[Python] Highlighting retrieved nested text chunks in elastic search?

i have a hybrid search index with a mapping like:

{
...
"raw_content": {
"type": "text",
"term_vector": "with_positions_offsets"
},
"text_passages": {
"type": "nested",
"properties": {
"vector": {"type": "dense_vector": "dims": 768},
"content": {"type":"text", "term_vector": "with_positions_offsets"}
...
}
}
...
}


i have a query that uses knn search to get chunks that match some query:

"knn": {
"field": "text_passages.vector",
"query_vector": image_embedding,
"k": 10,
"num_candidates": 10 * 3,
"boost": 1,
"inner_hits": {
"name": "matching_passages",
"size": 2,
"_source": ["text_passages.content", "text_passages.chunk_num"],
}
}


My hope is to use the highlighter to highlight the two returned inner hits in this query, but for the life of me i cannot figure this out. I've tried placing a highlighter:

"highlight": {
"order": "score",
"fields": {
"text_passages.content": {
"type":"unified",
"fragment_size": 150,
"number_of_fragments":3 ,
"pre_tags": ["<strong>"],
"post_tags": ["</strong>"]
}
}
}


inside knn, inside inner_hits, outside knn, i cant seem to get this work which is confusing because the docs say they support highlighting in inner_hits. Has anyone done before? am i approaching embeddings wrong, and the text chunks/vectors of a document should be in an array instead of a nested object? very confused. any help is appreciated.

btw im using elastic 8.17 and python 3.12.7.

Continue reading...