Introduction
Hi, I am Akira, the editor-in-chief of Data Without Code. In our previous Use Case tutorial, we solved a massive supply chain problem by automating our inventory safety stock calculations. We are now officially experts at processing numbers.
But numbers only tell you what happened. They don’t tell you why it happened. If sales of your flagship product suddenly dropped this month, the answer isn’t in your SQL database—it is out there on the internet. It is in customer reviews, YouTube comments, and social media posts.
If you have 10,000 product reviews or tweets about your brand, you cannot read them all manually. As a DX manager, I teach my team to use Sentiment Analysis. This is a technique that automatically reads text and scores it as Positive, Negative, or Neutral.
You might think you need a PhD in Python and Natural Language Processing (NLP) to do this. You do not. In this tutorial, I will show you how to perform social media sentiment analysis in KNIME without writing a single line of code.
Step 1: Prepare Your Text Data
First, you need data. This could be a CSV file downloaded from your customer survey tool, or raw text scraped from a website. (If you don’t know how to get web data, check out my guide on web scraping basics using KNIME.)
Let’s assume you have a dataset with a column called “Customer Review” containing thousands of text entries.
Step 2: Convert Strings to Documents
In the Data Prep module, we used the String Manipulation node to clean short text. But for advanced text analysis, KNIME needs to treat the text as a formal “Document.”
To do this, we need the KNIME Textprocessing extension. (If you haven’t installed it yet, read my guide on installing KNIME extensions.)
- Search for the Strings to Document node and connect it to your raw data.
- Double-click the node. In the configuration, select your “Customer Review” column as the Title/Text source.
- Execute the node. Your raw text is now converted into a formal Document object that KNIME can mathematically analyze.
Step 3: The Dictionary-Based Approach (No Code Required)
There are many ways to do machine learning sentiment analysis, but the absolute easiest way for beginners is the Dictionary-Based Approach.
This simply means giving KNIME two lists of words (dictionaries). One list contains positive words (like “amazing”, “love”, “great”), and the other contains negative words (like “terrible”, “hate”, “broken”).
KNIME will scan every single customer review and count how many positive or negative words it finds!
How to Tag the Words
You can easily find free sentiment dictionaries online and load them into KNIME using the CSV Reader node.
- Add a Dictionary Tagger node to your canvas. Connect your Document data to the top port, and your list of Positive Words to the bottom port.
- Configure the node to tag any matching words as “Positive”.
- Add a second Dictionary Tagger node right after it, but this time connect your list of Negative Words to the bottom port. Configure it to tag matches as “Negative”.
Now, every word in your 10,000 reviews has been scanned and tagged. No coding required!
Step 4: Count the Score
The final step is to count the tags. We use the Bag of Words Creator node to extract all the words, and then the TF (Term Frequency) node to count them.
Finally, we summarize the results using our favorite aggregation tool: the GroupBy node. Group the data by “Document” and sum up the number of Positive and Negative tags.
If a review has 5 Positive tags and 1 Negative tag, the overall sentiment is overwhelmingly Positive. You have just built a fully automated sentiment scoring engine!
Conclusion: Your Next Steps
Congratulations! You have successfully unlocked the power of unstructured text analytics. You can now connect this workflow to a live data source, calculate the daily sentiment score of your brand, and send an automated email alert to your PR team if the negative score suddenly spikes.
We have optimized marketing, supply chain, and customer success. But what about the internal heartbeat of your company—your employees?
If you want to know who is likely to quit before they actually hand in their resignation, you need HR analytics. Are you ready to predict human behavior? Join me in our next Use Case: Employee turnover prediction: HR analytics with KNIME!
