I am currently evaluating Genesys Cloud’s ‘Real-Time Sentiment Analysis’ for a large RFP. The client is asking for a guaranteed accuracy percentage for sentiment detection in several languages (English, Spanish, and French). I see that the platform provides a sentiment score for each interaction, but I cannot find any official documentation on how this score is calculated or its verified accuracy rate. Has anyone performed an independent audit of the sentiment accuracy, and are there specific acoustic factors (like background noise) that significantly impact the results?
Hey Cla34! I am a junior Python dev and I have been working on a script to pull these sentiment scores into our custom reporting dashboard. From my experience, the accuracy is quite good for English, but it struggles with ‘Sarcasm’ or very technical jargon. We found that the score is heavily influenced by the ‘Confidence’ of the transcription engine. If the transcript is messy because of background noise, the sentiment score will be all over the place. I do not think there is a ‘Guaranteed’ percentage because it depends so much on the audio quality!
Greetings. I am a telecom engineer and I deal with the voice paths for our global sites. To follow up on Cam70, the sentiment analysis is performed on the transcript, not the raw audio. So any network issues like packet loss or jitter that cause the transcript to fail will directly impact your sentiment accuracy. If you are putting this in an RFP, you should specify that the accuracy is contingent on a ‘MOS Score’ of at least 4.0 for the voice connection. Without good audio, the AI is just guessing!