Data format
What to send SymNexusExtract
The inputs the platform expects, and the labels it learns from.
Input
Input format
Plain text: one message, document section or post per row (CSV) or per line (JSONL).
| id | text |
|---|---|
| T-1001 | I was charged twice for my Pro plan, please refund it. |
| T-1002 | The dashboard crashes when I open reports. |
id,text
T-1001,"I was charged twice for my Pro plan, please refund it."
T-1002,"The dashboard crashes when I open reports."For training
Label format
For categories: a label per text. For details: the words that make up each detail, marked by type (B- starts a detail, I- continues it, O is everything else).
| text / tokens | label / tags |
|---|---|
| I was charged twice… | billing |
| ["Call","Maria","Garcia"] | ["O","B-PERSON","I-PERSON"] |
Tips for good results
- A few hundred labelled examples per category is a strong start.
- Keep categories distinct and well-defined.
- Label details consistently, e.g. always include both first and last names.