Update pages/3_Earnings_Semantic_Search_🔎_.py

pages/3_Earnings_Semantic_Search_🔎_.py

@@ -10,6 +10,8 @@ st.sidebar.header("Semantic Search")
 
 st.markdown("Earnings Semantic Search with LangChain, OpenAI & SBert")
 
+starter_message = "Ask me anything about the Earnings Call!"
+
 st.markdown(
  """
  <style>
@@ -59,7 +61,8 @@ st.markdown(
 )
 
 bi_enc_dict = {'mpnet-base-v2':"all-mpnet-base-v2",
- 'instructor-base': 'hkunlp/instructor-base'}
+ 'instructor-base': 'hkunlp/instructor-base',
+ 'FlagEmbedding': 'BAAI/bge-base-en'}
 
 search_input = st.text_input(
  label='Enter Your Search Query',value= "What key challenges did the business face?", key='search')
@@ -73,69 +76,89 @@ overlap_size = 50
 
 try:
 
- if search_input:
- 
- if "sen_df" in st.session_state and "earnings_passages" in st.session_state:
  
- ## Save to a dataframe for ease of visualization
- sen_df = st.session_state['sen_df']
-
- title = st.session_state['title']
+ if "sen_df" in st.session_state and "earnings_passages" in st.session_state:
+ 
+ ## Save to a dataframe for ease of visualization
+ sen_df = st.session_state['sen_df']
+
+ title = st.session_state['title']
+
+ earnings_text = st.session_state['earnings_passages']
+
+ print(f'earnings_to_be_embedded:{earnings_text}')
+
+ st.session_state.eval_set = generate_eval(
+ earnings_text, 10, 3000)
+
+ # Display the question-answer pairs in the sidebar with smaller text
+ for i, qa_pair in enumerate(st.session_state.eval_set):
+ st.sidebar.markdown(
+ f"""
+ <div class="css-card">
+ <span class="card-tag">Question {i + 1}</span>
+ <p style="font-size: 12px;">{qa_pair['question']}</p>
+ <p style="font-size: 12px;">{qa_pair['answer']}</p>
+ </div>
+ """,
+ unsafe_allow_html=True,
+ )
+
+ embedding_model = bi_enc_dict[sbert_model_name]
+ 
+ with st.spinner(
+ text=f"Loading {embedding_model} embedding model and Generating Response..."
+ ):
+ 
+ docsearch = create_vectorstore(earnings_text,title, embedding_model)
 
- earnings_text = st.session_state['earnings_passages']
+ memory, agent_executor = create_memory_and_agent(search_input,docsearch)
 
- print(f'earnings_to_be_embedded:{earnings_text}')
+ if "messages" not in st.session_state or st.sidebar.button("Clear message history"):
+ st.session_state["messages"] = [AIMessage(content=starter_message)]
 
- st.session_state.eval_set = generate_eval(
- earnings_text, 10, 3000)
+ for msg in st.session_state.messages:
+ if isinstance(msg, AIMessage):
+ st.chat_message("assistant").write(msg.content)
+ elif isinstance(msg, HumanMessage):
+ st.chat_message("user").write(msg.content)
+ memory.chat_memory.add_message(msg)
 
- # Display the question-answer pairs in the sidebar with smaller text
- for i, qa_pair in enumerate(st.session_state.eval_set):
- st.sidebar.markdown(
- f"""
- <div class="css-card">
- <span class="card-tag">Question {i + 1}</span>
- <p style="font-size: 12px;">{qa_pair['question']}</p>
- <p style="font-size: 12px;">{qa_pair['answer']}</p>
- </div>
- """,
- unsafe_allow_html=True,
+ if user_question := st.chat_input(placeholder=starter_message):
+ st.chat_message("user").write(user_question)
+ 
+ with st.chat_message("assistant"):
+ 
+ st_callback = StreamlitCallbackHandler(st.container())
+ 
+ response = agent_executor(
+ {"input": user_question, "history": st.session_state.messages},
+ callbacks=[st_callback],
+ include_run_info=True,
  )
 
- embedding_model = bi_enc_dict[sbert_model_name]
- 
- with st.spinner(
- text=f"Loading {embedding_model} embedding model and Generating Response..."
- ):
+ answer = response["output"]
  
- docsearch = process_corpus(earnings_text,title, embedding_model)
-
- result = embed_text(search_input,docsearch)
-
-
- references = [doc.page_content for doc in result['source_documents']]
-
- answer = result['answer']
-
- sentiment_label = gen_sentiment(answer)
+ st.session_state.messages.append(AIMessage(content=answer))
  
- ##### Sematic Search #####
- 
- df = pd.DataFrame.from_dict({'Text':[answer],'Sentiment':[sentiment_label]})
- 
- 
- text_annotations = gen_annotated_text(df)[0] 
- 
- with st.expander(label='Query Result', expanded=True):
- annotated_text(text_annotations)
+ st.write(answer)
+ 
+ memory.save_context({"input": user_question}, response)
  
- with st.expander(label='References from Corpus used to Generate Result'):
- for ref in references:
- st.write(ref)
+ st.session_state["messages"] = memory.buffer
  
- else:
+ run_id = response["__run"].run_id
+ 
+ col_blank, col_text, col1, col2 = st.columns([10, 2, 1, 1])
+
  
- st.write('Please ensure you have entered the YouTube URL or uploaded the Earnings Call file')
+ with st.expander(label='Query Result with Sentiment Tag', expanded=True):
+ 
+ sentiment_label = gen_sentiment(answer)
+ df = pd.DataFrame.from_dict({'Text':[answer],'Sentiment':[sentiment_label]})
+ text_annotations = gen_annotated_text(df)[0] 
+ annotated_text(text_annotations)
+ 
  
  else: