import streamlit as st import random from helpers import query_you_com, query_tavily, query_perplexity, query_brave from provider_info import search_providers from mongod_db import MongoDBHandler from bson.objectid import ObjectId # from swarms.utils.loguru_logger import logger import time import uuid # mongo = MongoDBHandler() # Set Streamlit to wide mode st.set_page_config(layout="wide", page_title="SearchArena") # Add information to sidebar # st.sidebar.title("About the App") # st.sidebar.write(""" # This app allows you to compare responses from different search engines. # Submit a question, and you'll receive answers from two randomly selected search engines. # You can then vote on which response you prefer. # """) # st.sidebar.write(""" # **[GitHub](https://github.com/leowalker89/SearchArena)**""") # Header section st.title("⚔️ Search Arena: Evaluating and Comparing Search Providers") # Subheader with introduction st.header("Welcome to Search Arena") st.write(""" Welcome to Search Arena, an open platform for evaluating and comparing search providers through crowdsourced preferences. Inspired by the groundbreaking work of LMSYS's Chatbot Arena in benchmarking large language models (LLMs), Search Arena aims to bring a similar approach to the world of search. Our platform allows you to input a query and receive results from two anonymous search providers. After reviewing the results, you can vote for the provider that delivered the most relevant and helpful information. The search providers' identities are kept hidden during the voting process to ensure unbiased evaluation. You can continue refining your query and interacting with the search results until you are satisfied with the outcome. Currently, Search Arena compares results from four leading search providers: Tavily, Brave Search, Perplexity, and You.com. By collecting votes from a wide user base, we aim to establish a robust leaderboard that reflects the real-world performance and user preferences of these search engines. Join us in our mission to advance search technology through open collaboration and data-driven insights. Your participation will contribute to a growing dataset that will be made available to the research community, fostering innovation and improvement in the field of information retrieval. """) # Define the function to process the question def ProcessQuestion(question): document_id = None # Randomly select two out of the four functions functions = [query_you_com, query_tavily, query_perplexity, query_brave] selected_functions = random.sample(functions, 2) # Get answers from the selected functions answer_a = selected_functions[0](question) answer_b = selected_functions[1](question) # Log into mongodb mongo = MongoDBHandler() try: # logger.info(f"Logging question: {question}") document_id = mongo.add( { "question": question, "answer_a": answer_a, "answer_b": answer_b, "selected_functions": [f.__name__ for f in selected_functions], "query_time": time.time(), "session_id": st.session_state.session_id, } ) # logger.info("Successfully logged into mongodb") except Exception as e: # logger.error(f"Error logging into mongodb: {e}") print(f"Error logging into mongodb: {e}") return answer_a, answer_b, selected_functions, document_id def UpdateVote(session_id, vote): mongo = MongoDBHandler() try: mongo.update( {"session_id": session_id}, {"$set": {"vote": vote}} ) except Exception as e: print(f"Error updating vote in mongodb: {e}") def UpdateFeedback(session_id, feedback): mongo = MongoDBHandler() try: mongo.update( {"session_id": session_id}, {"$set": {"feedback": feedback}} ) except Exception as e: print(f"Error updating feedback in mongodb: {e}") # Initialize session state if not already done default_values = { "state": "arena_ready", "question": "", "answer_a": "", "answer_b": "", "source_a": "", "source_b": "", "winner": "", "selected_button": "", "document_id": "", "feedback": "", "session_id": str(uuid.uuid4()) } for key, value in default_values.items(): if key not in st.session_state: st.session_state[key] = value # Streamlit app layout st.title("Search Engine Agent Comparison") def on_submit(): question = st.session_state["question_input"] if question: answer_a, answer_b, selected_functions, document_id = ProcessQuestion(question) st.session_state["question"] = question st.session_state["answer_a"] = answer_a st.session_state["answer_b"] = answer_b st.session_state["source_a"] = selected_functions[0].__name__ st.session_state["source_b"] = selected_functions[1].__name__ st.session_state["state"] = "arena_review" st.session_state["document_id"] = document_id def handle_vote(vote): st.session_state["winner"] = vote st.session_state["state"] = "arena_results" UpdateVote(st.session_state["session_id"], vote) def get_provider_info(provider_function_name): provider_name_map = { 'query_you_com': 'You.com', 'query_tavily': 'Tavily', 'query_perplexity': 'Perplexity AI', 'query_brave': 'Brave Search' } provider_name = provider_name_map.get(provider_function_name) return next((provider for provider in search_providers if provider['company_name'] == provider_name), {}) def render_ready_state(): st.text_area("Enter your question here (max 1000 characters):", max_chars=1000, key="question_input", on_change=on_submit) st.button("Submit", on_click=on_submit) def render_review_state(): st.write("## Results") col1, col2 = st.columns(2) with col1: st.write("### Answer A") st.write(st.session_state["answer_a"]) with col2: st.write("### Answer B") st.write(st.session_state["answer_b"]) st.write("### Vote for the Best Answer") col1, col2, col3, col4 = st.columns(4) if col1.button("It's a Tie 🤝"): handle_vote("Tie") if col2.button("A is better 💪"): handle_vote("A") if col3.button("B is better 🥇"): handle_vote("B") if col4.button("Both are bad 👎"): handle_vote("Both are bad") def render_results_state(): st.write("## Results") st.write(f"### Question: {st.session_state['question']}") provider_info_a = get_provider_info(st.session_state["source_a"]) provider_info_b = get_provider_info(st.session_state["source_b"]) col1, col2 = st.columns(2) with col1: if st.session_state["winner"] == "A": st.write(f"### ⭐ {provider_info_a['company_name']} 🥇") elif st.session_state["winner"] == "Tie": st.write(f"### 🤝 {provider_info_a['company_name']} 🤝") elif st.session_state["winner"] == "Both are bad": st.write(f"### 👎 {provider_info_a['company_name']} 👎") else: st.write(f"### {provider_info_a['company_name']} 🥈") st.write("**Response:**") st.markdown(f"
{st.session_state['answer_a']}
", unsafe_allow_html=True) with col2: if st.session_state["winner"] == "B": st.write(f"### ⭐ {provider_info_b['company_name']} 🥇") elif st.session_state["winner"] == "Tie": st.write(f"### 🤝 {provider_info_b['company_name']} 🤝") elif st.session_state["winner"] == "Both are bad": st.write(f"### 👎 {provider_info_b['company_name']} 👎") else: st.write(f"### {provider_info_b['company_name']} 🥈") st.write("**Response:**") st.markdown(f"
{st.session_state['answer_b']}
", unsafe_allow_html=True) st.write("### Feedback") feedback = st.text_area("Please provide feedback on why you chose the winner:") if feedback: UpdateFeedback(st.session_state["session_id"], feedback) st.write("### About the search providers:") col1, col2 = st.columns(2) with col1: st.write(f"**Website:** [{provider_info_a['website']}]({provider_info_a['website']})") st.write(f"**Overview:** {provider_info_a['overview']}") with col2: st.write(f"**Website:** [{provider_info_b['website']}]({provider_info_b['website']})") st.write(f"**Overview:** {provider_info_b['overview']}") if st.session_state["state"] == "arena_ready": render_ready_state() elif st.session_state["state"] == "arena_review": render_review_state() elif st.session_state["state"] == "arena_results": render_results_state() # Apply custom CSS to highlight the selected button selected_button = st.session_state.get("selected_button", "") if selected_button: st.markdown( f""" """, unsafe_allow_html=True )