Addsome changes 5

app.py

@@ -12,9 +12,25 @@ with open("model/sbert_embeddings2.pkl", "rb") as f:
     book_embeddings = pickle.load(f)
 
 # Load models
+retriever_model = SentenceTransformer("all-mpnet-base-v2")  # More accurate than MiniLM
 reranker_model = CrossEncoder("cross-encoder/stsb-roberta-large")  # More accurate ranking
 
-recommender = BookRecommender()
+def retrieve_candidates(book_title, top_n=10):
+    """Retrieve top-N similar books using SBERT embeddings"""
+    if book_title not in df["title"].values:
+        return ["Error: Book title not found in dataset!"]
+
+    # Get book index
+    book_idx = df[df["book_name"] == book_title].index[0]
+
+    # Compute cosine similarity
+    query_embedding = book_embeddings[book_idx]
+    scores = util.cos_sim(query_embedding, book_embeddings)[0]
+
+    # Get top-N similar books (excluding the book itself)
+    top_indices = torch.argsort(scores, descending=True)[1:top_n+1]
+
+    return df.iloc[top_indices][["title", "summary"]].values.tolist()
 
 def rerank_books(query_title, candidates):
     """Re-rank books using a cross-encoder"""
@@ -28,15 +44,15 @@ def rerank_books(query_title, candidates):
 
 def recommend_books(book_title):
     """Complete recommendation pipeline"""
-    candidates = recommender.recommend(book_title, top_n=5)
+    candidates = retrieve_candidates(book_title, top_n=10)
     if isinstance(candidates, list) and "Error" in candidates[0]:
         return candidates[0]
     
-    return candidates
+    return rerank_books(book_title, candidates)
 
 # Gradio Interface
 with gr.Blocks() as demo:
-    gr.Markdown("# 📚 Content-Based Book Recommendation")
+    gr.Markdown("# Content-Based Book Recommendation")
     gr.Markdown("Enter a book title to find similar books based on summaries.")
     
     with gr.Row():