Select Page

Natural Language Processing is a difficult topic in the deep learning area. For example, if we want to look at the sentiment of thousands of customer reviews without any coding knowledge, the Alteryx sentiment Analysis tool is a perfect way to do it.

In this blog, I will show how to use this tool, and more importantly, I will introduce how the VADER algorithm behind it works to proceed with a reliable result.

This is the sentiment analysis tool. All the input is a column of customer review. In this case, I use the data from the web scripting and JSON parsing. The configuration window of this tool is quite simple.

The Algorithm is called VADER – it measures the valence and magnitude of emotion in the text. The feeling of the text refers to whether the comment is negative or positive. On the computer side, the negative and positive field is defined by the Compound Sentiment Score.

So, the question is, what is Compound Sentiment Score? In the VADER algorithm, each comment’s score is between -1 to 1. If a comment is exceptionally Negative, the score is -1. In contrast, if a comment is hugely Positive, the score is 1. But there is a middle ground there when the comment is not too bad or good. It would be Neutral.

So, what’s the range score value of these three terms? Based on the countless experiment of many talented people who developed this algorithm, the maximum negative value is -0.1 and the minimum positive value is 0.5. So, this range is from -1 to -0.1, which is negative. From -0.1 to 0.5, is neutral. A value larger than 0.5, is positive.

Let’s look at some examples from real comments from the datasets. For this sentence, we can see many good words here, from our human eyes, we can see this comment is positive. And the Algorithm showed the score of this sentence is 0.79. which is quite good.

Also, these comments here showed up some bad words, we can see it’s negative. And the score of it is -0.8. Match our thoughts.

Last, we can see some good words here, and some bad words here. Which is neutral, and the score is very close to 0.

From these examples, we can see how powerful this tool is, and it helps us to identify the sentiment easily and reliably.