How to recognize a brand in a text with Elasticsearch - elasticsearch

Question

I have been stuck for 2 days on this and I am sure it can be done with Elasticsearch. Any help would be really appreciated!
I receive products from various sources and I want to integrate them to my current inventory.
Products reach me in the form of text. They generally have a brand and a name:
1000 Stories Zinfandel Bourbon Barrel Aged
1000 Stories Gold Rush Red Blend Bourbon Barrel
1000 Stories Cabernet Bourbon Barrel Aged
^^^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Brand Product Name
But one can be missing or they can be mixed. That's why I want to recognize what is there and what is missing.
I generally already know the brands and products from my inventory. How can I get Elasticsearch to tell me which is what?
Ideally, I would get something like:
<brand>1000 Stories</brand> <name>Zinfandel Bourbon Barrel Aged</name>
<brand>1000 Stories</brand> <name>Gold Rush Red Blend Bourbon Barrel</name>
<brand>1000 Stories</brand> <name>Cabernet Bourbon Barrel Aged</name>
Though just recognizing the brand would already be a big step.
I was hoping to make it work with phrase suggestion, because it already includes the matches in the results. I have tried to map my brands and names with all those analyzers:
settings: {
analysis: {
analyzer: {
trigram: {
type: "custom",
tokenizer: "standard",
filter: ["lowercase","shingle"]
},
reverse: {
type: "custom",
tokenizer: "standard",
filter: ["lowercase","reverse"]
},
raw_analyzer: {
tokenizer: "keyword",
filter: [
"lowercase",
"asciifolding"
]
}
},
filter: {
shingle: {
type: "shingle",
min_shingle_size: 2,
max_shingle_size: 3
}
}
}
},
mappings: {
product: {
properties: {
brand: {
type: "text",
fields: {
trigram: {
type: "text",
analyzer: "trigram"
},
reverse: {
type: "text",
analyzer: "reverse"
},
raw: {
type: "text",
analyzer: "raw_analyzer"
}
}
}
And searched them with each analyzer:
phrase_name_suggestion: {
phrase: {
field: 'name.trigram',
max_errors: 0.99,
size: 5,
gram_size: 3,
direct_generator: [
{
field: "name.trigram",
suggest_mode: "always"
},
{
field: "name.reverse",
suggest_mode: "always",
pre_filter: "reverse",
post_filter: "reverse"
} ],
highlight: {
pre_tag: "<name>",
post_tag: "</name>"
}
}
},
phrase_name_raw_suggestion: {
phrase: {
field: 'name.raw',
max_errors: 0.99,
size: 5,
gram_size: 3,
highlight: {
pre_tag: "<name>",
post_tag: "</name>"
}
}
}
I am only getting random suggestions on the wrong terms, or no results at all. Like a term of a brand suggested for a term of a name, instead of just recognizing the brand.
Note in case it narrows the options: names are manually input, so I can get all variety of text: missing name, typos ("Sinfandel" for "Zinfandel"), abbreviations ("cab sauv" for "cabernet sauvignon"), ... That's a separate issue but if it can be included into this solution I'll happily take it.
I am running Elasticsearch 6.4.2. I can work with a more recent version if needed.