What cartolex does, in its own words

cartolex turns the publications of a group of people into a map of their field. This page is the vocabulary you need before your first project: each word as the app uses it, what it means, and the screen where you meet it. Ten minutes of reading; the tutorials then put each word to work.

A project

A project is one map and everything it is made from, kept in a folder of your computer: the people, their texts, your decisions, the results. You create, open and switch projects from the Projects screen (the project’s name at the top left of every screen, then « All projects… »). The demo project is an invented community of about forty people in coastal and marine sciences, with texts in English and French: try anything on it, nothing can break.

When you create a project you give it a name, the field’s title (what it maps, in a few words, as a specialist would say it), a short description (read by the AI, if you use one, to judge what belongs to the field), the languages of the texts and a reference language: keywords of every language are gathered under their name in that language.

The Overview is the project’s home: its Next step always names the one thing to do now, and the checklist « Your first map » shows the whole way (see Your first map, with the demo project).

People

The map is made of people and of what they write. The People screen lists them, with three things for each.

  • Role: what the person is to the map.

    • Mapped: their texts draw the map, and they are on it. Most people are.

    • Context: their texts help find the keywords of the field, but they are not on the map (the co-authors of your mapped people, for instance).

    • Projected: placed on the finished map by their texts, without moving it (newcomers, applicants, another team you want to compare).

    • Excluded: left out; Undecided: not chosen yet.

  • Identity: whether cartolex knows who the person is in the bibliographic services (OpenAlex, the ORCID registry, HAL). To check: candidate records wait for you on the Identities tab; Confirmed; Accepted automatically (a single clear match, still marked for review); No record. Several people with the same name are common: nothing is collected for a person until their record is confirmed.

  • Coverage: whether enough texts were found. Good (three texts with an abstract or more), Thin, Failed (a service failed: retry it) or No data. The Coverage tab counts them and says why.

People come in by Import (a list of names, a folder of documents, a prepared corpus) or by Collect (the people of an institution, the co-authors of your people). Rows that may be one person are proposed on the Duplicates tab, two or more at a time; people you know to be one are merged with Same person…; a merge can always be undone. A person’s sheet (Enter on a row, or « Open the sheet ») says what was found for them and why.

Texts

A text is a publication: its title, its abstract when there is one, and sometimes its full text, in any of the project’s languages. The Texts tab lists them. The harvest (Collect › Harvest texts) gathers the works of the people whose identity is confirmed; the same work found twice (the same DOI, or the same title, author and year) counts once. Documents you import (PDF or text files) are texts too.

Collecting is the only step that sends something over the network, and only to open bibliographic services. Before it starts, a notice says what leaves the computer (names, identifiers), to which service, and what never does; it starts only when you have read it (Privacy and personal data).

Organisations

Organisations are where people work: a team, a laboratory, a department, an institution. They come with the people’s affiliations (each with its years) or from a column of your list. They sit on levels (a lab inside an institution), can have several parents (a joint unit), and can be renamed or merged on the Organisations tab. On the map, an organisation sits at the average place of its mapped people.

Keywords

cartolex finds the keywords in the texts themselves; nothing comes from an outside list of terms.

  • A candidate is a phrase built like a technical term (sediment transport, trait de côte), used by at least three people and by at most 60 % of them.

  • Each candidate falls in a band, with the reason shown beside it: Kept (a phrase of several words), To check (a single word, often too general), Set aside (a fragment of a longer phrase, a grammatical word, a word everyone uses), or Rejected automatically (a phrase that is never a keyword in any field).

  • Your decisions: Keep, Exclude, Merge (two spellings, or a term and its translation, become one keyword). Every decision can be undone from the History.

  • The AI clean-up (optional) judges the candidates and gives each accepted one a category (a concept, a method, an object of study, a place, a field) and its name in the reference language, so that a term and its translation become one keyword.

  • The lexicon is what the last build kept: the keywords the themes and the map use, on the Final lexicon tab of the Lexicon screen, with a word cloud.

How keywords emerge explains each rule in plain words.

Themes

The keywords are grouped into themes: keywords that the same people use together end up in the same theme. The themes form a tree of one to four levels (themes, then topics inside them, for instance). The grouping only proposes a tree: on the Themes screen you rename, move, merge, split, create and set aside, with undo and redo, then save (a new version, which you can restore later) and apply (the map is drawn again with your themes). A theme’s share is the part of all keyword use that counts toward it.

The map

The Map screen places on one picture:

  • the people: two people are close when they use the same keywords in the same proportions. Each point takes the colour of the theme that weighs most in their texts;

  • the keywords, near the people who use them;

  • the organisations, at the average place of their mapped people;

  • the texts, at the average place of the keywords found in them;

  • the projected people, and the time windows (a person’s texts over a few years, joined by a line in time order).

The treemap beside it shows the themes and their shares; the panel shows what is selected. The distances on the map are a picture: the similarity shown in the panel (from 0, nothing in common, to 1, the same use of the vocabulary) is measured in the space the themes are drawn from, and is the number to quote. A map version records how the map was drawn: the pinned one is the one every build redraws, so the map people know does not move.

The build

Every result is computed by the build, in stages: gather the texts, find keyword candidates, AI clean-up, build the vocabulary, place keywords in a common space, group keywords into topics and themes, apply your themes, draw the map, change over time, place projected people. Each stage is Never built, Up to date, Needs update (something it read changed: the line says what), Running, Failed (its previous results are kept) or Skipped. A build runs only what needs it.

Build… on the Overview opens the Build screen: before anything runs, it says what will run and why, the time and memory it should take, and asks how the AI steps are done. A build runs in the background: the Activity button at the top right follows it and every collection, with a Stop button. Each screen has a Tune panel (« Tune the keywords », « Tune the map »…) with the settings of its step; the Recipe tab of the Build screen lists them all, with what differs from the defaults. Building a project says more.

The AI steps

Two steps can use an AI, and both work without one. For each, the Build screen asks how, and the project remembers:

  • No AI: the keywords are kept by their statistics, the themes by the grouping; you review them by hand.

  • With your copilot: cartolex gives you a bundle (one zip file) for an AI assistant that can run code. It holds the candidate keywords with counts, and people as numbers in a random order: never a text, a name, an identifier or an organisation. The assistant judges with cartolex’s own tools and gives back a result.json; you import it and accept what you want. No key, and cartolex itself sends nothing (Clean the keywords with an AI copilot).

  • By API (keyword clean-up only): cartolex sends the keyword strings, with the field’s title and description, to the AI provider set in the settings, with your key. It asks first, and the provider bills the calls; an answer already paid for is never asked again.

Sharing

The Share screen builds an offline site: a folder that opens in any browser, without a server or an internet connection, with the map, the themes, a page per person and per organisation, and the method. At each build you choose names or pseudonyms for the people, and whether it carries the titles or the abstracts of the texts; checks list what to look at before sending it. The same screen gives the map as an image, the theme table and other files (Share a site).

Where things are kept

Everything stays in the project’s folder on your computer: what was collected (sources/), what you decided (decisions/, with every earlier version), what was computed (derived/), and what was shared (outputs/). Nothing is kept elsewhere except the app’s own settings and keys, on this computer (The project format, cartolex-project/1 for the details).

Next