This guide will walk you through how Lumoa measures Sentiment, and the typical use cases.
Use cases
You can use Sentiment analytics in the following use cases:
Gain insights into customer emotions and opinions, and use this information to improve your customer experience.
Track changes in customer sentiment over time.
Track sentiment for each topic.
Ultimately, all the above use cases, help you monitor the sentiment of your data, and understand how your customers are feeling about your products, services, and brands.
Sentiment in Lumoa
To see Sentiment in Customer Feedback, follow these steps:
Click on the Setting gear button.
Navigate to the Impact page.
Under Feedback, click on the upper left Arrow, and open the feedback.
Hoover over the Statement, to see whether the Sentiment is positive, negative, or neutral.
How is Sentiment measured?
The sentiment model uses AI to detect whether the sentiment of each sentence is negative, neutral, or positive. The model uses the RoBERTA architecture and benefits from seeing data from the CommonCrawl corpus. The model is further trained with examples of Lumoa’s own data and other data sources.
You can manually update the Sentiment from the feedback, and help train the model by correcting incorrect sentiment in the UI.
The model splits comments into sentences before it does predictions.
The model looks at a single sentence when it makes the prediction as opposed to the whole comment.
In addition, if there is a score attached to the feedback the score can be used to help detect irony. This process is called "sarcasm detection". For example, if a user comments "solid quality!" and gives 0 points, it will not mark it as positive.
How is Sentiment Score measured?
The model outputs a sentiment prediction of negative/neutral/positive and a sentiment score [-1,1]. The sentiment scores are derived from the outputs of the sentiment model.
For neutral the score is always 0.
Scores for negatives and positives can be interpreted as how sure the model was about its choice. E.g. if the score is -0.999, then the model was very sure that the sentence was negative. And if the score is 0.5, the model thinks the sentence was positive, but it was less confident about it, since it is not very close to 1.
Get in touch
📧 Do you have any questions or comments about using Lumoa? Please don't hesitate to email us at [email protected].