elasticsearch-flavor

Apache Mahout base Recommendation engine. Use elasticsearch to storage and REST interface.

Version

Flavor	Elasticsearch
0.0.3	1.5.0

Installation

$ plugin --url 'https://github.com/f-kubotar/elasticsearch-flavor/releases/download/v0.0.3/elasticsearch-flavor-0.0.3.zip' --install flavor

Getting Start

Setup Preference index

curl -XPOST localhost:9200/my_index -d '{
    "mappings" : {
        "preference" : {
            "preference": {
                "properties": {
                    "user_id": {
                        "type": "long"
                    },
                    "item_id": {
                        "type": "long"
                    },
                    "value": {
                        "type": "float"
                    }
                }
            }
        }
    }
}'

Name	Type	Description
user_id	long	Unique User ID for this plugin.
item_id	long	Unique Item ID for this plugin.
value	float	Value rated by user_id for item_id.

Recommendation

Similar Items

GET /{index}/{type}/_flavor/{index}/{type}/similar_items/{item_id}

Query Parameters

Name	Type	Description
index	String	Index name that there is preference.
type	String	Document Type name of the preference.
size	int	Number of recommend items
similarity	String	ItemSimilarity algorithm name. Default value is PearsonCorrelationSimilarity. Other values: EuclideanDistanceSimilarity LogLikelihoodSimilarity TanimotoCoefficientSimilarity

Curl Example

$ curl 'localhost:9200/my_index/preference/_flavor/similar_items/5803?size=3'
HTTP/1.1 200 OK
Content-Length: 126
Content-Type: application/json; charset=UTF-8

{
    "hits": {
        "hits": [
            {
                "item_id": 40891,
                "value": 1.0
            },
            {
                "item_id": 48541,
                "value": 1.0
            },
            {
                "item_id": 151,
                "value": 1.0
            }
        ],
        "total": 3
    },
    "took": 4
}

Similar Users

GET /{index}/{type}/_flavor/similar_users/{user_id}

Query Parameters

Name	Type	Description
index	String	Index name that there is preference.
type	String	Document Type name of the preference.
size	int	Number of recommend items
similarity	String	UserSimilarity algorithm name. Defaualt value is PearsonCorrelationSimilarity. Other values: EuclideanDistanceSimilarity
neighborhood	String	UserNeighborhood algorithm name. Defaualt value is NearestNUserNeighborhood. Other values: ThresholdUserNeighborhood
neighborhoodN	int	neighborhood size. capped at the number of users in the data. Allow if using NearestNUserNeighborhood.
neighborhoodThreshold	float	User similarity threshold. Allow if using ThresholdUderNeighborhood.

Curl Example

$ curl 'localhost:9200/my_index/preference/_flavor/similar_users/9?size=3'
HTTP/1.1 200 OK
Content-Length: 86
Content-Type: application/json; charset=UTF-8

{
    "hits": {
        "hits": [
            {
                "user_id": 15719
            },
            {
                "user_id": 2
            },
            {
                "user_id": 78
            }
        ],
        "total": 3
    },
    "took": 55
}

Recommend

GET /{index}/{type}/_flavor/user_based_recommend/{user_id}

Or

GET /{index}/{type}/_flavor/item_based_recommend/{user_id}

Query Parameters

Name	Type	Description
index	String	Index name that there is preference.
type	String	Document Type name of the preference.
size	int	Number of recommend items
similarity	String	UserSimilarity algorithm name. Defaualt value is PearsonCorrelationSimilarity. Other values: EuclideanDistanceSimilarity
neighborhood	String	UserNeighborhood algorithm name. Defaualt value is NearestNUserNeighborhood. Other values: ThresholdUserNeighborhood
neighborhoodN	int	neighborhood size. capped at the number of users in the data. Allow if using NearestNUserNeighborhood.
neighborhoodThreshold	float	User similarity threshold. Allow if using ThresholdUderNeighborhood.

Curl Example

$ curl 'localhost:9200/my_index/preference/_flavor/user_based_recommend/9?size=3'
HTTP/1.1 200 OK
Content-Length: 86
Content-Type: application/json; charset=UTF-8
{
    "hits": {
        "hits": [
            {
                "item_id": 58506,
                "value": 2.5583827
            },
            {
                "item_id": 222964,
                "value": 2.460672
            },
            {
                "item_id": 7015,
                "value": 2.4304452
            }
        ],
        "total": 3
    },
    "took": 947
}

Preload

Read all preference data into memory using the FastIDSet of Mahout.

curl -XPOST localhost:9200/_flavor/reload -d '{
    "preference" : {
        "index" : "my_index",
        "type" : "preference"
    }
}'

Response

{"dataModel":"ElasticsearchPreloadDataModel[index:suzuri_preference_development type:preference]","total_users":6907,"total_items":40695}

Preload data Usage

If you omit the {index}/{type}, to become using preload data.

$ curl localhost:9200/_flavor/similar_items/101