Text Classifier
Sort a long list without reading every line
On it. Running Text Classifier now and I'll bring the results straight back here.
Try asking
About
Text Classifier looks through a pile of snippets, tickets, reviews, or log lines so you do not have to read every one.
Tell it what you want. It keeps the useful rows, puts a label on each item, or just counts how many fall in each bucket. If it is not sure, it keeps the row instead of throwing it away. It returns a label and a confidence score, not a written summary.
Use this when you have dozens of items. If only a handful are already on screen, decide yourself.
What you can do
- filter — drop the junk and keep the useful bits
- classify_many — put one label on each row
- tag — apply every label that fits, not just one
- count — get totals, not a row for every item
- classify, check, score, evaluate — the same idea for one piece of text
Related tools sort search results, check messages for harm, route support tickets, pick an email or amount, check a citation, compare options, match a need, say what to do first, approve a queue item, and spot duplicates.
Who it's for
Anyone sifting search results, tickets, reviews, or logs who does not want to paste every line into chat.
How to use it
- Start with filter or count when the list is long
- Use classify_many or tag when each row needs a label, then only read the unsure ones
- Pass a file_id if the list lives in a file
- Use classify, check, or evaluate for a single item
Getting started
Call filter with your question and the texts. Read only the rows it kept.
Pick one label for a piece of text
Put one label on each item in a list of up to 100
Keep the texts that help answer your question
Apply every label that fits, not just one
Pick one label for a piece of text. Returns the winner, how sure it is, and whether nothing fitted.
Put one label on each item in a list of up to 100. Returns the label and how sure it is, so you need not reread the list.
Keep the texts that help answer your question. Unsure items stay in the list so useful rows are not dropped.
Apply every label that fits, not just one. Use this when a row can be several things at once, such as topics on a ticket.
Label a list and return totals. Use this when you need counts, not a row to read for every item.
Ask one or more yes/no questions about the same text. Each answer is a score you can act on, not a written verdict.
Rate text on a scale you write, such as calm to angry. Returns a score that can land between levels, plus how sure it is.
Ask several questions about one piece of text in a single call: pick a label, yes/no, or a rating.
- Rerank, screening, and ticket triage now live in their own TypeSafe tools
- Added filter, classify_many, tag, and count for labelling piles of text without reading every row
- Initial release with classify, check, score, evaluate, rerank, screen, and support triage
Powered by
TypeSafeRelated
