Open Forum

Hi all - just wondering what peoples feelings are around the QM/SA transcription elements at the moment. A while back we had issues with transcripts not picking up the correct wording but after an upgrade/change of the AI engine that 8x8 uses it did appear to get better. However, the more I look at call transcripts the more I think it is getting worse. Here's a few bullet points of issues i see daily, be interested to see if anyone else has similar issues...

1. Poor call tagging. Many calls show as #Complaint. However when going to that section of the call the transcript show the agent used the phrase, "That's no problem". As the word "Problem" is in the Complaint topic it has tagged it which is incorrect.
2. Issues surrounding transcription. Not sure if it doesn't like our Yorkshire accents but many calls I listen to and then read the transcription does not match up. Lots of errors.
3. Incorrect emotion allocation. Many calls that are classed as "Overall Negative Emotion", really aren't negative when listening to them. It seems really sensitive in terms of how it evaluates emotion.

Any feedback/opinions appreciated.

Thanks

We have had the same issue and are working with 8x8 support to resolve this. This is a recent response i have received: 

Speech Analytics sentiment in 8x8 QM is automatically calculated from the full call transcript using an internal word-based algorithm and not just the phrases or topics you configure. Because the system relies on a broader built-in dictionary and pattern detection, calls can be marked negative even when none of your visible negative phrases are matched. Neutral sentiment does exist but is hidden in the interaction view, and any post-call evaluations do not change the emotion result.

& 

We received an update from our Applications Team. Please note that this behavior is expected because you are using a wide search by adding * at the end of the phrase.

For example:


disappoint*

annoy*

irritat*


Even though the phrase "ridiculous" was not added to the topic "Caller Negative Emotion," other phrases with * will match similar words or variations.

To avoid this behavior, you will need to remove the * so that Speech Analytics (SA) matches only the exact word instead of similar words or meanings.

Hi all - just wondering what peoples feelings are around the QM/SA transcription elements at the moment. A while back we had issues with transcripts not picking up the correct wording but after an upgrade/change of the AI engine that 8x8 uses it did appear to get better. However, the more I look at call transcripts the more I think it is getting worse. Here's a few bullet points of issues i see daily, be interested to see if anyone else has similar issues...

1. Poor call tagging. Many calls show as #complaint. However when going to that section of the call the transcript show the agent used the phrase, "That's no problem". As the word "Problem" is in the Complaint topic it has tagged it which is incorrect.
2. Issues surrounding transcription. Not sure if it doesn't like our Yorkshire accents but many calls I listen to and then read the transcription does not match up. Lots of errors.
3. Incorrect emotion allocation. Many calls that are classed as "Overall Negative Emotion", really aren't negative when listening to them. It seems really sensitive in terms of how it evaluates emotion.

Any feedback/opinions appreciated.

Thanks

We have had the same issue and are working with 8x8 support to resolve this. This is a recent response i have received: 

Speech Analytics sentiment in 8x8 QM is automatically calculated from the full call transcript using an internal word-based algorithm and not just the phrases or topics you configure. Because the system relies on a broader built-in dictionary and pattern detection, calls can be marked negative even when none of your visible negative phrases are matched. Neutral sentiment does exist but is hidden in the interaction view, and any post-call evaluations do not change the emotion result.

& 

We received an update from our Applications Team. Please note that this behavior is expected because you are using a wide search by adding * at the end of the phrase.

For example:


disappoint*

annoy*

irritat*


Even though the phrase "ridiculous" was not added to the topic "Caller Negative Emotion," other phrases with * will match similar words or variations.

To avoid this behavior, you will need to remove the * so that Speech Analytics (SA) matches only the exact word instead of similar words or meanings.

Hi all - just wondering what peoples feelings are around the QM/SA transcription elements at the moment. A while back we had issues with transcripts not picking up the correct wording but after an upgrade/change of the AI engine that 8x8 uses it did appear to get better. However, the more I look at call transcripts the more I think it is getting worse. Here's a few bullet points of issues i see daily, be interested to see if anyone else has similar issues...

1. Poor call tagging. Many calls show as #Complaint. However when going to that section of the call the transcript show the agent used the phrase, "That's no problem". As the word "Problem" is in the Complaint topic it has tagged it which is incorrect.
2. Issues surrounding transcription. Not sure if it doesn't like our Yorkshire accents but many calls I listen to and then read the transcription does not match up. Lots of errors.
3. Incorrect emotion allocation. Many calls that are classed as "Overall Negative Emotion", really aren't negative when listening to them. It seems really sensitive in terms of how it evaluates emotion.

Any feedback/opinions appreciated.

Thanks