Skip to content

Latest commit

 

History

History
133 lines (97 loc) · 5.52 KB

File metadata and controls

133 lines (97 loc) · 5.52 KB

Glossary

Domain terms that appear in the data, the schema or the UI and may not be immediately obvious to a non-specialist reader.

Bibliographic sources

DBLP. The Digital Bibliography & Library Project — a computer science bibliography from the University of Trier, indexing journals, conferences, books and other publications since 1993. Source for the two fact_*_articles tables. Site: https://dblp.org.

Scimago / SJR. Scimago Journal Rank — a journal-level prestige metric based on the eigenvector centrality of citations, weighted by the citing journals' own SJR. Higher is better, no upper bound. Backed by Elsevier Scopus data. We obtain SJR plus eight more metrics per journal from a Kaggle dump.

iCore / CORE2026. The Computing Research and Education Association of Australasia (CORE) maintains a public ranking of computing conferences. Tiers run A*, A, B, C and Multiconference from top to bottom. Site: http://portal.core.edu.au/conf-ranks/.

Kaggle. Public dataset platform. The Scimago dump we ingest was published there: "Journal Rankings" by Khaled-El-Sayed.

Schema patterns

Lookup table. A dimension table that contains the conformed, unique entities of one type. Example: lookup_journals is the single-version-of-truth for the 1 423 journals in the corpus.

Fact table. A table whose rows represent individual events or measurements at the lowest grain of analysis. Example: fact_journal_articles — one row per article ever published in any indexed journal.

Bridge table. Resolves an N-to-N relationship between a fact row and a dimension. bridge_journal_article_authors connects an article to each of its authors with a composite primary key.

Star schema. Kimball dimensional model: a small number of fact tables surrounded by dimension/lookup tables. The opposite of a snowflake schema (further normalisation of dimensions). Optimised for analytical reads.

Wide dimension. A dimension table that absorbs the attributes of the entity it represents — even if some attributes are sparsely populated — rather than splitting them across normalised sub-tables. Trade-off: faster reads, slightly larger per-row storage.

Materialised table. A regular CREATE TABLE whose contents are populated from a SELECT over views. Acts as a cached snapshot of an expensive aggregate. Read-only between rebuilds. The pattern used throughout sql_scripts/03_views.sql for the heavy aggregates.

Rejection log. Quarantine table for rows that fail validation during ETL. Captures source file, source id, reason and the raw row as JSON, so failures are visible and recoverable.

Ranking metrics (in lookup_journals)

Best quartile. Categorical Q1Q4 ranking of the journal relative to all journals in its best_subject_area. Q1 = top 25 %.

Best subject area. The subject category where the journal has its strongest position (highest quartile). Each journal can appear in several subject areas; the "best" is what the UI uses.

H-index. The largest integer h such that the journal has published at least h papers cited h or more times. Resists single-paper outliers; favours sustained productivity.

Cite score. Average citations per article over the last 4 years.

Total docs / total docs 3y. Number of indexed documents ever / in the last 3 years.

Total refs. Total references contained in articles published in the journal.

Total cites 3y. Total citations the journal received in the last 3 years.

Citable docs 3y. Documents counted as citable in the 3-year window (research articles, reviews; excludes editorials, errata, etc.).

Cites per doc 2y. Citations per article in the last 2 years — the building block of Journal Impact Factor.

Refs per doc. Average reference count per article. Higher in review-heavy journals.

Ranking metrics (in lookup_conferences)

Acronym. The short form (e.g. EDBT, VLDB) used by DBLP as booktitle. Acronym → iCore is the strict entity-resolution edge.

Rank value. The iCore tier (A* to Multiconference).

Primary FoR. Australian Field of Research code that classifies the conference's topic. The code is opaque (e.g. 4605) but each one has a human description (e.g. "Distributed and parallel computing"). The lookup_field_of_research_categories table maps codes to descriptions.

API and front-end concepts

RFC 7807 Problem Details. Standard JSON shape for HTTP error responses (application/problem+json). Fields: type, title, status, detail, instance. Used by every error path of the API.

Stale-while-revalidate (SWR). Caching strategy where a stale cache entry is served immediately while the fetch refreshes it in the background. TanStack Query implements this with staleTime and gcTime.

Visx. Airbnb's collection of low-level visualisation primitives that wrap D3 scales, shapes and axes as React components. Trade-off vs Recharts/Chart.js: more code to write, but pixel-perfect control.

Voronoi nearest-x overlay. Hover-detection layer that maps each mouse position to the nearest data point in X without forcing the cursor onto a tiny marker. Implemented in LineChart.tsx.

Drag-to-zoom. Brush interaction where dragging across the X axis selects a new view window. Implemented client-side; the data array is sliced, not re-fetched.

Path / search params. TanStack Router treats route parameters (e.g. $journalId) and URL search parameters (?start_year=…) as typed, schema-validated inputs. Filters live in the URL, so every page state is shareable.