apps/similarity.py

import streamlit as st
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
from sentence_transformers import SentenceTransformer

def app():
    st.title("Text Similarity")
    with st.expander("ℹ️ - About this app", expanded=True):

        st.write(
            """     
            Information cartography - Get your word/phrase/sentence/paragraph embedded and visualized.
            The (English) sentence-transformers model "all-MiniLM-L6-v2" maps sentences & paragraphs to a 384 dimensional dense vector space This is normally used for tasks like clustering or semantic search, but in this case, we use it to calculate the (cosine) similarity. The sentence transformer is context sensitive and works best with whole sentences, to account for that we extend your text with "The book is about <text>" if its less than 15 characters.
            
            Simply put in your text and press COMPARE, the higher the similarity the closer the text in the embedding space (max 1).
            """)
    
    model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
    
    with st.container():
        col1, col2 = st.columns(2)
        with col1:
            word_to_embed1 = st.text_input("Text 1", value="The most vulnerable countries are seeking ‘loss and damage’ compensation from the biggest polluters.",)
        with col2:                                 
            word_to_embed2 = st.text_input("Text 2", value="COP27 opens amid compounding crises of war, warming and economic instability.",)
            
    if st.button("Comapre"):
        with st.spinner("Embedding comparing  your inputs"):
            
            document = [word_to_embed1 ,word_to_embed2]
            documents_embed = ["The book is about "+ wte for wte in document if len(wte) <15]
            #Encode paragraphs
            document_embeddings = model.encode(documents_embed , show_progress_bar=False)
            #Compute cosine similarity between labels sentences and paragraphs
            similarity_matrix = cosine_similarity(document_embeddings)
            
            st.write("Text similarity:", round(similarity_matrix[0][1]*100,2),"%")