This is the Recommendtion Service for Model AirBnb. It takes into account information from User Behavior, Inventory and Search and provides recommendations on how to price future properties. These recommendation are used by Search to sort Search Results.
The recommendations service consumes messages from search query (Search query lets analytics know which searches were actually performed to get the search results), search results (Search results are consumed to see what results are actually being published after the weights, and whether those do lead to purchases) and booking details (booking details are provided when a booking is performed. The booking is consumed by Inventory, and Inventory adds on additional values related to the listing).
The recommendations servive publishes messages that has recommendation weights for the search service. The sort order scores are used by the search service to sort search results.
See CONTRIBUTING.md for contribution guidelines.
- Usage
- Requirements
- Development
- System Overview
- System Service Architecture
- Messages Consumed
- Messages Published
- Schema Design
To install, run the following commands. Make sure to update the proper locations for Elastic Search, Postgres, Redis, and SQS Services.
npm install
- aws-sdk 2.141.0
- body-parser 1.17.0
- elasticsearch 13.3.1
- express 4.15.0
- pg 7.3.0
- redis 2.8.0
- winston 3.0.0-rc1
- winston-elasticsearch 0.5.3
Search query lets analytics know which searches were actually performed to get the search results. The weights used will be seen here.
| Property | Description | Required |
|---|---|---|
| Search Query ID | TBD range | yes |
| User ID | (coordinate range with Behaviour) | yes |
| Market ID | Identifier for destination being searched | yes |
| Check-in | date | yes |
| Check-out | date | yes |
| Room Type | string | no |
| Amenities | List of amenity IDs | no |
| Neighbourhood ID | TBD range | no |
Search results are consumed to see what results are actually being published after the weights, and whether those do lead to purchases.
| Property | Possible Values | Required |
|---|---|---|
| Search Query ID | TBD range | yes |
| Available Listings | Sorted list of listings provided to the client in response to a search query | yes |
| Listing ID | Identifier for the listing | yes |
| Neighbourhood ID | Identifier for the neighbourhood | yes |
| Room Type | string | yes |
| Amenities | List of amenity IDs | yes |
| Nightly Prices | list | yes |
| Date | date | yes |
| Price | float | yes |
| Average Rating | Float (NULL of number of reviews = 0) | yes |
| Recommendation Weight | Object | yes |
Booking details are provided when a booking is performed. The booking is consumed by Inventory, and Inventory adds on additional values related to the listing.
| Property | Possible Values | Required |
|---|---|---|
| Listing ID | Identifier for the listing | yes |
| User ID | (coordinate range with Behaviour) | yes |
| Search ID | Identifier for search | yes |
| Neighbourhood | Identifier for the neighbourhood | yes |
| Room Type | string | yes |
| Amenities | List of amenity IDs | yes |
| Nightly Prices | list | yes |
| Date | date | yes |
| Price | float | yes |
| Average Rating | Float (NULL of number of reviews = 0) | yes |
If there are no values provided for coefficients, assume weight is 1.0. The sort order scores are used by the search service to sort search results.
Subscribers: Search Service
| Property | Description | Required |
|---|---|---|
| Recommendations ID | integer | yes |
| Date | date | yes |
| Rules | Object (with below keys/values) | yes |
| room type | string | no |
| market id | integer | no |
| neighborhood | string | no |
| amenities | array of integers | no |
| user id | integer | yes |
| price range | array [min price, max price] | no |
| Coefficients | Object (with below keys/values) | yes |
| price | float (-2.00 - 2.00) | no |
| amenities | Set of coefficient to apply for available amenities | no |
| id | integer | no |
| coefficient | float (-2.00 - 2.00) | no |
| room type | object (key as room type_id (string), value as float) | no |
| type | string | no |
| coefficient | float (-2.00 - 2.00) | no |
| rating | float (-2.00 - 2.00) | no |
For example, a scoring update message could look like:
[
{
recommendationId: 12345,
date: '2017-10-20T21:39:37Z',
rules: {
userId: 54321,
marketId: 10,
neighborhood: 'Castro',
amenities: [1, 3, 4, 5],
roomType: 'entire home'
}
coefficients:
priceCoefficient: -1.2
}
}
]
