How to configure a summarizer?

Different prepackaged summarization components (extractive and abstractive) are available to compute a short summary of a longer piece of text.

Abstractive Summarization

The algorithm creates an entirely new text, based on the content of the document. Large Language Models are very well fitted for this task.

  • Go to the Processing view
  • Create a new pipeline
  • Select Off-the-shelf component in the drop-down list
  • Select a cloud provider of Large Language Models (LLM)
  • Select the LLM
  • Edit the prompt and write: “Summarize in english the following text: $text
  • Give a name to the “Completion Altext” to store the summary in this metadata field.
  • Save your pipeline
  • Test it in the Annotation test view as above. See How to test a model?

Extractive Summarization

An algorithm identifies the most informative sentences from a text and brings them together to form the new summary. This approach is quick and easy, but does not generate new sentences which may result in incoherent texts.

  • Proceed as above
  • Select “textranksummarizer” component
  • Adjust available parameters with the settings icon.
  • Give a name to the Alternative text (Alttext). The system will generate an additional text field called “ExtrSummary” that holds the newly generated summary.
  • Save the pipeline