Update app.py

app.py

@@ -7,6 +7,10 @@ from langchain_community.embeddings import HuggingFaceEmbeddings
 from langchain_groq import ChatGroq
 from langchain.memory import ConversationBufferMemory
 from langchain.chains import ConversationalRetrievalChain
+from nltk import word_tokenize, FreqDist, sent_tokenize
+from textblob import TextBlob
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.decomposition import NMF
 import os
 import nltk
 nltk.download('punkt')
@@ -31,11 +35,56 @@ def setup_vectorstore(documents):
     vectorstores = FAISS.from_documents(doc_chunks, embeddings)
     return vectorstores
 
-def create_chain(vectorstores):
+# Function to extract mentality traits (viewpoints, reasoning patterns, etc.)
+def extract_mentality_traits(documents):
+    text = "\n".join(documents)
+    blob = TextBlob(text)
+
+    # Analyze tone
+    sentiment = blob.sentiment.polarity
+    tone = "neutral"
+    if sentiment > 0.2:
+        tone = "positive"
+    elif sentiment < -0.2:
+        tone = "negative"
+
+    # Common phrases and reasoning patterns
+    words = word_tokenize(text.lower())
+    fdist = FreqDist(words)
+    common_phrases = [word for word, count in fdist.most_common(10) if len(word) > 3]
+
+    # Topic Modeling to identify key viewpoints
+    vectorizer = TfidfVectorizer(stop_words='english')
+    doc_term_matrix = vectorizer.fit_transform([text])
+    nmf = NMF(n_components=2, random_state=1)
+    nmf.fit(doc_term_matrix)
+    topics = []
+    for topic in nmf.components_:
+        topic_words = [vectorizer.get_feature_names_out()[i] for i in topic.argsort()[-5:]]
+        topics.append(", ".join(topic_words))
+
+    mentality_traits = {
+        "tone": tone,
+        "common_phrases": common_phrases,
+        "reasoning_pattern": "deductive" if sentiment > 0 else "inductive",
+        "key_viewpoints": topics
+    }
+    return mentality_traits
+
+def create_chain(vectorstores, mentality_traits):
+    # Create style prompt dynamically based on mentality traits
+    style_prompt = (
+        f"Think and respond in the same mindset as the person from the uploaded document. "
+        f"The author's tone is generally {mentality_traits['tone']}. "
+        f"Their reasoning style is mostly {mentality_traits['reasoning_pattern']} and they often discuss themes such as {', '.join(mentality_traits['key_viewpoints'])}. "
+        f"Use phrases like {', '.join(mentality_traits['common_phrases'])}. "
+        f"Respond as if you are the author, mirroring their values, reasoning style, and mentality in all replies."
+    )
+    
     llm = ChatGroq(
         api_key=secret,
         model="llama-3.1-70b-versatile",
-        temperature=0.7  # Slightly higher for more natural, expressive language
+        temperature=0.7
     )
     retriever = vectorstores.as_retriever()
     memory = ConversationBufferMemory(
@@ -48,6 +97,7 @@ def create_chain(vectorstores):
         llm=llm,
         retriever=retriever,
         memory=memory,
+        style_prompt=style_prompt,
         verbose=True
     )
     return chain
@@ -72,12 +122,14 @@ if uploaded_file:
         f.write(uploaded_file.getbuffer())
 
     # Load and process the PDF document
+    documents = load_documents(file_path)
     if "vectorstores" not in st.session_state:
-        st.session_state.vectorstores = setup_vectorstore(load_documents(file_path))
+        st.session_state.vectorstores = setup_vectorstore(documents)
 
-    # Create the conversational chain with style adaptation
+    # Extract mentality traits and create the conversational chain with mental adaptation
+    mentality_traits = extract_mentality_traits(documents)
     if "conversation_chain" not in st.session_state:
-        st.session_state.conversation_chain = create_chain(st.session_state.vectorstores)
+        st.session_state.conversation_chain = create_chain(st.session_state.vectorstores, mentality_traits)
 
 for message in st.session_state.chat_history:
     with st.chat_message(message["role"]):
@@ -90,17 +142,11 @@ if user_input:
     with st.chat_message("user"):
         st.markdown(user_input)
 
-    # Add prompt for style adaptation
-    style_prompt = (
-         f"Respond in the same style and with the same mentality as the person in the uploaded document. "
-         f"Analyze the PDF to understand the author’s way of thinking, attitudes, personality, and values. "
-         f"Replicate this mentality by capturing their perspective, reasoning style, and tone in your response. "
-         f"Answer the question below as if you were the author, preserving their personality, mindset, and outlook:\n\n{user_input}"
-                   )
-
-    response = st.session_state.conversation_chain({"question": style_prompt})
+    # Use the conversation chain to generate the response
+    response = st.session_state.conversation_chain({"question": user_input})
 
     assistant_response = response["answer"]
     st.session_state.chat_history.append({"role": "assistant", "content": assistant_response})
     with st.chat_message("assistant"):
         st.markdown(assistant_response)
+