Why is my query to an index in azure cognitive search not always returning a result in the field @search.answers?
I am sending the following hybrid query to my index:
"search": "Why is Copenhagen a regional hub?",
"vectorQueries": [
{
"kind": "text",
"text": "Why is Copenhagen a regional hub?",
"k": 5,
"fields": "vector"
}
],
"queryType": "semantic",
"semanticConfiguration": "my-config",
"captions": "extractive",
"answers": "extractive|count-3",
"queryLanguage": "en-US"
}
This query does not have the field @search.answers in the response filled which can be seen here:
For most other search strings the required field is filled out with a text and highlight field.
According to the documentation an empty @search.answers field means that the search result is undetermined, which makes no sense because I get back some result values (even with high scores).
I am asking if anyone knows why this could happen. I wanted to retrieve the @search.answers because I have the feeling that this offers me the best search results.
The call looks right (“answers”: “extractive|count-3”).
Answers are only returned when there is a definitive answer contained within the corpus of the documents retrieved. Only verbatim text from a document can appear as an answer. High search result scores but unpopulated “@search.answers” array suggest that the documents returned are relevant but may not contain a definitive and direct answer to the question. The @search.rerankerScore isn't visible in the screenshot so it is unclear how the ranker scored this result.
The relevant questions are:
Return a semantic answer - Azure AI Search | Microsoft Learn