Troubleshooting Phrase Match AI rules that aren't rating submissions the way you expected
Duplicate Values
You might have terms entered that the algorithm considers to be duplicates. The Phrase Match algorithm:
- Conjugates verbs
- Lets learners insert 5 tokens (aka Words) into their answer
and still matches it to the term you entered. What does this look like in practice?
Term You Entered | Acceptable Conjugations | Acceptable Added Tokens |
Thank calling | Thanks for calling |
Thank you for calling Thanks so much for calling |
Charge card end now |
Charging card ending now |
Charge the card that ends in 5454 now Charging the card ending in 5454 right now |
What this means for you when writing Phrase Match rules:
- Do not enter terms that are the exact same phrase with one different word in them. The AI thinks these are duplicates and will struggle to know which term is hitting.
- Do not enter terms that are the exact same phrase in different tenses (e.g. "raise your rate" and "raised your rate"). Pick one tense and let the algorithm conjugate the verb for you!
"Thank for calling," "Thank you for calling" and "Thanks for calling" are duplicate terms when using Phrase. Just use "Thank calling"
Sneaky Stop Words
Stop words are common, trivial words like "the," "an," "for" that get filtered out by AI/NLP systems when analyzing text and they're throwing off your phrase match rules!
Bright recommends excluding prepositions and adjectives that aren't critical to the rule. Here's a full list of common stop words that we recommend you use as sparingly as possible:
- a, an, the
- to, for, from, by, with
- in, on, at, of, off
- this, that, these, those
- etc., e.g., i.e.