Cum să folosești embeddings pentru căutare semantică

import pandas as pd

documente = [
    "Cum instalez Python pe Linux",
    "Ghid de instalare aplicatie",
    "Erori de compilare in C++]
]

cuvant = "instalare"
rezultate = [d for d in documente if cuvant in d.lower()]
print(rezultate)

from sentence_transformers import SentenceTransformer
import numpy as np

documente = [
    "Cum instalez Python pe Linux",
    "Ghid de instalare aplicatie",
    "Erori de compilare in C++",
    "Tutorial machine learning pentru incepatori]
]

model = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2")
embeddings_doc = model.encode(documente)

intrebare = "cum se configureaza un program"
embedding_intrebare = model.encode([intrebare])

similaritati = model.similarity(embedding_intrebare, embeddings_doc)[0]
cele_mai_bune = np.argsort(similaritati, axis=0)[::-1][:2]

for idx in cele_mai_bune:
    print(f"{documente[idx]} (scor: {similaritati[idx]:.3f})")

# Pentru romană / multilingv: "paraphrase-multilingual-MiniLM-L12-v2"
# Pentru engleză: "all-MiniLM-L6-v2" (mai rapid)
# Via OpenAI: text-embedding-3-small

from sentence_transformers import SentenceTransformer
model = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2")

import numpy as np
from sentence_transformers import SentenceTransformer

model = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2")
documente = ["Text 1", "Text 2", "Text 3"]
embeddings = model.encode(documente, show_progress_bar=True)

# Salvează vectorii
np.save("embeddings.npy", embeddings)

def cauta_semantic(intrebare, documente, embeddings, top_k=3):
    emb_intrebare = model.encode([intrebare])
    similaritati = model.similarity(emb_intrebare, embeddings)[0]
    indici = np.argsort(similaritati, axis=0)[::-1][:top_k]
    return [(documente[i], similaritati[i]) for i in indici]

rezultate = cauta_semantic("eroare la compilare", documente, embeddings)
for doc, scor in rezultate:
    print(f"{doc} ({scor:.3f})")