e-satisfaction

Topic detection

Topic detection reads your open-text answers and works out what each one is about. It runs quietly in the background, classifying every comment against a set of themes, so you can see which topics come up most often and where your customers' attention concentrates — without anyone reading thousands of comments by hand. You configure it here in the Admin Panel; the results are explored over in Insights. Admin only

How it works

The analysis follows three simple steps, repeated automatically as new feedback arrives:

Read the responses

Topic detection looks at the open-text answers customers leave on your surveys.

Match against your themes

Each comment is compared with the themes in the model you've selected.

Tag automatically

Every comment is labelled with the themes it matches, so they can be counted, charted and filtered.

Two kinds of models

You choose which model decides what counts as a theme. There are two kinds, and you can use them together:

  • Pre-defined models — ready-made models that recognise common themes out of the box, with no setup required. They're the fastest way to get started.
  • Custom models — models tuned to your own language and the themes that matter to your business, for when the standard themes aren't specific enough. Custom models can run alongside the pre-defined ones.

Start ready-made, refine later

Most organizations begin with a pre-defined model to see results immediately, then add a custom model once they know which themes the standard set misses.

Turn it on

Enable topic detection

Switch on the master setting. When enabled, comments are analysed automatically using the selected model and topics.

Choose your model

Pick a pre-defined model, or choose Custom and name the custom model you want to use.

Set the depth and scope

Choose how deep the themes should go, the language of your comments, and any time window or limit you'd like to apply (see below).

Let it run

Save, and new comments are classified automatically from then on. Head to Text Analysis to see the themes emerge.

Settings

These options control exactly how comments are analysed:

  • Model — choose a pre-defined model, or select Custom to use one of your own.
  • Custom model — when the model is set to Custom, name which custom model should run.
  • Topic level — how deep the themes go. Basic keeps to top-level themes, while Advanced also surfaces sub-themes nested beneath them, for a more detailed breakdown.
  • Start date and End date — restrict the analysis to a specific time window, so you can focus on a campaign or season.
  • Language — the language of your comments, either Greek or English.
  • Classified comments limit — cap how many comments are analysed. Set it to 0 for no limit.
  • Custom topics — an optional list of your own themes. If you define them, the model's default topics are ignored and only your list is used.

Custom topic configurations Admin only

Beyond the main setup, you can run additional topic-detection models on your data, each producing its own dedicated reports — handy when different teams want themes analysed in different ways from the same feedback.

Administrators can:

  • Add a configuration — create a new model setup with the same fields as above.
  • Edit a configuration whenever your themes evolve.
  • Delete a configuration. This removes the configuration and its dedicated reports for good, so delete with care.

Each configuration also has a Template toggle. Templates are shared, reusable configurations available across your organization, so a setup that works well can be reused rather than rebuilt. Non-administrators can see existing configurations in read-only form, but only administrators can add, edit or delete them.

Deleting is permanent

Deleting a custom topic configuration also removes its dedicated reports for good. There's no undo, so make sure no one still relies on those reports first.

Where the results appear

Once topic detection is running, the themes it finds power the Text Analysis report in Insights. There you'll see the most-mentioned topics ranked, how they distribute across the theme hierarchy, and how each one trends over time. See Text Analysis for the full picture.

Pair it with sentiment

Topic detection tells you which themes appear; sentiment analysis tells you how customers feel about each one. Run both and you can see, for example, that delivery is praised while pricing is criticised.