Update app.py

app.py

@@ -1,5 +1,5 @@
 import gradio as gr
-from sentence_transformers import SentenceTransformer
+from sentence_transformers import SentenceTransformer, util
 import pandas as pd
 from rapidfuzz import fuzz, process
 
@@ -13,9 +13,9 @@ try:
     df = pd.read_csv(data_file, nrows=1000)  # Limit rows for testing
 except FileNotFoundError:
     df = pd.DataFrame({
-        "ProductName": ["Gucci Shoes", "Nike Sneakers", "Louis Vuitton Handbag"],
-        "Category": ["Shoes", "Bags"],
-        "SubCategory": ["Sneakers", "Totes"]
+        "ProductName": ["Gucci Sunglasses", "Nike Sneakers", "Louis Vuitton Handbag"],
+        "Category": ["Accessories", "Shoes"],
+        "SubCategory": ["Sunglasses", "Totes"]
     })  # Fallback sample data
 
 # Extract relevant fields
@@ -25,24 +25,35 @@ subcategories = df["SubCategory"].dropna().unique().tolist()
 
 # Merge into one dataset for autocomplete
 autocomplete_data = product_names + categories + subcategories
+autocomplete_data = [str(item).strip('"') for item in autocomplete_data]  # Clean text
 
-# Clean data by removing unnecessary characters
-autocomplete_data = [str(item).strip('"') for item in autocomplete_data]
+# Encode all items in the dataset into embeddings
+autocomplete_embeddings = model.encode(autocomplete_data, convert_to_tensor=True)
 
-# Autocomplete function
+# Synonym Expansion Function
+def find_synonym(word, top_n=1):
+    """Finds the most similar word in the dataset based on embeddings."""
+    query_embedding = model.encode(word, convert_to_tensor=True)
+    results = util.semantic_search(query_embedding, autocomplete_embeddings, top_k=top_n)
+    return [autocomplete_data[result['corpus_id']] for result in results[0]]
+
+# Autocomplete function with synonym handling
 def autocomplete(query):
     if not query.strip():
         return []  # Avoid empty queries
 
-    # Fuzzy matching with typo tolerance
-    matches = process.extract(query, autocomplete_data, scorer=fuzz.partial_ratio, limit=5)
+    # Find synonyms dynamically
+    synonyms = find_synonym(query, top_n=3)
+    
+    # Perform fuzzy matching with synonyms included
+    matches = process.extract(query, autocomplete_data + synonyms, scorer=fuzz.partial_ratio, limit=5)
 
-    # Return list of suggestions (Gradio will display them in separate lines)
+    # Return list of suggestions
     return [match[0] for match in matches]
 
 # Gradio interface
 with gr.Blocks() as demo:
-    gr.Markdown("### Improved Autocomplete for Luxury Products")
+    gr.Markdown("### AI-Powered Autocomplete with Synonyms")
 
     query = gr.Textbox(label="Start typing for autocomplete")
     autocomplete_output = gr.Textbox(label="Autocomplete Suggestions", lines=5, interactive=False)