PosNeg Tracking

PosNeg tracking refers to the recording of variably weighted user interactions with search results. As the name suggests, PosNeg tracking can be used to record both positive and negative interactions on query results. The resulting data is used for analytics and to power machine learning.

When initiating PosNeg tracking, both pos and neg tokens are generated for query-result pairs. Your app should store the tokens the user interacts with in ways you find relevant, e.g. a like, dislike, click, vote, product view, cart add or purchase. When these interaction events occur, your app should then send them back to our API along with the name of the event and optionally a weight to indicate strength when you have events of differing importance.

PosNeg tracking corrects for position bias by:

  1. penalizing negatively interacted results
  2. penalizing results occurring above those with positive interactions that were not interacted with
  3. boosting results with positive interactions

The above scoring system works slightly differently to Click based tracking. It is better for apps where:

  • clicks are not a good measure of success
  • multiple different events with different values that optimize differently (e.g. cart adds, purchases, wishlists)
  • implicit negative interactions occur (e.g. tinder-style thumbs up/down)

Otherwise it is more complex to implement and if your app does not fit into the above we recommend click tracking.

Documentation relating to PosNeg tracking for customers using our API can be found in relevant API clients