Use Case

You can use this plugin to calculate the relevance score of the two feature vector, such as:

• Personalized Search;
• Find similar product;
• Product recommendation;

Build Plugin from Code

1. git clone https://github.com/ginobefun/elasticsearch-feature-vector-scoring.git
2. mvn clean package -DskipTests
3. create vector scoring plugin dircectoy, such as ${ES_HOME}/plugins/vector-scoring
4. copy target/releases/elasticsearch-feature-vector-scoring-5.3.0.zip to your plugin directory and unzip it
5. restart elasticsearch

Script Parameters

• field: required, field in index to store the vector of document;
• inputFeatureVector: required, the condition vector, a string connected with comma;
• version: version of vector, if it isn't null, it should match the version of vector of document(if use version, the field value should start with ‘$VERSION|’, such as '20170331|0.1,2.3,-1.6,0.7,-1.3');
• baseConstant and factorConstant: used to calculate the final score, default value are 1. final_score = baseConstant + factorConstant * cos(X, Y)

Example

create a test index

PUT /test
{
  "mappings": {
    "product": {
      "properties": {
        "productName": {
          "type": "text",
          "analyzer": "standard"
        },
        "productFeatureVector": {
          "type": "keyword"
        }
      }
    }
  },
  "settings": {
    "index": {
      "number_of_shards": 1,
      "number_of_replicas": 0
    }
  }
}

index some documents

POST /test/product/1
{
  "productName": "My favorite brand of shoes",
  "productFeatureVector": "0.1,2.3,-1.6,0.7,-1.3"
}

POST /test/product/2
{
  "productName": "Normal brand of shoes",
  "productFeatureVector": "-0.5,1.6,1.1,0.9,0.7"
}

POST /test/product/3
{
  "productName": "The shoes I don't like",
  "productFeatureVector": "1.2,0.1,0.4,-0.2,0.3"
}

normal search

POST /test/_search
{
  "query": {
    "match": {
      "productName": "shoes"
    }
  }
}

the the result is:

{
  "took": 8,
  "timed_out": false,
  "_shards": {
    "total": 1,
    "successful": 1,
    "failed": 0
  },
  "hits": {
    "total": 3,
    "max_score": 0.14181954,
    "hits": [
      {
        "_index": "test",
        "_type": "product",
        "_id": "2",
        "_score": 0.14181954,
        "_source": {
          "productName": "Normal brand of shoes",
          "productFeatureVector": "-0.5,1.6,1.1,0.9,0.7"
        }
      },
      {
        "_index": "test",
        "_type": "product",
        "_id": "1",
        "_score": 0.1273061,
        "_source": {
          "productName": "My favorite brand of shoes",
          "productFeatureVector": "0.1,2.3,-1.6,0.7,-1.3"
        }
      },
      {
        "_index": "test",
        "_type": "product",
        "_id": "3",
        "_score": 0.1273061,
        "_source": {
          "productName": "The shoes I don't like",
          "productFeatureVector": "1.2,0.1,0.4,-0.2,0.3"
        }
      }
    ]
  }
}

search with feature vector score

POST /test/_search
{
  "query": {
    "function_score": {
      "query": {
        "match": {
          "productName": "shoes"
        }
      },
      "functions": [
        {
          "script_score": {
            "script": {
              "inline": "feature_vector_scoring_script",
              "lang": "native",
              "params": {
                "inputFeatureVector": "0.1,2.3,-1.6,0.7,-1.3",
                "field": "productFeatureVector"
              }
            }
          }
        }
      ],
      "boost_mode": "multiply"
    }
  }
}

and the result is:

{
  "took": 35,
  "timed_out": false,
  "_shards": {
    "total": 1,
    "successful": 1,
    "failed": 0
  },
  "hits": {
    "total": 3,
    "max_score": 0.2546122,
    "hits": [
      {
        "_index": "test",
        "_type": "product",
        "_id": "1",
        "_score": 0.2546122,
        "_source": {
          "productName": "My favorite brand of shoes",
          "productFeatureVector": "0.1,2.3,-1.6,0.7,-1.3"
        }
      },
      {
        "_index": "test",
        "_type": "product",
        "_id": "2",
        "_score": 0.17267346,
        "_source": {
          "productName": "Normal brand of shoes",
          "productFeatureVector": "-0.5,1.6,1.1,0.9,0.7"
        }
      },
      {
        "_index": "test",
        "_type": "product",
        "_id": "3",
        "_score": 0.10233018,
        "_source": {
          "productName": "The shoes I don't like",
          "productFeatureVector": "1.2,0.1,0.4,-0.2,0.3"
        }
      }
    ]
  }
}

A personalized search case details

see http://ginobefunny.com/post/personalized_search_implemention_based_word2vec_and_elasticsearch/