Cannot set up script_score for dense_vector using Java REST API

Elastic Stack Elasticsearch
1

I want to search similar sentences using dense vectors calculated by BERT.

I am following this great paper in french or english

I indexed sentences with each having a dense_vector calculated by BERT.

I ran this query with success in KIbana giving wonderful results (I shortened the dense_vector values for this post) :

    GET webmonitor_phrases/_search
    {
      "size" : 5,
      "query": {
        "script_score": {
          "query" : {
            "match_all" : {}
          },
          "script": {
            "source": "cosineSimilarity(params.queryVector, 'Phrase_vector') + 1.0",
            "params": {
              "queryVector": [-0.01927643,-0.018217519,-0.021058332,0.027151054,0.05784438,-0.008460273,-0.027456086,0.042760707,-0.08016594,-0.017899536,-0.041639842,0.02960743,-0.01679815,0.1015308,0.05703567,0.0069785477,0.009718613,0.008082357,-0.012589352,0.12446663,0.018861143,0.053296693,-0.03215604,0.010972261,-0.10968127,0.012995018,0.07786082,-0.017690469,0.03248199,-0.04271781,-0.06806753,-0.07297233,0.0035993687,0.011882267,0.016708337,-0.019617425,-0.047242224,0.014557032,0.04139613,-0.008320617,-0.04017011,-0.03703198,0.022198584,-0.021771815,0.02410845,0.016920311,-0.02317638,0.004767835,0.023395542,0.007503315,0.025918411,-0.016145818,-0.06478562,-0.08040128,0.002344442,-0.008921131,0.047266897,0.047718693,-0.0011429694,0.05584679,-0.024764488,0.0192903,0.024397915,-0.045773678,-0.024413446,0.021329394,-0.0469761,0.063250914,-0.046951707,0.0404466,-0.065411255,-0.038457282,0.029206796,-0.01588464,0.04883571,0.026628332,0.05048967,-0.012432704,-0.024493853,0.003305153,0.0039466517,-0.04878099,-0.033499695,0.040233597,0.01040456,0.0076247444,-0.052615903,0.036758807,-0.0438332,-0.015946992,-0.08943201,0.014444613,-0.0019423354,-0.020825846,-0.022945268,0.0867511,-0.09420672,0.011214681,-0.06201148,0.10487903,-0.015105568,-0.00086394855,0.04614532,0.05401257]
            }
          }
        }
      }
    }

I am facing an issue when trying to do exactly the same thing using Java REST client.

Following this, I coded the following snippet :

                // construction de la request
                request = new SearchTemplateRequest();
                request.setRequest(searchRequest);

                request.setScriptType(ScriptType.INLINE);

                //                            "            \"source\": \"cosineSimilarity([{{value_vector}}], doc['{{interro_vector}}']) + 1.0\"  " +
                //                            "            \"source\": \"cosineSimilarity([" + dense_vector + "], doc['" +  enrg.getInterrogation() + "']) + 1.0\"  " +


                script =
                    "{" +
                        "    \"script_score\": {       " +
                        "        \"query\": { " +
                        "           \"query_string\" : { " +
                        "               \"query\" : \"" + query + "\", " +
                        "               \"default_field\" : \"tout\" " +
                        "            }  " +
                        "        },   " +
                        "        \"script\": { " +
                        "        \"source\": \"cosineSimilarity([" + dense_vector + "], '" +  enrg.getInterrogation() + "' ) + 1.0\"  " +
                        "        } " +
                        "    }  " +
                        "}";
                request.setScript(script);

                Map<String, Object> scriptParams = new HashMap<>();
                //scriptParams.put("interro_vector", enrg.getInterrogation());
                //scriptParams.put("value_vector",  dense_vector );
                request.setScriptParams(scriptParams);

I get this in the logs (I shortened the dense_vector values for this post):

    |2020/06/26 13:17:37 075-qeswebmonitor-demoqwam-TRACE-QESHost_elastic62-buildRequestForDenseVector()|: Fin:
{    
"script_score": {               
  "query": {            
    "query_string" : {                
      "query" : "*:*",                
      "default_field" : "tout"             
    }          
  },           
  "script": {         
    "source": 
"cosineSimilarity([-0.01927643,-0.018217519,-0.021058332,0.027151054,0.05784438,-0.008460273,-0.027456086,0.042760707,-0.08016594,-0.017899536,-0.041639842,0.02960743,-0.01679815,0.1015308,0.05703567,0.0069785477,0.009718613,0.008082357,-0.012589352,0.12446663,0.018861143,0.053296693,-0.03215604,0.010972261,-0.10968127,0.012995018,0.07786082,-0.017690469,0.03248199,-0.04271781,-0.06806753,-0.07297233,0.0035993687,0.011882267,0.016708337,-0.019617425,-0.047242224,0.014557032,0.04139613,-0.008320617,-0.04017011,-0.03703198,0.022198584,-0.021771815,0.02410845,0.016920311,-0.02317638,0.004767835,0.023395542,0.007503315,0.025918411,-0.016145818,-0.06478562,-0.08040128,0.002344442,-0.008921131,0.047266897,0.047718693,-0.0011429694,0.05584679,-0.024764488,0.0192903,0.024397915,-0.045773678,-0.024413446,0.021329394,-0.0469761,0.063250914,-0.046951707,0.0404466,-0.065411255,-0.038457282,0.029206796,-0.01588464,0.04883571,0.026628332,0.05048967,-0.012432704,-0.024493853,0.003305153,0.0039466517,-0.04878099,-0.033499695,0.040233597,0.01040456,0.0076247444,-0.052615903,0.036758807,-0.0438332,-0.015946992,-0.08943201,0.014444613,-0.0019423354,-0.020825846,-0.022945268,0.0867511,-0.09420672,0.011214681,-0.06201148,0.10487903,-0.015105568,-0.00086394855,0.04614532,0.05401257], 'Phrase_vector' ) + 1.0"          
    }     
  }  
}|
    |2020/06/26 13:17:37 076-qeswebmonitor-demoqwam-TRACE-QESHost_elastic62-getNombreReference()|: search template avant:SearchRequest{searchType=DFS_QUERY_THEN_FETCH, indices=[webmonitor_phrases], indicesOptions=IndicesOptions[ignore_unavailable=false, allow_no_indices=true, expand_wildcards_open=true, expand_wildcards_closed=false, allow_aliases_to_multiple_indices=true, forbid_closed_indices=true, ignore_aliases=false, ignore_throttled=true], types=[], routing='null', preference='null', requestCache=null, scroll=null, maxConcurrentShardRequests=0, batchedReduceSize=512, preFilterShardSize=128, allowPartialSearchResults=null, localClusterAlias=null, getOrCreateAbsoluteStartMillis=-1, ccsMinimizeRoundtrips=true, source={"from":0,"size":20,"explain":false,"track_total_hits":2147483647,"aggregations":{"FOLDER_ID.verbatim":{"terms":{"field":"FOLDER_ID.verbatim","size":10,"min_doc_count":1,"shard_min_doc_count":0,"show_term_doc_count_error":false,"order":[{"_count":"desc"},{"_key":"asc"}]}},"SOURCE.verbatim":{"terms":{"field":"SOURCE.verbatim","size":10,"min_doc_count":1,"shard_min_doc_count":0,"show_term_doc_count_error":false,"order":[{"_count":"desc"},{"_key":"asc"}]}},"DATE_COLLECT":{"terms":{"field":"DATE_COLLECT","size":10,"min_doc_count":1,"shard_min_doc_count":0,"show_term_doc_count_error":false,"order":{"_key":"desc"}}},"QES_Person.verbatim":{"terms":{"field":"QES_Person.verbatim","size":20,"min_doc_count":1,"shard_min_doc_count":0,"show_term_doc_count_error":false,"order":[{"_count":"desc"},{"_key":"asc"}]}},"QES_Company.verbatim":{"terms":{"field":"QES_Company.verbatim","size":20,"min_doc_count":1,"shard_min_doc_count":0,"show_term_doc_count_error":false,"order":[{"_count":"desc"},{"_key":"asc"}]}},"QES_Location.verbatim":{"terms":{"field":"QES_Location.verbatim","size":20,"min_doc_count":1,"shard_min_doc_count":0,"show_term_doc_count_error":false,"order":[{"_count":"desc"},{"_key":"asc"}]}},"QES_Event.verbatim":{"terms":{"field":"QES_Event.verbatim","size":20,"min_doc_count":1,"shard_min_doc_count":0,"show_term_doc_count_error":false,"order":[{"_count":"desc"},{"_key":"asc"}]}},"QES_Concept.verbatim":{"terms":{"field":"QES_Concept.verbatim","size":20,"min_doc_count":1,"shard_min_doc_count":0,"show_term_doc_count_error":false,"order":[{"_count":"desc"},{"_key":"asc"}]}},"QES_Organization.verbatim":{"terms":{"field":"QES_Organization.verbatim","size":20,"min_doc_count":1,"shard_min_doc_count":0,"show_term_doc_count_error":false,"order":[{"_count":"desc"},{"_key":"asc"}]}}}}}||
    |2020/06/26 13:17:37 085-qeswebmonitor-demoqwam-ERROR-QESHost_elastic62-getNombreReference()|: Recuperation nombre reference impossible:Elasticsearch exception [type=parsing_exception, reason=Unknown key for a START_OBJECT in [script_score].]||

The request looks pretty much the same.
Cannot figure out what is going wrong and why I am getting this message:

Elasticsearch exception [type=parsing_exception, reason=Unknown key for a START_OBJECT in [script_score].]

Can you help me ?

2

Shouldn't your script begin with query:

script = "{ "query":" + " { "script_score" ?

3

It works !

Here is the correct snippet:

            // construction de la request
            request = new SearchTemplateRequest();
            request.setRequest(searchRequest);
            request.setScriptType(ScriptType.INLINE);

            //                            "            \"source\": \"cosineSimilarity([{{value_vector}}], doc['{{interro_vector}}']) + 1.0\"  " +
            //                            "            \"source\": \"cosineSimilarity([{{value_vector}}], '" +  enrg.getInterrogation() + "') + 1.0\"  " +
            //                            "            \"source\": \"cosineSimilarity([" + dense_vector + "], doc['" +  enrg.getInterrogation() + "']) + 1.0\"  " +

            script =
                "{ \"query\":   " +
                    "{" +
                    "    \"script_score\": {       " +
                    "        \"query\": { " +
                    "           \"query_string\" : { " +
                    "               \"query\" : \"" + query + "\", " +
                    "               \"default_field\" : \"tout\" " +
                    "            }  " +
                    "        },   " +
                    "        \"script\": { " +
                    "            \"source\": \"cosineSimilarity([" + dense_vector + "], doc['" +  enrg.getInterrogation() + "']) + 1.0\"  " +
                    "        } " +
                    "    }  " +
                    "}" +
                "}  ";
            request.setScript(script);

            Map<String, Object> scriptParams = new HashMap<>();
            //scriptParams.put("interro_vector", enrg.getInterrogation());
            //scriptParams.put("value_vector",  dvalues );
            request.setScriptParams(scriptParams);

I believed I tried this before calling for help but you proved me that it was not the case....

Could not make it work with aggregation but I know that you are working hard on full integration.

I was following your work since the day you made this post

It took me several months to fully apprehend the range of your post and be able to use it.
I feel ashamed to post silly questions about bad formed syntax regarding the elegance of the solution you implemented in Elasticsearch and the power it brings by marrying the speed of Elasticsearch and the power of BERT.

The results are simply atonishing ! The dream of semantic search with Elastic comes true !
Thank you Mayya for this wonderful accomplishment !

2 Likes
4

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.