You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Handling synonyms is like an ABC thing when it comes to semantic search, however the current neural search plugin does not handle synonyms.
I suggest a feature to enable that out of the box without really going through the expense of fine-tuning embedding model. One simple way to do it is by simply replacing a word with both the word and its synonyms before generate a vector for them. The source text remains the same but the embedding gets to reflect the synonyms. The option can also be disabled in case the embedding model already understands synonyms.
Example:
Src="OS is amazing"
dst="[[Open Search]]os is amazing"
vec= getEmbedding(dst)
POST test/_doc/1
{"content":src,
"Vec":vec
}
The text was updated successfully, but these errors were encountered:
@asfoorial This is quite interesting! Do you happen to have any benchmark tests that demonstrate how handling synonyms improves search accuracy in semantic search?
Handling synonyms is like an ABC thing when it comes to semantic search, however the current neural search plugin does not handle synonyms.
I suggest a feature to enable that out of the box without really going through the expense of fine-tuning embedding model. One simple way to do it is by simply replacing a word with both the word and its synonyms before generate a vector for them. The source text remains the same but the embedding gets to reflect the synonyms. The option can also be disabled in case the embedding model already understands synonyms.
Example:
Src="OS is amazing"
dst="[[Open Search]]os is amazing"
vec= getEmbedding(dst)
POST test/_doc/1
{"content":src,
"Vec":vec
}
The text was updated successfully, but these errors were encountered: