Plan requirement
| Subscription | Suite Professional or higher, Support Professional or higher |
| Access | Admin |
Defining what to extract and every way customers name it. The synonym list is where this succeeds or fails, and corrections afterwards do not feed back.
Add one
- In Admin Center, click AI in the sidebar and open the entity settings.
- Add an entity and name it.
- List its values.
- Add the synonyms for each value.
- Watch what gets extracted from new tickets.
Get the synonyms from tickets, not from marketing
Search your tickets for a product and read how customers refer to it. You will find the abbreviation your team uses internally, the name from two versions ago, three misspellings, and a description by colour.
That list is the entity. The official product name is one entry in it.
Include the wrong versions
Common misspellings, missing hyphens, the name run together. Customers type quickly on phones, and an entity that only matches the correct spelling misses a substantial share.
Cover each language
Synonyms are language-specific, and a list built in English extracts nothing from Dutch tickets. Where you support several languages, each needs its own entries. Language classification itself covers roughly 150 languages, so the constraint is your synonym list rather than the detection.
Watch for values that overlap
Two products with similar names, or a synonym that belongs to both, produce confident wrong extractions. Where that happens, make the synonyms more specific rather than adding more of them.
Review after a launch
A new product arrives with no synonyms and gets extracted as nothing, or as its predecessor. Adding entity values belongs in the launch checklist alongside the article.
A change does not reach back
An entity is applied when the conversation happens, so editing it today leaves yesterday's tickets as they were. Testing a new definition against existing examples therefore proves nothing, and this is where people conclude the entity is broken when it was not there yet.
It bites hardest on entities that match a pattern rather than a list of values: if the pattern does not fit the way your identifiers are written, it will never match anything, old or new.
Check what is being missed
Tickets that clearly mention a product and have no entity are your synonym gaps, and reading ten of them fills most of them.
Comments
0 comments
Article is closed for comments.