app.py

import streamlit as st
import torch
from sentence_transformers import SentenceTransformer

# Load SBERT model (choose a suitable model from https://www.sbert.net/docs/pretrained_models.html)
@st.cache_resource
def load_sbert():
    model = SentenceTransformer('all-MiniLM-L6-v2')  # Example model
    return model

model = load_sbert()


def calculate_similarity(word1, word2):
    embeddings1 = model.encode(word1)
    embeddings2 = model.encode(word2)

    # Convert NumPy arrays to tensors
    embeddings1 = torch.tensor(embeddings1)
    embeddings2 = torch.tensor(embeddings2)

    cos_sim = torch.nn.functional.cosine_similarity(embeddings1, embeddings2, dim=0)
    return cos_sim.item()

def display_top_5(similarities):
    # Sort by similarity (descending)
    top_5_similarities = sorted(similarities, key=lambda item: item[1], reverse=True)[:5]

    st.subheader("Top 5 Most Similar Words:")
    for word, similarity in top_5_similarities:
        st.write(f"- '{word}': {similarity:.4f}")

        
# Streamlit interface
st.title("Sentence Similarity Checker")

reference_word = st.text_input("Enter the reference Sentence:")
word_list = st.text_area("Enter a list of sentences or phrases (one word per line):")

if st.button("Analyze"):
    if reference_word and word_list:
        # Calculate similarities for the reference phrase against the word list
        similarities = []
        for word in word_list.splitlines():
            similarity = calculate_similarity(reference_word, word)
            similarities.append((word, similarity)) 

        #  Find top 5 (We should only do this once outside the loop)
        display_top_5(similarities) 
    else:
        st.warning("Please enter a reference word and a list of words.")