Simply create the semantic_search.py in which i linked it app/__init__py
To ensure that my application run on flask run i installed python-dotenv and in .flaskenv i set FLASK_APP=semantic_search.py and FLASK_DEBUG=1
Now in app folder I created routes.py in which I just render the templates.
In utils/semantic_search.py I wrote the main logic of search which I will explain later.
data folder have NCO-2015 data which I imported it in semantic_search.py
And now my application is ready!!
Be sure to install flask,dotenv,SentenceTransformer
I have selected all-MiniLM-L6-v2 model
Now created data corpus with title and description
Created corpus embeddings using model.encode()
Now to get similar occupations and get top_k=5 results
First user type query which i encode it to the same embedding used to encode corpus
Then I find cosine similarities between corpus embeddings and query embedding
top_hits will store the top top_k numbers of similar corpus embeddings
Now top hits similarities have format in which there is values and indices
now item in selected by idx
result is appended in which code, title, description and score is appended by score.item()
Return the result which will be stored in results in route.py and then passed along render_template.