"Delicious meat restaurant near Seomyeon"
I'm trying to implement retrieving answers from pre-prepared data when you search like this.
The method AI suggested was,
========================
[stores.json]
↓ (run once)
[Textualization + Embedding Generation]
↓
[FAISS Vector Index]
========================
[User Question]
↓
[Question Embedding]
↓
[FAISS Similarity Search]
↓
[Operating / Closed Filter]
↓
[Final Results]
========================
That's what it was. I don't really understand it, but I completed it the way Claude Code suggested and ran it, but there are too many errors.
It can't search using related words that aren't exact words in the data.
"Near Seomyeon" -> Couldn't retrieve areas close to Seomyeon
"Delicious" -> Excluded because this word doesn't exist in the data
"Meat restaurant" -> A raw fish restaurant appears. It only searches when you say "pork" or "beef".
What I'm thinking now is that this will only be limited to "narrow-scope help".
"Near Seomyeon" -> Extract only the word "Seomyeon"
"Delicious" -> Create a set of words like "delicious", "popular", "recommended", etc., and substitute them with a representative word
"Meat restaurant" -> Create a set of words like "meat restaurant", "meat", "pork belly restaurant", "raw meat restaurant", etc., and substitute them with a representative word
I thought AI would understand like humans and search the data for me, but... is this the best approach?
I'm looking for advice on whether there's a better way.
