Thematic theme creation

Thematic creates themes using it's unassisted NLP AI, which is then refined by Thematic data scientists, and the data set stakeholders.

The initial set of themes created by Thematics algorithm are very specific and cover a wide variety of topics. Each theme contains mapped phrases that are used to tag responses.

Example mapped phrases for "Good seats"

  • amazing seating - I really enjoyed the airline's amazing seating.
  • fantastic seats - I recommend this airline because of the fantastic seats.
  • seats are great - The seats are great and the staff are polite.

To tidy the themes Thematic will merge two themes together which it believes have a similar meaning (eg. "Good experience" and "Great time") and will combine their mapped phrases. Each comment can be tagged by more than one theme, but never the same theme twice.

flight delays is an example theme with no delays and delayed us as merged themes

Thematic then tries to group the themes into a 2 level hierarchy of similar topics (eg. Base theme "People", with sub-themes "Friendly", "Grumpy" or "Rude")

Example base theme with it's sub-themes sorted by Thematic

This list of themes is then sense checked by a data scientist at Thematic and then handed to the data set stakeholders, who can make more refinements using the in portal theme editor. The types of refinements typically include

  • Removing unhelpful or unactionable themes
  • Merging similar themes together
  • Creating new themes and moving themes under them
  • Adding contextual understanding to the themes (eg. Users rarely say UI, but a UI base theme can be useful for analysis)
  • Adding or removing existing mapped phrases to tighten up the accuracy of the themes

Thematic can also run more theme discovery when new data is loaded, or if a specific topic is desired. These new themes with typically be put into a holding theme until a data set stakeholder can double check them, and then move them into the existing hierarchy.