Recommendations

Recommend Items to User

Based on the user's past interactions (purchases, ratings, etc.) with the items, recommends top-N items that are most likely to be of high value for the given user.

The most typical use cases are recommendations on the homepage, in some "Picked just for you" section, or in email.

The returned items are sorted by relevance (the first item being the most relevant).

Besides the recommended items, also a unique recommId is returned in the response. It can be used to:

Let Recombee know that this recommendation was successful (e.g., user clicked one of the recommended items). See Reported metrics.
Get subsequent recommended items when the user scrolls down (infinite scroll) or goes to the next page. See Recommend Next Items.

It is also possible to use POST HTTP method (for example in the case of a very long ReQL filter) - query parameters then become body parameters.

post/{databaseId}/recomms/users/{userId}/items/

Path parameters

databaseIdstring required

ID of your database.

userIdstring required

ID of the user for whom personalized recommendations are to be generated.

Request body

countinteger required

Number of items to be recommended (N for the top-N recommendation).

scenariostring

Scenario defines a particular application of recommendations. It can be, for example, "homepage", "cart", or "emailing".

You can set various settings to the scenario in the Admin UI. You can also see the performance of each scenario in the Admin UI separately, so you can check how well each application performs.

The AI that optimizes models to get the best results may optimize different scenarios separately or even use different models in each of the scenarios.

cascadeCreateboolean

If the user does not exist in the database, returns a list of non-personalized recommendations and creates the user in the database. This allows, for example, rotations in the following recommendations for that user, as the user will be already known to the system.

returnPropertiesboolean

With returnProperties=true, property values of the recommended items are returned along with their IDs in a JSON dictionary. The acquired property values can be used to easily display the recommended items to the user.

Example response:

  {
    "recommId": "ce52ada4-e4d9-4885-943c-407db2dee837",
    "recomms": 
      [
        {
          "id": "tv-178",
          "values": {
            "description": "4K TV with 3D feature",
            "categories":   ["Electronics", "Televisions"],
            "price": 342,
            "url": "myshop.com/tv-178"
          }
        },
        {
          "id": "mixer-42",
          "values": {
            "description": "Stainless Steel Mixer",
            "categories":   ["Home & Kitchen"],
            "price": 39,
            "url": "myshop.com/mixer-42"
          }
        }
      ],
     "numberNextRecommsCalls": 0
  }

includedPropertiesstring[]

Allows specifying which properties should be returned when returnProperties=true is set. The properties are given as a comma-separated list.

Example response for includedProperties=description,price:

  {
    "recommId": "a86ee8d5-cd8e-46d1-886c-8b3771d0520b",
    "recomms":
      [
        {
          "id": "tv-178",
          "values": {
            "description": "4K TV with 3D feature",
            "price": 342
          }
        },
        {
          "id": "mixer-42",
          "values": {
            "description": "Stainless Steel Mixer",
            "price": 39
          }
        }
      ],
    "numberNextRecommsCalls": 0
  }

filterstring

Boolean-returning ReQL expression, which allows you to filter recommended items based on the values of their attributes.

Filters can also be assigned to a scenario in the Admin UI.

boosterstring

Number-returning ReQL expression, which allows you to boost the recommendation rate of some items based on the values of their attributes.

Boosters can also be assigned to a scenario in the Admin UI.

reqlExpressionsobject

A dictionary of ReQL expressions that will be executed for each recommended item. This can be used to compute additional properties of the recommended items that are not stored in the database.

The keys are the names of the expressions, and the values are the actual ReQL expressions.

Example request:

{
  "reqlExpressions": {
    "isInUsersCity": "context_user[\"city\"] in 'cities'",
    "distanceToUser": "earth_distance('location', context_user[\"location\"])"
  }
}

Example response:

{
  "recommId": "ce52ada4-e4d9-4885-943c-407db2dee837",
  "recomms": 
    [
      {
        "id": "restaurant-178",
        "reqlEvaluations": {
          "isInUsersCity": true,
          "distanceToUser": 5200.2
        }
      },
      {
        "id": "bar-42",
        "reqlEvaluations": {
          "isInUsersCity": false,
          "distanceToUser": 2516.0
        }
      }
    ],
   "numberNextRecommsCalls": 0
}

diversitynumber double

Expert option: Real number from [0.0, 1.0], which determines how mutually dissimilar the recommended items should be. The default value is 0.0, i.e., no diversification. Value 1.0 means maximal diversification.

minRelevance'low' | 'medium' | 'high'

Expert option: Specifies the threshold of how relevant must the recommended items be to the user. Possible values one of: "low", "medium", "high". The default value is "low", meaning that the system attempts to recommend a number of items equal to count at any cost. If there is not enough data (such as interactions or item properties), this may even lead to bestseller-based recommendations to be appended to reach the full count. This behavior may be suppressed by using "medium" or "high" values. In such a case, the system only recommends items of at least the requested relevance and may return less than count items when there is not enough data to fulfill it.

rotationRatenumber double

Expert option: If your users browse the system in real-time, it may easily happen that you wish to offer them recommendations multiple times. Here comes the question: how much should the recommendations change? Should they remain the same, or should they rotate? Recombee API allows you to control this per request in a backward fashion. You may penalize an item for being recommended in the near past. For the specific user, rotationRate=1 means maximal rotation, rotationRate=0 means absolutely no rotation. You may also use, for example, rotationRate=0.2 for only slight rotation of recommended items. Default: 0.

rotationTimenumber double

Expert option: Taking rotationRate into account, specifies how long it takes for an item to recover from the penalization. For example, rotationTime=7200.0 means that items recommended less than 2 hours ago are penalized. Default: 7200.0.

expertSettingsobject

Dictionary of custom options.

returnAbGroupboolean

If there is a custom AB-testing running, return the name of the group to which the request belongs.

Response

Successful operation.

recommIdstring required

Id of the recommendation request

numberNextRecommsCallsinteger

How many times Recommend Next Items have been called for this recommId

abGroupstring

Name of AB-testing group to which the request belongs if there is a custom AB-testing running.

Changes

No recorded changes to this endpoint across all 1 revision of this API.