Holst

Holst is a (minimalist) Recommentations Engine as a Service, which uses collaborative filtering to compute the recommendations. Using a RESTful API, it allows users to create and manage their own sets of data (comprised of Users, Items, and Links), and to retrieve customized recommendations, based on that information.

In a nutshell, a user can have one or multiple links with an item, which are subsequently used to generate item recommendations for a given user.

Disclaimer: This tool is not ready for production use yet. I started this a side project that allowed me to play around with Neo4j and graph database concepts.

Getting started

• Install and start Neo4j.
• Install dependencies, and build the app: go get & go build.
• Copy the sample config file config.json.sample into a new file config.json, under the same directory, and edit this file according to your system configuration
• Run the app: ./holst

Authorization & Authentication

The current implementation includes a temporary (and rather naive) authentication mechanism. All requests must include an Authorization header, using Basic access authentication. The username and password are directly set up in the config.json file.

Hint: If you just want to test the API, you can simply use the default credentials in the sample config file, with the following authorization header:

Authorization: Basic aHR0cGNsaWVudDpzZWNyZXRzdHVmZg==

Manage data (CRUD operations)

The data model consists of users, items, and links. Every model contains a unique identifier, which can be explicitly provided, or automatically generated when issuing a POST request.

Users

• id [required]
• name [optional]

Create/Update users

• POST /user

• Payload: JSON-encoded user. Explicitly providing an ID will update the existing entity, if found. Otherwise a new user will be created. The lack of ID will always result in a new entity creation.

• Example Request:

   POST /user
   { 
     "name" : "Roger"
   }	

• Example Response

   201 Created
   {
     "id": "8a7ddc079f118bb1",
     "name": "Roger"
   }

Read users

• GET /user/{id}

• Example Request:

   GET /user/8a7ddc079f118bb1	
  

• Example Response

   200 OK
   {
     "id": "8a7ddc079f118bb1",
     "name": "Roger"
   }

Delete users

• DELETE /user/{id}

• Example Request:

   DELETE /user/8a7ddc079f118bb1	
  

• Example Response

   200 OK
  

Items

• id [required]
• name [optional]
• categories [optional]

Create/Update items

• POST /item

• Payload: JSON-encoded item. Explicitly providing an ID will update the existing entity, if found. Otherwise a new item will be created. The lack of ID will always result in a new entity creation.

• Example Request:

   POST /item
   { 
     "name" : "Pulp Fiction",
     "categories": ["movie", "1994"]
   }	

• Example Response

   201 Created
   {
     "id": "8a7ddc079f118bb2",
     "name": "Pulp Fiction",
     "categories": ["movie", "1994"]
   }

Read items

• GET /item/{id}

• Example Request:

   GET /item/8a7ddc079f118bb2
  

• Example Response

   200 OK
   {
     "id": "8a7ddc079f118bb2",
     "name": "Pulp Fiction",
     "categories": ["movie", "1994"]
   }

Delete items

• DELETE /item/{id}

• Example Request:

   DELETE /item/8a7ddc079f118bb2
  

• Example Response

   200 OK
  

Links

• id [required]
• userId [required]
• itemId [required]
• type [optional]
• score [optional]

Create/Update links

Although a link's ID can also be automatically generated, both the userId and itemId need to be specified upfront. Additionally, you can pass the link type and score. These can be used for filtering results. A type can be any keyword relevant to your application (for example: buy, like, etc)

• Example Request:

   POST /link
   {
     "userId" : "8a7ddc079f118bb1",
     "itemId" : "8a7ddc079f118bb2",
     "type" : "like"
   }	

• Example Response

   201 Created
   {
     "id": "8a7ddc079f118bb3",
     "userId" : "8a7ddc079f118bb1",
     "itemId" : "8a7ddc079f118bb2",
     "type" : "like"
   }

Read links

• GET /link/{id}

• Example Request:

   GET /link/8a7ddc079f118bb3
  

• Example Response

   200 OK
   {
     "id": "8a7ddc079f118bb3",
     "userId" : "8a7ddc079f118bb1",
     "itemId" : "8a7ddc079f118bb2",
     "type" : "like"
   }

Delete links

• DELETE /link/{id}

• Example Request:

   DELETE /item/8a7ddc079f118bb3
  

• Example Response

   200 OK
  

Recommendations

Recommendations are performed using collaborative filtering. Providing a user id, and optionally, item categories and link type, the system will return a list of recommended items.

Getting recommendations

• GET /recommend/{userId}?[category=cat1&category=cat2]&[type=t]

Where category can hold multiple values (just concatenate the parameters in the query), and type is the link type. For example, querying only for type=like will only return results based on that relationship type. Each recommendation has a strength attribute, which represents the number of occurrences of such item. Results are ordered by this attribute.

• Example Request:

   GET /recommend/8a7ddc079f118bb1?category=movie&type=buy
  

• Example Response

   200 OK
   [
      {
        "item": {
            "id": "aaaa",
            "name": "Casablanca"
        },
        "strength": 7
      },
      {
        "item": {
            "id": "bbbb",
            "name": "Star Wars"
        },
        "strength": 4
      },
      {
        "item": {
            "id": "cccc",
            "name": "The Lord of the Rings"
        },
        "strength": 3
      }
   ]

License

This software is distributed under the BSD-style license found in the LICENSE file.