I wonder if anyone can share any information on topic strictness please? I cannot see anything in the Resource Center on it at all and am keen to understand how strictness is defined. It is clearly on a scoring scale, but an explanation of this measurement would really help.

Thanks

Heather

Hi Heather. 

It is basically, tool's recognition confidence level, of how sure it is that the word transcribed is actually word that was spoken. 

For Example: you set your strictness to 50 for "buy food"; the only time the system is going to add tag for this phrase if it is at least 50% confident that the caller said "buy food". If it is only 49% confident, the call will not get tagged for the phrase. 

Majority of mine are set to 65, unless those some difficult words/phrases that normally get transcribed incorrectly. 

Hope this helps.

Larissa

I wonder if anyone can share any information on topic strictness please? I cannot see anything in the Resource Center on it at all and am keen to understand how strictness is defined. It is clearly on a scoring scale, but an explanation of this measurement would really help.

Thanks

Heather

Hi Larissa and Heather,

Topic strictness is not related to the transcription confidence filter we have in the Speech and Text Analytics Admin settings. 

Topic strictness indicates how strict the matching algorithm is when matching phrases in a transcript. The matching algorithm looks at phrase confidence after stemming to determine if the phrase is a match. 

In practice, the way we recommend customers to determine the strictness is by trial and error.  Starting at some default, we examine the events captured by this phrase, and then increase or decrease as necessary.

One thing to note, longer phrases that have more content words (words that aren't stop words) require lower strictness and vice versa – for example if you have only 3 content words in a phrase you can decide on a strictness of Medium which means that you need to capture two of the three words, but customers sometimes prefer to match all three words (High). 

If you have (say) 5 content words, then many times 3 of them as enough for a good match, and then Med-Low would be the best tradeoff between precision and recall, but this is dependent on the phrase.

Hope that makes sense.

Hi Heather. 

It is basically, tool's recognition confidence level, of how sure it is that the word transcribed is actually word that was spoken. 

For Example: you set your strictness to 50 for "buy food"; the only time the system is going to add tag for this phrase if it is at least 50% confident that the caller said "buy food". If it is only 49% confident, the call will not get tagged for the phrase. 

Majority of mine are set to 65, unless those some difficult words/phrases that normally get transcribed incorrectly. 

Hope this helps.

Larissa

I wonder if anyone can share any information on topic strictness please? I cannot see anything in the Resource Center on it at all and am keen to understand how strictness is defined. It is clearly on a scoring scale, but an explanation of this measurement would really help.

Thanks

Heather