Segment creation via Recommendation Engine

Here, the user can enter in the name of the new segment.

The language and target market of the segment are predetermined by the settings of the taxonomy it’s being created for. This is not modifiable at this stage.

The user can start adding seedwords by typing their words into the “Seedwords” field. Autocompletion will allow the display of matching available words as the user types. Once a word is selected, it will be added under the “Words in Segment” column, which contains the content of the segment (refer to Words in Segment). The word will be color-coded in blue, and an “Audience Estimate” will be recomputed.

Recommendations are words that the user can add to enrich his or her segment and are proposed based on their similarity in relation to the seedword. The user can see recommendations for a particular seedword by clicking on it. Once the recommendations have loaded, the user can then click on the words they would like to add to the segment. To see more recommendations, the user can click on “Next” in the bottom right corner of the section or “Prev” to see the previous set of words. A whole page of recommendations can also be added to the “Words” column by clicking on the “Add this page” button in the bottom left corner.

Specificity

The words that appear in the “Recommendations” section are color-coded according to how specific they are. Specificity is an absolute value attached to a given word; it reflects how rare a word is in the language. The value is computed on a wide reference corpus.

Words with a red circle next to them have a low specificity (ex. lip) and are more common than those with a high specificity. While you may have a bigger audience with these words, they can be “dangerous” as they can generate an audience of users that do not show a specific interest. (For example, the word “lip” will refer to anyone who has been exposed to it at least once while browsing, however this doesn’t necessarily mean that he/she would be interested in “lipstick” specifically and therefore is not guaranteed to be our target.)

Words with an orange circle are neither too common nor too rare (ex. Besame). These words are more specific and guarantee a better targeting (ex. “Besame” rather than the word “lip”, which is too generic).

Words with a green circle have the highest specificity (ex. Le Disko). Although you may have a smaller audience with these words, they provide a guarantee of reaching a very specific target. In this case, it is a matter of quality over quantity.

On the right in the “Words in Segment” section, the user will be able to see all of the words being added to the new segment, which includes seedwords and any recommendations added. A maximum of 100 words can be added. Seedwords are color-coded in blue. Suggestions are color-coded according to their specificity (red, orange, green) and will appear in descending similarity to the seedword. The Audience Estimate of each word in the segment will be visible to the right of the word. The words in the segment can also be ordered by either Weight or Audience Estimate.

The Audience Estimate for the segment will be visible in the upper right corner of the segment creation screen. Audience Estimate is computed through the Semantic Matching technique. The matching audience number will rely on specificity and the number of added words. Words with a low specificity will generate a larger audience as the words are common, whereas words with a higher specificity will have a smaller audience. Audience estimate takes account of all words in the segment which is being created. It is therefore a lower number than the sum of the audience estimates attached to all words, one by one.

Once satisfied with the segment, the user can click on “Create” in the upper right corner to add the new segment to his or her taxonomy. If the user wishes to edit the segment, it can be accessed by selecting the relevant taxonomy from the Folder View.