_core_web_lg) over dissertation titles, paper keywords, and subject terms — no abstracts were available. A 200+ entry normalization table handles historical states (USSR → Russia, Ottoman Empire → Turkey), ancient entities (Babylonia → Iraq), city-to-country mapping (London → United Kingdom, Beijing → China), and modifier stripping (Ancient Egypt → Egypt). Keyword similarity across departments is computed with TF-IDF vectorization and pairwise cosine similarity, producing a 44×44 similarity matrix used by the ring diagram and network views.

Establishing the building history for each of the 44 departments was its own research effort, separate from the data cleaning work. The Heritage Map supplies building polygons but does not record which departments occupied which buildings or when, so each department's physical history had to be reconstructed from scratch. Primary sources included department websites with "Our History" or "About" pages, university directories, library exhibits, facilities records, and archived administrative pages. The range of difficulty varied widely: some departments have well-maintained written histories; others — particularly discontinued programs like the Graduate Library School (1926–1989) and Home Economics (1918–1956), or departments with multiple relocations like Astronomy and Astrophysics (Ryerson → Astronomy and Astrophysics Center → Eckhardt Research Center) and Economics (SSRB → Saieh Hall)—required piecing together sources that weren't written with this question in mind. Historical department names were tracked alongside building timelines, so the same research that produced the routing table also produced the historical names, notes, and source citations visible in the department cards in the Dept Histories tab.

The dataset

The pipeline produces a shareable 15-column CSV with canonical department names, historical campus building assignments, and geographic entities. The pipeline is fully deterministic and all normalization decisions live in explicit lookup tables in the source code, making the output reproducible from the raw inputs.

32,013
dissertation records
132
years of coverage
44
canonical departments
96.9%
dept + building coverage
4,068
geo-enriched records

Features

  • Campus Choropleth — building polygons colored by dissertation count, with a year slider and animated playback. Departments are routed to their actual building at each point in time. Click any building to explore its dissertations.
  • Global Research Focus — countries shaded by mention count in dissertation titles, keywords, and subject terms. Decade toggle and a line chart trace how global research interests have shifted.
  • Keyword Similarity — a ring of all 44 departments arranged in clockwise campus order, with chord edges scored by TF-IDF cosine similarity. The layout is conceptually inspired by ring species: adjacent departments share vocabulary the way neighboring species share genes. Includes a force-directed network, distance vs. similarity scatter, and a time series of mean pairwise similarity by year.
  • Network Analysis — clustering departments, keywords, advisors, and committee members to surface interdisciplinary connections across the dataset. In development.
  • Department Histories — a Gantt chart of all 44 departments colored by academic division, with sourced building timelines and detailed department cards including historical names, addresses, and citations.
  • Advisors & Committee — testing whether a small number of advisors produce a disproportionate share of PhDs (a "Matthew Effect"), visualized as a Sankey diagram faceted by department. In development.
  • Dissertation Explorer — search by author name, keyword, or topic content with filtering by year and department. In development.
  • ORCID & Post-Graduation — where do graduates go after UChicago? How much do they publish? What academic placements do they take? In development.
  • Download Data — the cleaned CSV, README, license, and citation guidance.

Known limits

  • No abstracts — the ProQuest export did not include abstracts, so all topical and geographic analysis relies on titles, paper keywords, class terms, and subject terms. Coverage is real but partial — a dissertation on medieval France whose title gives no geographic signal will not appear in the global map.
  • Historical departments are messy — names change, committees become departments, programs are dissolved or folded into others, and some records list only a division. All of those decisions are documented in the Dept Histories tab.
  • Building history is evidence-based — locations are routed using cited sources, but where records are sparse or sources conflict, judgment calls were made and are noted.
  • Some early records cannot be placed — bureaucratic changes, incomplete convocation programs, and OCR errors in HathiTrust mean a portion of pre-1932 records remain UNKNOWN on department and building.

Year

2025
— dissertations

Color scale

0
—

Dissertations assigned to building in selected year.

Top — all years

    Dissertations over time

    All assigned dissertations by year. Click a bar to jump to that year.

    Methods & Sources
    Department assignment — three-source pipeline

    31,019 of 32,013 records (96.9%) carry a mapped modern department and campus building geometry. Departments were assigned from three sources, applied in priority order:

    1. ProQuest Dissertations & Theses Global — 6,310 records (2009–2025) have departments directly from ProQuest. These are the only records where department data was entered at source; pre-2010 ProQuest records were retrodigitised with title and date only. Source: ProQuest PQDT Global, exported via TDM Studio.
    2. UChicago Convocation Programs, 1932–2009 (convocation_data.xlsx) — 22,993 records matched to ProQuest by title+year (case-insensitive, whitespace-collapsed). 97% match rate by exact title; 40-character prefix fallback recovers an additional ~363 records where titles differ in punctuation or subtitle handling. Column width varies by era: 9-column rows (pre-~1990) store department in column 8; 10-column rows (~1990–2009) store it in the last column. We always read the last column and fall back to the second-to-last (division) when the last is empty. Source: University of Chicago Library dissertation and convocation office archives.
    3. HathiTrust Digitized Catalog, 1893–1931 (hathi_1893-1931.csv) — 2,617 records with department and division fields. Strategy A (title match): 1,499 records matched by title+year (exact or 40-char prefix). Strategy B (author+year match): when title strategies fail, the student's surname is extracted from the ProQuest author field ("LASTNAME, FIRSTNAME" format) and matched against HathiTrust's "FIRSTNAME LASTNAME" field for the same graduation year. Only unambiguous single matches are accepted; ties (two authors with the same surname in the same year) are silently discarded to prevent misassignment. This recovered an additional 869 records. Source: HathiTrust Digital Library digitized catalog (hathitrust.org), via UChicago Library.

    Remaining unknowns (994 records, 3.1%): Primarily records where no supplementary source matched, or where the source listed only a division (e.g., "Division of Social Sciences") too broad to assign to a specific department. The Graduate Library School (141 records, 1926–1989) is mapped to Harper Memorial Library (1926–1969) then Joseph Regenstein Library (1970–1989). Home Economics (72 records, 1918–1956) is mapped to 5740 S. Woodlawn Ave from 1929, confirmed by CEHD building history. Radiology and medical-school clinical departments are excluded — they were on the medical campus outside the academic quad covered by this visualization.

    Department normalization & building history
    All 44 canonical departments — historical names, building timelines, judgment calls, and sources — are in the Dept Histories tab → Click any bar to open the full department card.

    Year

    All
    — dissertations with geo focus

    Color scale

    1
    —

    Top countries

      Shifts over time — top 6 countries

      Dissertation count by decade, all available years.

      Methods & Sources
      Geographic NER pipeline

      Geographic research focus is extracted using spaCy en_core_web_lg Named Entity Recognition on each dissertation's Title, Paper Keywords, and Subject Terms. GPE and LOC entities are collected, then normalized to ISO country names via a 200+ entry alias table covering historical state names (USSR → Russia, Persia → Iran), major cities (London → United Kingdom, Beijing → China), ancient entities (Babylonia → Iraq, Ottoman Empire → Turkey), and UK constituent nations. US states and US cities are excluded to avoid inflating the US count from domestic references. Each country is counted once per dissertation (deduplicated).

      No abstract field: ProQuest did not include abstracts in the export. Coverage is limited to dissertations where the title or keywords name a place — a structural limitation of the TDM export. 4,068 of 32,013 records (12.7%) have a detectable geographic entity. Top countries: United States (379), China (239), France (166), UK (165), India (148).

      Sources

      Keyword Similarity Ring

      How to read this diagram
      How to Read: Ring Diagram
      Nodes Each dot is a department. Order around the ring is a shortest-distance campus tour: departments in the same building are always adjacent, then buildings are connected by geographic proximity. Minimizing total tour length is equivalent to ensuring each building neighbors its closest buildings in the ring. Color = academic division (legend below). Ring edges Arcs connect consecutive ring neighbors. Color and thickness are normalized to this ring's min/max similarity range so the full scale is always visible: red = weakest, blue = mid-range, green = strongest. A red arc between nearby buildings suggests spatial proximity without much shared vocabulary in that period. Hover chords Mouse over a node to reveal maroon chords connecting it to every department it shares keyword overlap with, not just its ring neighbors. Chord thickness = cosine similarity. The tooltip shows ring-neighbor scores and the single strongest non-adjacent connection. Ring break The reddest, thinnest arc marks the weakest consecutive link in the displayed ring. The breakpoint label below names both departments and gives their similarity score and distance. Year slider Recomputes the ring for the selected year using only departments with enough keyword data in that year to build a similarity matrix. The tour order, edge colors, and breakpoint are recalculated for that yearly network. Sparse early years may not have enough data to draw a full ring. ⇌ Campus Morphs ring positions into actual building locations on the underlying campus map — lets you verify that the tour order matches real geography and that the ring break occurs between genuinely proximate buildings.

      Year

      All time
      Edge similarity weak
      strong

      Similarity Network

      Force-directed graph: nodes cluster by keyword similarity. Edge weight = cosine similarity. Drag nodes freely. Color = academic division. Switch to Scatter for distance vs. similarity.

      How to read · Analysis · Methods & Sources
      How to Read: Right Panel Views
      Scatter Axes X = haversine distance between building centroids (meters, ~50–2,000 m). Y = cosine similarity of TF-IDF keyword vectors (0–1). A downward slope would indicate proximity correlates with similarity. Dot color Same-division pairs use their division color; cross-division pairs are grey. Network Nodes / edges Each circle is a department (color = division). Lines connect pairs with cosine similarity ≥ 0.08; thickness scales with strength. Drag nodes freely. Node set mirrors the active ring year. Hover Highlights a node's edges in maroon; tooltip shows its three most similar departments and scores. Time Series Line Mean pairwise cosine similarity across all department pairs per year. Rising = vocabulary converging across campus; falling = increasing specialization. Gold dot marks the active ring year. Hover for exact values.

      Top Keywords by Department

      Select a department to see its TF-IDF top terms (all-time corpus).

      Methods

      Keyword corpus construction For each department, all dissertations assigned to it contribute their Paper Keywords and Subject Terms fields from the ProQuest record. These are concatenated into a single keyword document per department (pool-of-words model). Title text is excluded — titles are too short and noisy for reliable topic signals. Departments with fewer than 30 total keyword terms are excluded from the ring. For per-decade matrices, the threshold is 15 terms.
      TF-IDF vectorization and cosine similarity Each department's keyword document is vectorized with scikit-learn TfidfVectorizer: up to 800 features, English stop words removed, minimum document frequency 2, unigrams only. TF-IDF down-weights terms that appear in almost every department (common academic vocabulary) and up-weights terms distinctive to a specific field. Pairwise cosine similarity between the resulting unit vectors produces a symmetric similarity matrix (up to 44×44, filtered to depts with sufficient keywords). Values range [0, 1]. Per-year matrices use the same method but restricted to dissertations from that year (300 features, min_df=1 due to smaller corpora). Decade matrices are also generated for summary views and historical comparison.

      Drives: ring edge color/width · hover chord thickness · scatter Y-axis · network edge weight · ring breakpoint · keyword display · yearly network · time series line
      Spatial distance and campus tour order Building positions are the mean centroid of each building's departments, derived from uchicago_locations.csv. Pairwise distance is the flat-Earth approximation in meters (accurate to <0.1% at campus scale). Campus range: ~50 m (Zoology Building ↔ Anatomy Building) to ~2,000 m (Art History ↔ Public Policy).

      Ring order minimizes total tour length using three stages. First, departments sharing a building are grouped as atomic units so they always appear consecutively. Second, a nearest-neighbor heuristic is run from every building as a starting point and the shortest result is kept — this avoids the greedy trap of a single starting choice locking in a long return jump. Third, 2-opt (reverse a segment when it shortens the tour) and Or-opt (relocate a single building to a cheaper position) alternate until neither can improve. Or-opt escapes local minima that 2-opt alone cannot, since it moves buildings across the ring rather than just reversing segments. Minimizing total distance is equivalent to ensuring each building neighbors its geographically closest available buildings — the prerequisite for the ring-species vocabulary test.

      Drives: ring node order · scatter X-axis · breakpoint distance label · campus morph layout · force simulation collision radius
      D3 force graph simulation The network layout uses D3 v7 forceSimulation with four forces: a link force attracting connected pairs with strength proportional to cosine similarity and distance inversely proportional to similarity (so highly similar departments pull together tightly); a many-body repulsion preventing overlap; a centering force; and a collision radius of 16 px. The simulation runs until convergence and remains live — dragging a node reactivates it. Only pairs above the similarity threshold (0.08) are drawn as edges. The year shown matches the year slider; the node set mirrors the active ring departments.
      Data provenance 31,019 / 32,013 records (96.9%) carry a mapped department and building geometry, filled from three sources: ProQuest (6,310 direct), UChicago Convocation Programs 1932–2009 (~22,993 matched by title+year), and HathiTrust Digitized Catalog 1893–1931 (~2,617, with author+year fallback). Some individual years remain too sparse to build a yearly keyword matrix; those years appear as gaps or reduced-node rings rather than silently falling back to all-time values. Full pipeline detail and department cards are in the Dept Histories tab →
      Data sources

      Network Analysis

      Methods & Sources
      What the nodes and links mean Each node is a school/division or department which you can click to explore further. Departments are normalized to their modern day counterpart departments (see tab department histories, but you can click to explore more to see the history of their naming conventions.
      What the cards include Clicking a node opens more detailed information about each node (dissertation, department, school, etc.) allowing you to further analyze networks between dissertations/departments/divisions.

      Click any node to see details.

      Department Timeline

      Methods & Sources
      What the bars mean Each bar segment is a department-building occupancy span from DEPT_HISTORY. The bars are not clipped to dissertation volume; sparse years do not shrink the historical housing span. Dissolved units end at their documented active end.
      What the cards add Clicking a department opens addresses, dissertation counts, historical names, notes/context, and row-level citations for every building move.
      Source policy Top department links point to current or historical department pages. Building-history citations support department location, founding, relocation, or continued occupancy claims.
      Pipeline connection The same building spans are mirrored in GIS/prepare_viz_data.py for historical campus routing. GeoJSON aliases handle cases like Walker Hall/Walker Museum and the Astronomy and Astrophysics Center/WERC site.

      Click any bar to see department details.

      Advisors & Committees

      This section explores advising structures within the University of Chicago dissertation archive. The visualizations highlight divisional growth and identify the most prolific dissertation advisors.

      Cleaned Dissertation Dataset

      32,013 University of Chicago dissertation records (1893–2025), cleaned and enriched with canonical department names, historical campus building assignments, and geographic research focus. Produced by the Data Visualization Fellowship 2025-2026.

      32,013
      dissertation records
      15
      columns
      96.9%
      dept coverage
      4,068
      geo-enriched records
      ⬇ Download ZIP (CSV + README + License)

      Includes: dissertations_clean_share.csv · DATA_README.md · LICENSE.txt

      Dataset Methods & Reproducibility
      Source assembly ProQuest supplies the base metadata. Convocation programs fill most 1932–2009 department gaps, and HathiTrust catalog records fill much of 1893–1931. Matching uses title+year first, then conservative prefix or author/year fallback when needed.
      Cleaning choices Raw department strings are preserved, then normalized into 44 canonical departments. Ambiguous division-only records stay UNKNOWN rather than being forced into a misleading department.
      Enriched fields campus_building comes from historical department-building routing. geo_entities and geo_primary come from spaCy NER plus country normalization. Keyword fields are retained for independent analysis.
      Reproducibility The pipeline is deterministic: normalization tables, building spans, country aliases, and visualization outputs are all explicit in the repository rather than inferred at runtime.

      Columns

      Column Description
      goidProQuest unique ID
      titleDissertation title
      dateYear of completion (4-digit)
      authorsAuthor(s)
      advisorsAdvisor(s)
      committee_membersCommittee members
      degreeDegree type (e.g. PhD)
      dept_rawOriginal department string from whichever source filled it
      dept_modernNormalized canonical department (44 values + UNKNOWN)
      campus_buildingBuilding assigned via historical routing — where the dept was at time of writing
      keywords_authorAuthor-assigned keywords (ProQuest)
      subject_termsProQuest controlled vocabulary subject terms
      class_termsProQuest classification terms
      geo_entitiesSemicolon-separated countries detected via NER (12.7% of records)
      geo_primaryPrimary detected geographic focus country
      License
      Released under Creative Commons Attribution 4.0 International (CC BY 4.0). Free to share and adapt for any purpose with attribution.

      Data Provenance
      Dissertation metadata originates from ProQuest TDM Studio, UChicago convocation programs (1932–2009), and HathiTrust digitized catalog (1893–1931). Enriched fields (dept_modern, campus_building, geo_entities, geo_primary) are original contributions of this project.

      Citation
      Jasmine Lu, Andrew McGallian, Khushi Desai. Visualizing University of Chicago Dissertation Topics (1893–2025). Data Visualization Fellowship, 2025-2026. github.com/xjasminelu/uchicago-dissertations-viz