Australian-Text-Analytics-Platform/ldaca-wordflow: Wordflow 0.7.12

Abstract

Wordflow 0.7.12 makes Quotation Runs several times faster and shows quotations as they are found; makes AI annotation safer and cheaper, keeping labels if a run stops and sending each text only once; lets you name topics in Topic Modelling; adds Map values to the Data Editor; lets you move result table columns by dragging their headers; shows passages that fit no topic as Ungrouped; and adds suggestions when a result or run looks off. Highlights Quotation Runs several times faster, with results as they come: Run uses several processor cores and shows quotations while it works. Stop keeps what was found as a result you can add to the Project, and Resume carries on from there, also after a restart. AI annotation that doesn't lose or repeat work: labels already returned are kept if a run stops, repeated texts are sent once, Run sends only rows without a label by default, and Preview pages you have seen aren't sent again. A slow request no longer fails the whole run. Local AI models work better: a Requests at once setting (2 for Custom providers) stops local servers timing out, and Thinking off now really turns thinking off for models such as Qwen, which is much faster. Name your topics: give Topic Modelling topics their own names; they appear in the topic list, the map and the Data Blocks you add to the Project. Ungrouped passages are shown: passages that fit no topic appear as Ungrouped instead of being left out, and giant topics are split more readily. Map values: a new Data Editor tool makes a column from a value you type for each value of a column, such as grouping parties into sides. Move table columns: drag a column header's grip to reorder Concordance, Quotation and Annotation results; downloads and Add to Project keep the order. Suggestions and calmer hints: a short suggestion appears when a result or run looks off, hints no longer block the page, and tabs are on by default. Concordance and Quotation Text mode uses no tokeniser: in Concordance's Text mode the tokeniser now shows None and is not used, also after you've tried Tokens mode. Excel downloads with long documents: Excel cells hold at most 32,767 characters, so a long document stopped an Excel download of Concordance or Quotation results. The document column is still ticked and marked recommended, but you can now untick it to download the other columns; its character positions (such as CONC_start_idx and CONC_end_idx) are left out with it, and the window reminds you that the results can't then be checked against the original text. Select none leaves the document ticked. The same applies to Add to Project. Opening a CSV in Excel: the downloaded CSV is complete and correct, but Excel splits text longer than 32,767 characters into extra rows when it opens the file, so the columns no longer line up. Leave the document column out, or open the CSV in another program. Hints, tabs and suggestions Turning hints off shows where they come back: a hint's Turn off hints… no longer turns hints off on the spot. It points at the Settings gear at the top right, then Settings points out Guidance and the Show contextual hints option, which you untick yourself, so you know where to turn hints back on. The Settings gear, then the Guidance tab, then the option pulse in turn, so you can see where to click next, and a click anywhere else (or Escape) closes the pointer if you'd rather carry on. Tabs on by default: each analysis tool now shows its tabs, so you can keep several analyses side by side; click + for another. A hint at + offers Turn off tabs…, which shows the way to Settings → General → Enable multi-tab step by step, and the + button's tooltip names the setting too. If you have already chosen a setting, it is kept. Suggestions: when a result or run looks off, a short suggestion with a lightbulb appears beside it, and the settings that may help pulse while it shows: a Concordance Preview that found nothing in the documents it checked, a Topic Modelling result with one huge topic or only a few topics, a Topic Modelling run still reading the text after five minutes, or a very large input before a run. Suggestions never stop you and need no closing; they fade as you carry on (typing doesn't count), or click Dismiss to close one now, and they come back when it happens again. If the settings a suggestion points at are scrolled out of view, they scroll back into view. Buttons that lose work, such as Clear and Stop, are named but never marked. Clicking a disabled Run outlines what it is waiting for, such as the tokeniser in Frequency. Turn them off under Settings → Guidance → Show suggestions. Hints no longer get in the way: a hint no longer dims or blocks the page. It outlines what it points at, and pulses when it points at a button to click; clicking that button does its job and closes the hint. The next hint waits until you have done something. Clearer Topic Modelling notes: the first-run time is now given for a fast recent computer, with a reminder that older or slower computers often take several times longer. When a suggestion gives the advice, the notes keep to the facts. Help tooltips no longer get in the way: a help tooltip closes as soon as the pointer leaves its icon, and the Switch Project tooltip no longer covers the Settings gear. Tables Move columns: in Concordance, Quotation and Annotation results, drag the grip at the left of a column header to move the column, for example a metadata column before the left context or between L1 and R1. Clicking the rest of the header still sorts. The tab remembers which columns you show and their order, and Add to Project and downloads keep that order. The analysed text in Show metadata: Concordance offers the analysed text column in Show metadata for Run results too, unticked at first. Find columns by name: when Show metadata lists many columns, a box at the top finds them by name. Narrower result columns: the grey labels under Concordance and Quotation column names are now a few words, such as L1 frequency; hover over a label for the full explanation. Export Double-click adds a Data Block: in Export, double-clicking a Data Block in the Project graph or the sidebar adds it straight away, without a second click. Data Editor Map values: a new Add column tool that makes a column from a value you type for each value of a column, for example to group parties into sides or correct spelling variants, with no regular expressions. Values are listed with their row counts, most common first or A to Z ignoring case; values you've typed are offered again so several values can share a group, and Fill the rest keeps the original value wherever you haven't typed one. Delete columns shows examples: each column in the Delete columns window is followed by its first value, so columns with unclear names, such as those in a survey export, are easy to recognise. A column with no values says (empty). Help Ungrouped text: the Topic Modelling help now explains why much of a corpus can be Ungrouped, what a document's main topic means, and why a theme every document shares does not become a topic. An informal benchmark: an appendix compares Max tokens 128 and 256 on one corpus. It is one ad hoc comparison for reference, not guidance. Topic Modelling Name your topics: give a topic its own name, such as Sleep instead of Topic 12, with the pencil on its card in the topic list or beside the examples title, or by double-clicking that title. The name shows in the topic list, the examples title and the bubble's tooltip, and named bubbles have a bold, underlined label. Names stay with their topics when you move the Topics slider: a merged topic shows which named topics it includes, and moving back brings the names back. Add to Project writes the names into a new TOPIC_top1_name or TOPIC_topic_name column, and the interactive HTML download shows them too. Latest selection on top: the topic you select last now goes to the top of the Selected group, instead of being sorted by size. Ungrouped instead of No topic: passages the analysis could not place in any topic are now called Ungrouped, with a grey bubble in the corner of the map and a grey entry in the topic list, so documents with no topic are no longer left out without notice. You can add the Ungrouped documents to the Project like a topic. A line under the results title gives the number of documents, segments and topics, and how many documents are in a topic or Ungrouped. When many documents are Ungrouped, a suggestion tells you which settings to change. Fewer giant topics: when one topic covers most documents, Auto max topic size now splits it even if part of it becomes Ungrouped, instead of keeping the giant topic. Suggestions explain what a lower or higher Max topic size does, with a Learn more link to the help. Colour by follows category order: after you make a column an ordered category, the Colour by legend uses that order straight away, instead of keeping the order it had before. Colour by explains itself: when no column has 2 to 8 values, or the result comes from two Data Blocks, Colour by no longer disappears. It stays on Data Block colour, greyed out, and hovering over it tells you why. Quotation Quotation Runs are several times faster: Run now uses several processor cores at once (half of those available, up to 8) and checks each distinct text once. On a 26,000-document corpus and an 18-core Mac it takes about 6 to 7 minutes instead of 27; a 2-core BinderHub session runs as before. See quotations as they are found: while Run works, Quotation Results so far shows the documents checked so far, in order, and updates every few seconds. Add to Project and downloads wait until the Run finishes or is stopped. Stop keeps what was found: Stop finishes the few documents being checked at that moment (a few seconds) and shows everything found as the result, marked Stopped early, so you can add it to the Project or download it. Choose