app.py

import streamlit as st
from sentence_transformers import SentenceTransformer, util
import pandas as pd
import numpy as np
from ast import literal_eval

# Load the model
model_name = "./Embedder-Typosquat" 
model = SentenceTransformer(model_name)

# Load the domains and embeddings
domains_df = pd.read_csv('domains_embs.csv')
domains_df.embedding = domains_df.embedding.apply(literal_eval)
corpus_domains = domains_df.domain.to_list()
corpus_embeddings = np.stack(domains_df.embedding.values).astype(np.float32)  # Ensure embeddings are float32

# Streamlit App
st.title("Mining Potential Legitimate Domains from a Typosquatted Domain")
st.write("Enter a potential typosquatted domain and select the number of top results to retrieve.")

# User Inputs
domain = st.text_input("Potential Typosquatted Domain")
top_k = st.number_input("Top K Results", min_value=1, max_value=50, value=5, step=1)

# Button to trigger search
if st.button("Search for Legitimate Domains"):
    if domain:
        # Perform Semantic Search
        query_emb = model.encode(domain).astype(np.float32)  # Ensure query embedding is also float32
        semantic_res = util.semantic_search(query_emb, corpus_embeddings, top_k=top_k)[0]
        ids = [r['corpus_id'] for r in semantic_res]
        scores = [r['score'] for r in semantic_res]
        
        # Create a DataFrame for the results
        res_df = domains_df.loc[ids,'domain'].copy()
        res_df['score'] = scores
        
        # Display the result DataFrame
        st.write("Mined Domains:")
        st.dataframe(res_df)
    else:
        st.warning("Please enter a domain to perform the search.")