KNIME Documentation

Summarize Product Reviews ​

In this tutorial, you will learn how to summarize long product reviews using a large language model in KNIME.

You will build a complete workflow from start to finish, beginning with raw review data and ending with concise summaries that can be reused in downstream analysis or reporting.

What you will build ​

You will create a workflow that:

  1. Reads product reviews from a file
  2. Builds one prompt per review
  3. Sends the prompts to a language model
  4. Stores the generated summaries in a new table column

By the end, you will have a working workflow that can be reused and adapted for new review data.

Before you begin ​

Make sure you have:

Step 1: Read the review data ​

Use the CSV Reader node to load the product reviews into your workflow.

  1. Download product_reviews.csv

  2. Add a CSV Reader node to your workflow.

  3. Configure the node to point to the downloaded file.

Each row in the table represents one customer review.

Step 2: Create one prompt per review ​

Use the Expression node to build a prompt dynamically for each review.

Configure the node to create a new column called Prompt with the following expression:

text
string(
  "You are a product quality expert.\n" +
  "Summarize the following review in 10 to 15 words:\n\n" +
  $["Review"]$
)

This creates one prompt per row by combining:

  • a fixed instruction
  • the review text from the table

Show prompt column

Expressions in KNIME

The Expression node uses the KNIME Expression Language to construct strings and transform data. If you are not familiar with the syntax, you can ask KAI, the KNIME AI assistant, to help you build or adapt the expression.

Step 3: Authenticate with the LLM provider ​

Authenticate with your chosen provider using:

Connect the credentials node to the authenticator using a flow variable. If authentication is successful, the authenticator node turns green.

Authenticate OpenAI

Step 4: Select a language model ​

Use the LLM Selector node to choose the model that will generate the summaries.

Typical settings include:

  • model name
  • maximum response length
  • temperature

For consistent summaries, consider setting the temperature to a low value.

Select model

Step 5: Prompt the model ​

Use the LLM Prompter node to send the prompts to the model.

Configure the node to receive as input:

Each input row is prompted independently, and the model response is stored in a new column (for example, Response).

LLM Prompter configuration

Result ​

After execution, the output table contains a new column with the generated summaries.

Each row now includes:

  • the original review
  • a short summary generated by the model

LLM Prompter results

If you want to review or compare your result, you can open the completed workflow:

➡️ Open the completed workflow on KNIME Hub

Next Steps ​

If you want to keep learning about AI and data workflows in KNIME, take one of our free courses and earn a microcredential badge to showcase your skills: