Elastic search searms to prioritize results with an isolated search term during a full text search - elasticsearch

Question

I am having problems with Elastic Search. It seams the search term is being isolated in search results.
We have a large subtitle database that was indexed using Elastic Search.
It seams however, that our searches prioritize search results where the search term is isolated.
Ie: the search for "Eat" produces:
Oh, skydiving. // Skydiving. // Oh, I got that one. // Eating crazy. // Eating, eating. // Just pass, just pass. // You guys suck at that. // What was that? // Synchronized swimming
AND
it's my last night so we're gonna live // life like there's no tomorrow. // - I think I'd just wanna, // - Eat. // - Bring all the food, // whether it's Mcdonald's, whether it's, // - Ice cream.
We need to INSTEAD prioritize search results where the searchTerm is found WITHIN the sentence, rather than just on its own.
I need help determining what needs to be fixed - The Mapping, the filters, the tokenizers etc.
Here are my settings:
static public function getSettings(){
return [
'number_of_shards' => 1,
'number_of_replicas' => 1,
'analysis' => [
'filter' => [
'filter_stemmer' => [
'type' => 'stemmer',
'language' => 'english'
]
],
'analyzer' => [
'text_analyzer' => [
'type' => 'custom',
"stopwords" => [],
'filter' => ['lowercase', 'filter_stemmer','stemmer'],
'tokenizer' => 'standard'
],
]
]
];
}
and here are my mapping:
https://gist.github.com/firecentaur/d0e1e196f7fddbb4d02935bec5592009
And here is my search
https://gist.github.com/firecentaur/5ac97bbd8eb02c406d6eecf867afc13c
What am I doing wrong?

Answer 1

This behavior must be caused by the TL/IDF algorithm.
If a query match a field, it will be more important if their is few words in the field.
If you want to adapt this to your use case, you can use a function_score query.
This post should help you to find a solution.
How can I boost the field length norm in elasticsearch function score?