Update app.py

app.py

@@ -8,6 +8,9 @@ import nltk
 from openai.embeddings_utils import cosine_similarity
 import spacy
 from spacy.cli import download
+import os
+
+openai.api_key = os.getenv("OPENAI_API_KEY")
 nltk.download('punkt')
 nltk.download('punkt_tab')
 
@@ -25,7 +28,7 @@ def bm25_rank(query, df, n=15):
   return top_results
 
 def search(query, df):
-  n = 1
+  n = 5
   query_embedding = model.encode(query)
   df = bm25_rank(query, df)
   df["similarity"] = df.embeddings.apply(lambda x: cosine_similarity(x, query_embedding.reshape(768,-1)))
@@ -50,7 +53,7 @@ def search(query, df):
             "image_url": smalldf.hotel_image[r],
             "score": smalldf.rate[r],
             "description": smalldf.hotel_description[r]
-            # "Relevant_reviews": [ smalldf.review_text[s] for s in smalldf.index]
+            "relevant_reviews": [ smalldf.review_text[s] for s in smalldf.index]
           })
           hlist.append(results.hotel_name[r])
   return resultlist
@@ -78,6 +81,7 @@ def get_hotel_info(query):
             'hotel_name': result['hotel_name'],
             'score': result['score'],
             'description': result['description']
+            'relevent_reviews': result['relevant_reviews']
         })
     return response
 
@@ -88,30 +92,66 @@ def get_hotel_info(query):
 #     response += f"![Hotel Image]({hotel_info['image_url']})\n"
 #     return response
 
-def chatbot_response(query):
-    hotel_infos = get_hotel_info(query)
-    image = hotel_infos[0]['image_url']
-    name = hotel_infos[0]['hotel_name']
-    score = df[df["hotel_name"] == name]['rate'].mean()
-    description = hotel_infos[0]['description']
-    return image, name, score, description
+def generate_answer(query, context):
+    prompt = f"""
+    Based on the following query from a user, please generate a detailed answer based on the context
+    focusing on which is the top hotel based on the query. You should respond as if you are a travel agent and are conversing with the
+    user in a nice cordial way. Remove any special characters and (\\n), make the output clean and concise.
+
+    ###########
+    query:
+    "{query}"
+
+    ########
+
+    context:"
+    "{context}"
+    #####
+
+    Return in Markdown format with each hotel highlighted.
+    """
+
+    messages = [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": prompt}
+    ]
+    response = openai.ChatCompletion.create(
+        model="gpt-4o-mini",
+        max_tokens=1500,
+        n=1,
+        stop=None,
+        temperature=0.2,  # Higher temperature means more creative or more hallucination
+        messages=messages
+    )
+
+    # Extract the generated response from the API response
+    generated_text = response.choices[0].message['content'].strip()
+
+    return generated_text
+
+def chatbot_response(message, history):
+    hotel_infos = get_hotel_info(message)
+    if hotel_infos:
+        context = "\n".join([
+            f"Hotel Name: {info['hotel_name']}, Score: {info['score']}, Description: {info['description']}, Reviews: {info['relevant_reviews']}"
+            for info in hotel_infos
+        ])
+        response = generate_answer(message, context)
+    else:
+        response = "No results found."
+
+    history.append((message, response))
+    return history, history
 
 with gr.Blocks() as interface:
-    with gr.Row():
-        with gr.Column(scale=1):
-            query_input = gr.Textbox(label="Enter Your Query")
-            submit_button = gr.Button("Submit")
-        
-        with gr.Column(scale=2):
-            image_output = gr.Image(label="Hotel Image")
-            name_output = gr.Textbox(label="Hotel Name")
-            score_output = gr.Textbox(label="Score")
-            description_output = gr.Textbox(label="Description")
+    chatbot = gr.Chatbot(label="Hotel Search Chatbot")
+    query_input = gr.Textbox(label="Ask me about hotels!")
+    submit_button = gr.Button("Send")
 
     submit_button.click(
-        fn=chatbot_response, 
-        inputs=query_input, 
-        outputs=[image_output, name_output, score_output, description_output]
+        fn=chatbot_response,
+        inputs=[query_input, chatbot],
+        outputs=[chatbot, chatbot]
     )
 
 interface.launch()