Versions (relevant - OpenSearch/Dashboard/Server OS/Browser):
2.11
Describe the issue:
Is it possible to use a hybrid query to perform a federated search?
I would like to combine the results of my regular text search with the results coming from another system.
I potentially could feed the results into the query using a function score query, which will give the rows to return and assign the specific scores.
A query like the following could be used:
http://xxx.com:9201/idxname/_search?pretty&search_pipeline=hybrid-normalization-pipeline
{
“_source”: {
“includes”: [
“Name”
]
},
“sort”: [
{
“_score”: “desc”
}
],
“query”: {
“hybrid”: {
“queries”: [
{
“function_score”: {
“query”: {
“bool”: {
“must”: [
{
“terms”: {
“_id”: [
“300100613471622”,
“300100613471877”
]
}
}
]
}
},
“functions”: [
{
“filter”: {
“term”: {
“_id”: {
“value”: 300100613471622
}
}
},
“weight”: 560
},
{
“filter”: {
“term”: {
“_id”: {
“value”: 300100613471877
}
}
},
“weight”: 133
}
],
“boost_mode”: “replace”
}
},
{
“simple_query_string”: {
“query”: “test”,
“fields”: [
“Name”
],
“default_operator”: “and”
}
}
]
}
}
}
While this returns results, the next task is to get the results to return in a usefull sort order.
Can the normalization processor be used for this process? I am passing:
&search_pipeline=hybrid-normalization-pipeline
The normalization-processor was defined as:
{
“description”: “Post processor for hybrid search”,
“phase_results_processors”: [
{
“normalization-processor”: {
“normalization”: {
“technique”: “min_max”
},
“combination”: {
“technique”: “arithmetic_mean”,
“parameters”: {
“weights”: [
0.3,
0.7
]
}
}
}
}
]
}
However - whenever I run the search this processor appears to have no effect on the scores at all.
The results coming from the function score part of the query always have the scores assigned from the functionscore query.
Is there any way to normalize the scores from both these searches to achieve better more relvant results?
Configuration:
Relevant Logs or Screenshots: