Automatically extract key points and themes from text

We can learn huge amounts from customer comments, emails and chat.

However, wading through text – takes time, often more than you have available.

  • It can get repetitive, leading to boredom and inaccurate coding of responses as volume increases

  • It can also lead to humans unintentionally giving higher weight to things they’ve read at the beginning of their analysis or things they’ve read more recently

Fortunately, Text Analytics / NLP has made unbelievable progress over the last few years, driven by:

  • Availability of text data

  • Deep Learning

  • Increasingly sophisticated language models which take word context and word order into account

  • Pre-trained models and transfer learning

  • Computing power which can handle all the above

NLP is now a very viable solution for a range of business applications. It won’t replace the need for human interpretation, but will lead to increased speed and consistency of analysis, thanks to:

  • Speech recognition

  • Keyword extraction

  • Auto-summarisation

  • Topic Modelling / Clustering

  • Automated coding / classification of text

  • Sentiment detection

Archery Board

Keyword Extraction

Keyword extraction shows you the key words and multi-word phrases used in text, so you can quickly understand what points are being discussed.
Keyword analysis has moved on beyond simple word counts, using:


  • AI models to identify parts of speech 


  • Weightings to make rare words specific to the document stand out


  • Network models to understand how keywords are linked semantically (e.g. customer service, excellent)


  • Algorithms to measure the importance of words in a document


Automated Summarisation

When Google returns search results, it shows you the most relevant results based on your query.

Auto-summarisation works in a similar way, picking out the most representative bits of text from your documents. 


This means you can access a quick headline, analyse text faster, and more reliably distinguish the core story from the noise.

Colorful Tulips

Topic Modelling / Clustering

Somewhat similar to clustering of numerical data, we can also cluster text (e.g. open-responses, blogs etc) into topics based on shared meaning.


This also makes it possible to size, drill into, or visualise different clusters to understand which topics different audiences discuss.