In the advanced edition of the recommendation engine it is possible to build different recommendation domains. It can be achieved over splitting all the products into different item types. There are several possible use-cases for item types.
- A publisher tracks articles, pictures and videos as three different types.
- A supermarket splits all the products in food and non-food product groups.
- An owner of several web-shops uses a single account for all of them
Every products must belong to one and only one item type. Using different item types is similar to having separated accounts in the recommendation system. Whereas the item types concept provides the possibility to make cross type recommendations (like "users who watched this film also read this book").
Additionally to the logical splitting of the product domains the item types provide another important advantage. It plays a role if different types of product are not equal popular but must be equal often recommended. For example on a publisher page users watch video less often than reading articles. If the most popular products are requested without item type splitting, there will be most likely no videos in the result. If articles and videos are split into different item types, it is possible to request explicitly popular videos. This characteristic is very similar to the one sub-models provide (see 10. Sub-Models).
Following is a comparison for different ways of splitting products in the recommendation engine.
|Use case||Item types solution |
|Attribute based sub-models |
|Different recommendation engine accounts|
|Cross group recommendation requests are possible.||yes||yes||no|
|Product must belong to one group.||mandatory||optional||mandatory|
|Product can belong to multiple groups||no||yes||no|
|Single recommendation request can contain products from different groups||no (this feature in under construction)||yes||no|
|Product can change the group it belongs to||no||yes||no|
|Different item types share the scenario and model configuration||yes||yes||no|
|Required Recommender Edition||single advanced||multiple basic|
If multiple item types are enabled one must define supported item types for every scenario.
Every scenario supports a single input type (see 4. Scenario Context) and multiple output types. Every recommendation request delivers only one output item type (even if multiple selected in the interface above). The input and output types are set during the recommendation request and they must be covered by the list of the supported item types.