1 · Venues

A venue is one review cycle: its submissions, its reviewer pool, an assignment, and a recruitment pipeline. Create one from the Venues page with the default number of reviewers per paper (K) and papers per reviewer (L); both can be changed later on the Assignment page. The two demo venues ship with real accepted-paper lists: the VLDB 2026 research track (PVLDB vol. 19 papers in the conference programme) and the EDBT 2026 demonstration track, each with a reviewer pool drawn from the expert database.

2 · Data

Submissions — CSV or OpenReview JSON with sub_id, title, abstract, authors, emails, affiliations, countries, track. Countries are inferred from affiliations when not given. Reviewers — JSON with publication histories, or one-row-per-publication CSV, with optional country, max_load and expert_id (the DBLP-based id from the expert database, which enables authorship checks). Conflicts — declared pairs as CSV; Derive authorship conflicts adds author and co-author matches from names, emails and publication lists. Sample files are linked on each page.

3 · Assignment

Matching scores come from one of two routes, and the same options apply to both:

• Import existing matching scores — rows of sub_id, rev_id, score (higher is better), e.g. from bidding or a previous cycle. TaxoReview only solves the assignment.
• Match and assign from scratch — TaxoReview embeds every text with SPECTER2, builds a local weighted taxonomy over the submissions and each reviewer's evidence publications (automatic or chair-picked), computes taxonomy affinities and solves. Scores are shown on a 0–100 scale.

Options: reviewers per submission in a range; submissions per reviewer in a range (per-reviewer caps override the maximum); a balance threshold so every assigned pair scores at least the given value; country diversity so that no paper is reviewed only by reviewers from its authors' own country (papers or reviewers without a known country are never blocked); and no co-authors, which excludes reviewers who co-authored with a submission author according to their publication lists and, for reviewers linked to the expert database, its authorship graph. The problem is solved as a mixed-integer program; locked and manual pairs survive re-solving, vetoed pairs stay out. The result page reports single-country panels, the lowest assigned score, load spread and papers left below the minimum.

On a paper's page you can lock, remove, or add reviewers, declare a conflict, and see each candidate's score and supporting publications.

4 · Reviewers in the wild

The recruitment page is backed by a standalone snapshot of the CS expert-profile database restricted to the database field (databases, data management, data mining and information retrieval): 269,256 experts, 249,104 papers in CCF A–C / ICORE A*–C venues, snapshot 2026-09-22. An expert's history holds their retained ranked-venue papers, not every publication; counts are evidence of activity, not a quality ranking.

• By domain — experts are organised by the venue-derived research domains of the database itself; each domain lists its active experts with paper, first-author and recency counts.
• Recruitment bar — choose the venues an expert should have published in, then the requirement: at least m papers in those venues in the recent k years as first author (a junior bar) or as any author (a senior bar). Qualified experts can be shortlisted for the recruitment pipeline or added straight to the reviewer pool with their publication histories, after which re-running the matching assigns them.

Identity is the DBLP-based expert id; ORCID selections carry their evidence status (verified, corroborated, reported, conflict, none). Country evidence is not yet part of the published profiles, so recruited reviewers have no country until a chair sets one.

5 · Recruits and exports

The recruitment pipeline tracks shortlisted → invited → accepted / declined; accepted experts join the pool as recruited reviewers. Assignments, coverage and recruits export as CSV from the venue overview.

Method summary

TaxoReview represents a venue by a local weighted topic taxonomy built from the submissions and the reviewers' evidence publications, and scores a paper–reviewer pair by how well the reviewer's evidence covers the paper's topics in that taxonomy. Assignment is a constrained optimisation over those scores. Recruitment looks outside the pool, here through the expert database's own venue taxonomy and publication records.