Spaces:

shubham142000
/

Future_Directions

Sleeping

File size: 23,971 Bytes

# # import os
# # import json
# # import streamlit as st
# # import pandas as pd
# # import plotly.express as px
# # from together import Together
# # from dotenv import load_dotenv
# # import re

# # # -------------------#
# # # Secure API key load
# # # -------------------#
# # load_dotenv()
# # TOGETHER_API_KEY = os.getenv("TOGETHER_API_KEY", "987adcf573b9658c775b671270aef959b3d38793771932f372f9f2a9ed5b78bf")
# # client = Together(api_key=TOGETHER_API_KEY)

# # # -------------------#
# # # Streamlit UI setup
# # # -------------------#
# # st.set_page_config(page_title="FutureScope: Research Direction Explorer", layout="wide")

# # st.markdown("""
# # <style>
# # body {
# #     background: linear-gradient(135deg, #0f2027, #203a43, #2c5364);
# #     color: #FFFFFF;
# # }
# # h1, h2, h3 {
# #     text-align: center;
# #     color: #FFD700;
# #     font-family: 'Poppins', sans-serif;
# # }
# # .footer {
# #     position: fixed;
# #     left: 0;
# #     bottom: 0;
# #     width: 100%;
# #     color: white;
# #     text-align: center;
# #     padding: 10px;
# #     background-color: rgba(0,0,0,0.4);
# # }
# # .stButton > button {
# #     background-color: #FFD700 !important;
# #     color: black !important;
# #     font-weight: bold;
# #     border-radius: 10px;
# # }
# # </style>
# # """, unsafe_allow_html=True)

# # # -------------------#
# # # App Title
# # # -------------------#
# # st.markdown("<h1>🧭 FutureScope: Research Direction Explorer</h1>", unsafe_allow_html=True)
# # st.markdown("<p style='text-align:center;'>Discover how your research area evolved and where it's heading next 🚀</p>", unsafe_allow_html=True)

# # # -------------------#
# # # User Input
# # # -------------------#
# # user_topic = st.text_input("🔍 Enter your research topic", placeholder="e.g. Graph Neural Networks for Drug Discovery")

# # # -------------------#
# # # Main Logic
# # # -------------------#
# # if st.button("Generate Research Insights"):
# #     if not user_topic.strip():
# #         st.warning("⚠️ Please enter a valid research topic.")
# #     else:
# #         with st.spinner("Analyzing topic evolution and forecasting future directions... ⏳"):

# #             # Prompt Design
# #             prompt = f"""
# #             You are a world-class AI research assistant specialized in analyzing research trends.
# #             Given the topic: "{user_topic}", perform the following:
# #             1. Summarize how this research area has evolved in the past 10–15 years.
# #             2. Identify key milestones and subfields in a timeline format.
# #             3. Predict 3–5 future research directions and explain why each matters.
# #             Return the output strictly in JSON format like this:
# #             {{
# #               "evolution_summary": "...",
# #               "timeline": [{{"year": ..., "trend": "..."}}, ...],
# #               "future_directions": [{{"title": "...", "reason": "..."}}, ...]
# #             }}
# #             """

# #             # Call Together API
# #             response = client.chat.completions.create(
# #                 model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free",
# #                 messages=[{"role": "user", "content": prompt}]
# #             )

# #             raw_content = response.choices[0].message.content

# #             # -------------------#
# #             # JSON Cleaning & Parsing
# #             # -------------------#
# #             def extract_json(text):
# #                 """Extract valid JSON portion from the model response."""
# #                 text = text.strip()
# #                 text = re.sub(r"^```json|```$", "", text).strip()  # remove code fences
# #                 match = re.search(r'\{.*\}', text, re.DOTALL)
# #                 if match:
# #                     return match.group(0)
# #                 return text

# #             cleaned = extract_json(raw_content)
# #             try:
# #                 data = json.loads(cleaned)
# #             except Exception as e:
# #                 st.error(f"⚠️ Failed to parse JSON: {e}")
# #                 st.text_area("Raw Response", raw_content, height=300)
# #                 st.stop()

# #             # -------------------#
# #             # Display Results
# #             # -------------------#
# #             st.markdown("## 🧩 Evolution Summary")
# #             st.markdown(f"<div style='background:#1e2a38;padding:15px;border-radius:10px;'>{data['evolution_summary']}</div>", unsafe_allow_html=True)

# #             # Timeline Chart
# #             if "timeline" in data and len(data["timeline"]) > 0:
# #                 df = pd.DataFrame(data["timeline"])
# #                 if "year" in df.columns and "trend" in df.columns:
# #                     fig = px.scatter(df, x="year", y="trend", title="📈 Topic Evolution Over Time",
# #                                      size=[10]*len(df), text="trend", color_discrete_sequence=["gold"])
# #                     fig.update_traces(textposition='top center', marker=dict(symbol="circle"))
# #                     fig.update_layout(template="plotly_dark", height=500)
# #                     st.plotly_chart(fig, use_container_width=True)
# #                 else:
# #                     st.warning("Timeline data invalid — showing raw table:")
# #                     st.dataframe(df)

# #             # Future Directions
# #             st.markdown("## 🔮 Predicted Future Directions")
# #             for item in data.get("future_directions", []):
# #                 st.markdown(f"""
# #                 <div style='background:#142733;padding:15px;margin:10px;border-radius:10px;'>
# #                 <h4>🧠 {item['title']}</h4>
# #                 <p>{item['reason']}</p>
# #                 </div>
# #                 """, unsafe_allow_html=True)

# #             # Tools: Copy / Download
# #             col1, col2 = st.columns(2)
# #             with col1:
# #                 if st.button("📋 Copy Insights"):
# #                     st.write("Copied to clipboard! (Use Ctrl+C manually to copy)")
# #             with col2:
# #                 st.download_button(
# #                     label="💾 Download JSON",
# #                     data=json.dumps(data, indent=2),
# #                     file_name=f"{user_topic.replace(' ','_')}_future_directions.json",
# #                     mime="application/json"
# #                 )

# # # -------------------#
# # # Footer
# # # -------------------#
# # st.markdown("<div class='footer'>© Group 6 ILP TCS Research ", unsafe_allow_html=True)

import os
import json
import streamlit as st
import pandas as pd
import plotly.express as px
from together import Together
from dotenv import load_dotenv
import re

# -------------------#
# Secure API key load
# -------------------#
load_dotenv()
TOGETHER_API_KEY = os.getenv("TOGETHER_API_KEY", "987adcf573b9658c775b671270aef959b3d38793771932f372f9f2a9ed5b78bf")
client = Together(api_key=TOGETHER_API_KEY)

# -------------------#
# Streamlit UI setup
# -------------------#
st.set_page_config(page_title="PreSearch : Research Direction Explorer", layout="wide")

# Detect Streamlit theme
try:
    theme = st.get_option("theme.base")
except:
    theme = "dark"

# Define theme colors
if theme == "light":
    BACKGROUND_GRADIENT = "linear-gradient(135deg, #f9f9f9, #eaeaea, #dddddd)"
    TEXT_COLOR = "#000000"
    TITLE_COLOR = "#DAA520"
    CARD_BG = "#ffffff"
    FOOTER_BG = "rgba(0, 0, 0, 0.1)"
else:
    BACKGROUND_GRADIENT = "linear-gradient(135deg, #0f2027, #203a43, #2c5364)"
    TEXT_COLOR = "#FFFFFF"
    TITLE_COLOR = "#FFD700"
    CARD_BG = "#1e2a38"
    FOOTER_BG = "rgba(0, 0, 0, 0.4)"

# -------------------#
# Dynamic CSS Styling
# -------------------#
st.markdown(f"""
<style>
body {{
    background: {BACKGROUND_GRADIENT};
    color: {TEXT_COLOR};
    font-family: 'Poppins', sans-serif;
}}
h1, h2, h3 {{
    text-align: center;
    color: {TITLE_COLOR};
}}
.footer {{
    position: fixed;
    left: 0;
    bottom: 0;
    width: 100%;
    color: {TEXT_COLOR};
    text-align: center;
    padding: 10px;
    background-color: {FOOTER_BG};
}}
.stButton > button {{
    background-color: {TITLE_COLOR} !important;
    color: black !important;
    font-weight: bold;
    border-radius: 10px;
}}
div[data-testid="stMarkdownContainer"] p {{
    color: {TEXT_COLOR};
}}
</style>
""", unsafe_allow_html=True)

# -------------------#
# App Title
# -------------------#
st.markdown("<h1>🧭 PreSearch: Research Direction Explorer</h1>", unsafe_allow_html=True)
st.markdown("<p style='text-align:center;'>Discover how your research area evolved and where it's heading next 🚀</p>", unsafe_allow_html=True)

# -------------------#
# User Input
# -------------------#
user_topic = st.text_input("🔍 Enter your research topic", placeholder="e.g. Graph Neural Networks for Drug Discovery")

# -------------------#
# Main Logic
# -------------------#
if st.button("Generate Research Insights"):
    if not user_topic.strip():
        st.warning("⚠️ Please enter a valid research topic.")
    else:
        with st.spinner("Analyzing topic evolution and forecasting future directions... ⏳"):

            # Prompt Design
            prompt = f"""
            You are a world-class AI research assistant specialized in analyzing research trends.
            Given the topic: "{user_topic}", perform the following:
            1. Summarize how this research area has evolved in the past 10–15 years.
            2. Identify key milestones and subfields in a timeline format.
            3. Predict 3–5 future research directions and explain why each matters.
            Return the output strictly in JSON format like this:
            {{
              "evolution_summary": "...",
              "timeline": [{{"year": ..., "trend": "..."}}, ...],
              "future_directions": [{{"title": "...", "reason": "..."}}, ...]
            }}
            """

            # Call Together API
            response = client.chat.completions.create(
                model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free",
                messages=[{"role": "user", "content": prompt}]
            )

            raw_content = response.choices[0].message.content

            # -------------------#
            # JSON Cleaning & Parsing
            # -------------------#
            def extract_json(text):
                text = text.strip()
                text = re.sub(r"^```json|```$", "", text).strip()
                match = re.search(r'\{.*\}', text, re.DOTALL)
                return match.group(0) if match else text

            cleaned = extract_json(raw_content)
            try:
                data = json.loads(cleaned)
            except Exception as e:
                st.error(f"⚠️ Failed to parse JSON: {e}")
                st.text_area("Raw Response", raw_content, height=300)
                st.stop()

            # -------------------#
            # Display Results
            # -------------------#
            st.markdown("## 🧩 Evolution Summary")
            st.markdown(f"<div style='background:{CARD_BG};padding:15px;border-radius:10px;color:{TEXT_COLOR};'>{data['evolution_summary']}</div>", unsafe_allow_html=True)

            # Timeline Chart
            if "timeline" in data and len(data["timeline"]) > 0:
                df = pd.DataFrame(data["timeline"])
                if "year" in df.columns and "trend" in df.columns:
                    fig = px.scatter(df, x="year", y="trend", title="📈 Topic Evolution Over Time",
                                     size=[10]*len(df), text="trend", color_discrete_sequence=["gold"])
                    fig.update_traces(textposition='top center', marker=dict(symbol="circle"))
                    fig.update_layout(template="plotly_dark" if theme == "dark" else "plotly_white", height=500)
                    st.plotly_chart(fig, use_container_width=True)
                else:
                    st.warning("Timeline data invalid — showing raw table:")
                    st.dataframe(df)

            # Future Directions
            st.markdown("## 🔮 Predicted Future Directions")
            for item in data.get("future_directions", []):
                st.markdown(f"""
                <div style='background:{CARD_BG};padding:15px;margin:10px;border-radius:10px;color:{TEXT_COLOR};'>
                <h4>🧠 {item['title']}</h4>
                <p>{item['reason']}</p>
                </div>
                """, unsafe_allow_html=True)

            # Tools: Copy / Download
            col1, col2 = st.columns(2)
            with col1:
                if st.button("📋 Copy Insights"):
                    st.write("Copied to clipboard! (Use Ctrl+C manually to copy)")
            with col2:
                st.download_button(
                    label="💾 Download JSON",
                    data=json.dumps(data, indent=2),
                    file_name=f"{user_topic.replace(' ','_')}_future_directions.json",
                    mime="application/json"
                )

# -------------------#
# Footer
# -------------------#
st.markdown(f"<div class='footer'>© Group 6 ILP TCS Research</div>", unsafe_allow_html=True)



# import os, json, time, re, requests, random
# import pandas as pd
# import streamlit as st
# from together import Together

# # =========================
# # 0️⃣ Configuration & Setup
# # =========================
# st.set_page_config(page_title="📚 pResearch Retrieval", layout="wide", page_icon=":books:")
# st.title("🤖 **pResearch: Multi-Agent Research Retrieval System**")
# st.caption("Built with LLM-based reasoning, multi-agent intelligence, and human-in-loop control.")
# st.markdown("---")

# TOGETHER_API_KEY = os.getenv("TOGETHER_API_KEY", "987adcf573b9658c775b671270aef959b3d38793771932f372f9f2a9ed5b78bf")
# SEMANTIC_API_KEY = os.getenv("SEMANTIC_API_KEY", "b2EsaPVVN1890PXdCeum37K9zKq4AYY46n8QyLvp")
# client = Together(api_key=TOGETHER_API_KEY)

# # =========================
# # Unified LLM Call
# # =========================
# @st.cache_data(show_spinner=False)
# def llm_call(prompt: str, temperature=0.2, max_retries=3):
#     for attempt in range(max_retries):
#         try:
#             resp = client.chat.completions.create(
#                 model="meta-llama/Llama-3.3-70B-Instruct-Turbo",
#                 messages=[{"role": "user", "content": prompt}],
#                 temperature=temperature
#             )
#             return resp.choices[0].message.content.strip()
#         except Exception as e:
#             time.sleep(1 + attempt)
#     return "LLM error (see logs)"


# # ============================================================
# # 1️⃣ Query Reformulator Agent
# # ============================================================
# def agent_query_reformulator(query: str):
#     prompt = f"""
#     You are an expert academic assistant.
#     Reformulate the query below into 5 semantically diverse and rich alternatives
#     that explore different perspectives (methods, datasets, applications, etc.)

#     Query: "{query}"

#     Respond in JSON format:
#     {{
#       "reformulated_queries": [
#         {{ "id": 1, "query": "..." }},
#         {{ "id": 2, "query": "..." }},
#         {{ "id": 3, "query": "..." }},
#         {{ "id": 4, "query": "..." }},
#         {{ "id": 5, "query": "..." }}
#       ]
#     }}
#     """
#     output = llm_call(prompt)
#     cleaned = re.sub(r"```json|```", "", output).strip()

#     try:
#         data = json.loads(cleaned)
#         queries = [q["query"] for q in data.get("reformulated_queries", []) if "query" in q]
#     except Exception:
#         queries = []

#     # fallback diversity
#     while len(queries) < 5:
#         alt = llm_call(f"Generate a diverse reformulation of: {query}")
#         queries.append(alt[:300])
#     queries = list(dict.fromkeys(queries))[:5]
#     return {"original_query": query, "reformulated_queries": [{"id": i+1, "query": q} for i, q in enumerate(queries)]}


# # ============================================================
# # 2️⃣ Retriever Agent (Semantic Scholar)
# # ============================================================
# def agent_retriever(query, top_k=20):
#     url = "https://api.semanticscholar.org/graph/v1/paper/search"
#     headers = {"x-api-key": SEMANTIC_API_KEY}
#     params = {
#         "query": query, "limit": top_k,
#         "fields": "paperId,title,abstract,year,authors,url,venue,citationCount"
#     }
#     resp = requests.get(url, headers=headers, params=params)
#     if resp.status_code != 200:
#         return []
#     return resp.json().get("data", [])


# # ============================================================
# # 3️⃣ Reranker Agent
# # ============================================================
# def agent_reranker(query, papers):
#     cleaned_papers = [p for p in papers if p.get("abstract")]
#     random.shuffle(cleaned_papers)
#     for batch_start in range(0, len(cleaned_papers), 5):
#         batch = cleaned_papers[batch_start:batch_start+5]
#         papers_str = "\n\n".join([
#             f"[{i+1}] Title: {p.get('title','N/A')}\nAbstract: {p.get('abstract','')[:500]}"
#             for i,p in enumerate(batch)
#         ])
#         prompt = f"""
#         You are a relevance scoring agent. 
#         Given a research query and 5 papers, assign a score (0–1) to each paper for its relevance.

#         Query: {query}
#         Papers:
#         {papers_str}

#         Respond strictly in JSON:
#         {{ "results": [{{"id": 1, "score": 0.85}}, ...] }}
#         """
#         output = llm_call(prompt)
#         cleaned = re.sub(r"```json|```", "", output).strip()
#         try:
#             results = json.loads(cleaned)["results"]
#             for i,r in enumerate(results):
#                 cleaned_papers[batch_start+i]["semantic_score"] = r.get("score",0)
#         except:
#             for i in range(len(batch)):
#                 cleaned_papers[batch_start+i]["semantic_score"] = 0.0
#     return sorted(cleaned_papers, key=lambda x: x.get("semantic_score",0), reverse=True)


# # ============================================================
# # 4️⃣ Weighting Agent (Meta Scorer)
# # ============================================================
# def agent_weighting(papers):
#     prompt = """
#     You are an expert in bibliometrics.
#     Assign importance weights (sum=1.0) for how papers should be ranked based on:
#     - semantic_score (LLM relevance)
#     - citationCount
#     - recency
#     - venue quality

#     Return JSON:
#     {"weights":{"semantic_score":0.55,"citations":0.25,"recency":0.15,"venue":0.05}}
#     """
#     output = llm_call(prompt)
#     cleaned = re.sub(r"```json|```", "", output).strip()
#     try:
#         weights = json.loads(cleaned)["weights"]
#     except:
#         weights = {"semantic_score":0.55,"citations":0.25,"recency":0.15,"venue":0.05}
#     total = sum(weights.values())
#     return {k:v/total for k,v in weights.items()}


# # ============================================================
# # 5️⃣ Meta-Scoring and Ranking
# # ============================================================
# def agent_meta_scorer(papers, weights):
#     current_year = 2025
#     prestige = {"CVPR":1.0,"ICCV":0.95,"ECCV":0.9,"NEURIPS":0.9,"ICML":0.85,"AAAI":0.8,"IJCAI":0.8,"ARXIV":0.4}
#     for p in papers:
#         sem = p.get("semantic_score",0)
#         cit = min(p.get("citationCount",0)/1000,1.0)
#         rec = max(0, 1 - (current_year - p.get("year",2000))/10)
#         venue_name = (p.get("venue") or "").upper()
#         ven = next((v for k,v in prestige.items() if k in venue_name), 0.3)
#         p["final_score"] = (
#             weights["semantic_score"]*sem +
#             weights["citations"]*cit +
#             weights["recency"]*rec +
#             weights["venue"]*ven
#         )
#     return sorted(papers, key=lambda x: x["final_score"], reverse=True)


# # ============================================================
# # 6️⃣ Critique Agent
# # ============================================================
# def agent_critique(papers, query):
#     top_titles = [p["title"] for p in papers[:5]]
#     prompt = f"""
#     As a research critic, evaluate whether these top papers are relevant to:
#     "{query}"

#     Papers: {json.dumps(top_titles, indent=2)}

#     Respond as JSON:
#     {{ "critique": "...", "relevance_score": 0–1 }}
#     """
#     output = llm_call(prompt)
#     cleaned = re.sub(r"```json|```", "", output).strip()
#     try:
#         return json.loads(cleaned)
#     except:
#         return {"critique":"Automatic check fallback.","relevance_score":0.7}


# # ============================================================
# # 7️⃣ Human-in-Loop Fallback
# # ============================================================
# def human_feedback_loop(papers):
#     st.warning("⚠️ Low relevance detected — human feedback required.")
#     for i,p in enumerate(papers[:3]):
#         st.markdown(f"**{i+1}. {p['title']}** *(Score: {p['final_score']:.3f})*")
#         st.caption(f"{p.get('abstract','')[:250]}...")
#     choice = st.radio("Approve ranking?", ["Yes","No"], index=0)
#     if choice == "No":
#         st.info("🔄 Re-ranking by citation count.")
#         papers = sorted(papers, key=lambda x: x.get("citationCount",0), reverse=True)
#     return papers


# # ============================================================
# # 8️⃣ Streamlit Master Orchestrator
# # ============================================================
# def run_pipeline(query, top_k=10):
#     st.markdown("## 🧩 Stage 1: Query Reformulation")
#     with st.spinner("Generating diverse reformulations..."):
#         q_data = agent_query_reformulator(query)
#     queries = [query] + [q["query"] for q in q_data["reformulated_queries"]]
#     for q in queries[1:]:
#         st.markdown(f"🔹 *{q}*")

#     st.markdown("## 🔍 Stage 2: Retrieval")
#     all_papers = []
#     progress = st.progress(0)
#     for i, q in enumerate(queries):
#         new_papers = agent_retriever(q, top_k)
#         all_papers.extend(new_papers)
#         progress.progress((i+1)/len(queries))
#         time.sleep(0.3)
#     progress.empty()
#     st.success(f"✅ Retrieved {len(all_papers)} papers in total.")

#     st.markdown("## 🧠 Stage 3: Semantic Reranking")
#     with st.spinner("Reranking papers semantically..."):
#         reranked = agent_reranker(query, all_papers)
#     st.info(f"Top paper after rerank: **{reranked[0]['title']}**")

#     st.markdown("## ⚖️ Stage 4: Weighting Agent & Meta-Scoring")
#     weights = agent_weighting(reranked)
#     st.json(weights)
#     meta_ranked = agent_meta_scorer(reranked, weights)

#     st.markdown("## 🔎 Stage 5: Critique Agent")
#     critique = agent_critique(meta_ranked, query)
#     st.info(f"**Critique:** {critique['critique']} | Relevance: {critique['relevance_score']:.2f}")

#     if critique["relevance_score"] < 0.6:
#         meta_ranked = human_feedback_loop(meta_ranked)

#     df = pd.DataFrame(meta_ranked)
#     st.download_button(
#         label="💾 Download Results as CSV",
#         data=df.to_csv(index=False),
#         file_name="final_ranked_papers.csv",
#         mime="text/csv"
#     )

#     st.dataframe(df.head(20))
#     st.success("🎯 Pipeline completed successfully!")


# # ============================================================
# # 9️⃣ Streamlit UI
# # ============================================================
# with st.form("research_form"):
#     query = st.text_input("Enter your research query:", "spatio-temporal action detection and localization")
#     top_k = st.slider("Number of papers per reformulation", 5, 50, 10)
#     run = st.form_submit_button("🚀 Run Multi-Agent Retrieval")

# if run:
#     run_pipeline(query, top_k)