Topic Modeler

Explore and edit the fingerprint of an expression

Clicking on the terms of the expression will
highlight the corresponding dots in the fingerprint.
You can enter an expression in any of the supported languages.


With our Topic Modeler, you can intuitively synthesize meaning by creating a highly customized semantic fingerprint. You just need to point and click the single dots corresponding to the topics you want to include or exclude.

With this tool, you can create a reference fingerprint that represents what you are interested in, what you are looking for: this is a great basis for a filter. The high throughput and fast responses delivered by the Retina API enable the application of this filter to any stream of text: mail messages, tweets, timeline entries, blog posts etc.

You can also use your reference fingerprint to create a filter to classify an existing set of documents. E.g. select all experts relevant to your industry area within LinkedIn profiles, or personalize the ranking of your result sets in a search engine. The filter can, of course, be saved to a file (in JSON format), and, if necessary, re-uploaded for further editing.

Use Cases

  • Streaming Text Filtering
  • Classification

How It Works

The Topic Modeler uses the underlying Retina API to obtain a semantic fingerprint of the input expression (representing 16,000 semantic features), which is rendered as an image.

Hovering the mouse over the image will cause a text box to appear displaying the terms associated with each particular pixel and help you get a feel for which meaning each of the pixels represents.

This semantic fingerprint is used to retrieve a list of similar terms related to the input expression. Clicking on any of the terms in the Input or Output section will highlight the features associated with the term in the fingerprint image. This will give you a feel for how individual words are represented by multiple features.

How To Use it

Begin by setting your input language from the list of available languages in the drop-down menu. Next replace the example expression in the Input section by entering a term or expression in order to create a semantic fingerprint for your topic of interest.

As you begin typing in the empty Input area, you will notice the display changes to show a list of commands and supported operators to build an expression:

  • Pressing the tab or enter key after a term or operator completes the term or operator (you will need to press the tab or enter key after each input term or operator). The semantic fingerprint image and the output terms are updated when you do this.
  • You can click on a term or operator to select it and drag it to a different location in your expression.
  • Shift and click or control and click can be used to select multiple items.
  • The operators allow you to perform Boolean (AND, OR, XOR) or arithmetical (SUB) operations on the semantic fingerprint. For instance, you may want to create a fingerprint about Java (the island) so an expression of the form java SUB computer could be used to filter out computer related contexts.
  • If your input expression contains a term that is a reserved word (one of: AND, OR, XOR, SUB) then you will need to escape that term with single or double quotes. E.g. ship OR ‘sub’ where the input term sub is a reserved word (otherwise used as operator).

As a second step you can begin editing your fingerprint. The Topic Modeler allows you to edit the 16,000 semantic features directly by modifying the image of the fingerprint using the following buttons:

  • The Point button: clicking on this button and then clicking on the image will add the feature represented by that point. (In the image, a point is represented as one pixel).
  • The Brush button: click on this button, select the width of your brush strokes, and then paint over the image holding down the left mouse button. This action will add multiple features to the fingerprint. Each of the painted pixels adds one semantic feature.
  • The Eraser button: this button can be used to remove features from a fingerprint. You can select fine point erasure (1px) or a broader scope (3px, 5px).
  • The Save button: when you have completed your fingerprint (or if you want to save your work for another day) you can use the save button to download your fingerprint in JSON format.
  • The Upload button: use this button if you want to upload and work on a previously created fingerprint.
  • The Grid button: this button can be used to overlay a grid on the image. This can be helpful in locating individual pixels.

Note: It is much easier to erase unwanted features than to add needed features. As a general rule, it is better to add needed features via the expression and then finely tune the fingerprint by deleting unwanted features using the Erase button on the fingerprint image.