Sentiment analysis

A sentiment is a feeling or judgment of something and in the field of analytics, sentiment analysis is a type of language or text-processing algorithm that will break down parts of a text grouping into at least three categories: positive, neutral, and negative.

In Power Query, the sentiment analysis capability will evaluate a text input and return a score. For example, if you have a query of customer survey results, there might be a free-text field where customers write down their thoughts about your retail organization. The survey takers will complete the survey and might respond with text such as “This was a wonderful shopping experience” or “You didn’t have the product I wanted.” When the survey results are compiled, these free-text values will be stored in a text field and the sentiment analysis feature can decide whether the sentiment of the text entered is positive, negative, or neutral. If there are 500 survey responses, then it’s possible to use this capability combined with other aggregation and transformation capabilities to determine the sum of positive, negative, and neutral results based on the free-text field alone.

The sentiment analysis capability takes the entire input field as input into the pretrained algorithm so, as with any text analytics process, it’s important to understand how this works, as there may be blocks of text such as “At first, I didn’t find the product I wanted but there was a very helpful associate who showed me where the product was and helped ensure I had everything I needed.” In this example, there are parts that could be interpreted as negative and others that could be interpreted as positive; if analysis is performed on the entire block of text, then it’s likely to have a positive sentiment.

Sentiment scoring will return a decimal value between 0 and 1, where negative sentiment can be interpreted as being between 0 and 0.5 while positive sentiment is 0.5 and greater. When sentiment is detected as neutral, a value of 0.5 is returned.

It’s also important to note that support for this feature varies by language. Currently, English, German, Spanish, and French are supported, with others in preview.