It was already mentioned that the recommendation engine collects events and computes recommendations based on the statistic. The most important events collected by the engine are click and buy-events. It is sufficient for providing basic recommendations. There are some additional events for creating more complex scenarios and providing statistic about acceptance of recommendations like conversion rate or revenue.
All possible events used in the system are described in the following table.
|CLICK||It should be sent to the recommendation engine, if customer opens any of the product or article detail page on the shop or website.|
|BUY||It should be sent, if something was bought.|
|CONSUME||Similar to the BUY event but without a payment. It is designed for publisher websites. This event is, if a article or a web page is consumed (usually "read" or "pre-listened") without paying money.|
|RENDER||It should be sent, if a recommendation (which was fetched from the recommendation engine) is shown on the web page. This information is used by filters to suppress repeated recommendations of the same item.|
|FOLLOW||If a user clicked on the recommended product the FOLLOW (a.k.a. "clickrecommended") event must be sent. It allows to build acceptance statistics and enables A/B testing.|
|TRANSFER||A special type of event to deal with a user login after the user allready surfed on the web page anonymously. It should always be sent if the identifier of the user changed. As a result the anonymous history of the user will be transfered from the old identifier to the new one. This workflow is automatically done in the recommender engine.|
Additional events for advanced edition
|BASKET||It should be sent when the user added the specified product to the shopping cart. It allows to create recommendation for products the customer is interested in but which were finally not bought by the user for whatever reasons.|
|Allows the user to suppress this recommendation. If the recommendation engine gets this event, the specified product or article won't be recommended anymore for the specified user. Default suppressing interval is one year.|
|OWNS||Same as BUY, but without influence on the statistic. It can be used, if a user already owns the product, but bought it somewhere else and to avoid to recommend this product again.|
|RATE||Additional models can be created for the advanced edition, using this type of events. It allows to build recommendations not only for implicit tracking events like "clicked" or "buy" but also for explicit events "rated" or "liked". This events need additional integration into the web page to allow the user to give an appropriate feedback. The event should be triggered as an result of this user feedback.|
All the events requires the current user ID and the ID of one ore more context items. Some events need additional information. For more sophisticated algorithms and result filtering event types with additional parameters are needed. In the table below is a brief overview to additional parameter information.
|CLICK||Category path of the item the customer clicked on can be attached to the event. It is the alternative way to provide information for the item store. This way is available for both basic and advanced editions. If you implemented the full item import, you should not provide this information over the click event.|
|BUY||The price a user paid for the product. This is an important parameter for the statistic and especially for A/B tests. Quantity of the products bought must be sent as well. This parameter does not update the item store. It is used only for revenue statistic, not for item filtering.|
|FOLLOW||The scenario which provided the recommendations must be sent in this event.|
|RATE||The rating (for example 1 to 5 stars) can be sent as an additional parameter. See the 1. Tracking Events for information about rating scales.|
The event tracker specification with all events and paramaters is available in the developer space. See the following document for more information: