Phani-ISB/Knowledge_Graphs

1# -*- coding: utf-8 -*-2"""Learn with Knowledge Graphs.ipynb3 4Automatically generated by Colab.5 6Original file is located at7    https://colab.research.google.com/drive/16UX6wbUmaLG6YBJKzH5YouNYYnw2mL8H8"""9 10# app.py11import streamlit as st12import wikipediaapi13import requests, json14import networkx as nx15import matplotlib.pyplot as plt16from neo4j import GraphDatabase17 18# ---------------------------19# CONFIGURATION20# ---------------------------21# API Key for Perplexity22PPLX_API_KEY = "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"23 24# Optional Neo4j credentials (leave empty if not using Neo4j)25NEO4J_URI = "neo4j+s://1a780c1e.databases.neo4j.io"26NEO4J_USER = "neo4j"27NEO4J_PASSWORD = "Xaabk9z1r5J-DPK6JPOH5QuOHL_MrTeFytx2c4sxjN4"28 29driver = None30if NEO4J_URI and NEO4J_USER and NEO4J_PASSWORD:31    driver = GraphDatabase.driver(NEO4J_URI, auth=(NEO4J_USER, NEO4J_PASSWORD))32 33 34# ---------------------------35# FUNCTIONS36# ---------------------------37def perplexity_chat(prompt, model="sonar-medium-online"):38    url = "https://api.perplexity.ai/chat/completions"39    headers = {40        "Authorization": f"Bearer {PPLX_API_KEY}",41        "Content-Type": "application/json",42    }43    data = {44        "model": model,45        "messages": [{"role": "user", "content": prompt}],46        "temperature": 0,47    }48    resp = requests.post(url, headers=headers, data=json.dumps(data))49    if resp.status_code != 200:50        return f"❌ Error {resp.status_code}: {resp.text}"51    return resp.json()["choices"][0]["message"]["content"]52 53 54def extract_triples_from_chunk(text, max_triples=5):55    prompt = f"""Extract up to {max_triples} subject-predicate-object triples56from the text below. Return only triples in the format (subject, predicate, object).57 58Text: {text}"""59 60    content = perplexity_chat(prompt)61    triples = []62    for line in content.splitlines():63        line = line.strip(" ()[]{}")64        if not line:65            continue66        parts = [p.strip() for p in line.split(",")]67        if len(parts) == 3:68            triples.append(tuple(parts))69    return triples70 71 72def build_kg_from_wiki_title(title, lang="en", chunk_chars=800, max_triples_per_chunk=5):73    wiki = wikipediaapi.Wikipedia(lang, user_agent = "MyKGApp/1.0 (contact: your_email@example.com)")74    page = wiki.page(title)75    if not page.exists():76        return []77 78    text = page.text79    chunks = [text[i:i+chunk_chars] for i in range(0, len(text), chunk_chars)]80 81    triples = []82    for chunk in chunks:83        chunk_triples = extract_triples_from_chunk(chunk, max_triples=max_triples_per_chunk)84        triples.extend(chunk_triples)85 86    return triples87 88 89def insert_triple(tx, subject, predicate, obj):90    tx.run(91        """92        MERGE (s:Entity {name: $subject})93        MERGE (o:Entity {name: $object})94        MERGE (s)-[:RELATION {type: $predicate}]->(o)95        """,96        subject=subject, predicate=predicate, object=obj97    )98 99def insert_triples(triples):100    if not driver:101        return102    with driver.session() as session:103        for s, p, o in triples:104            session.execute_write(insert_triple, s, p, o)105 106 107def answer_with_kg(question, triples, top_k=10, model="sonar-medium-online"):108    context_triples = triples[:top_k]109    context_str = "\n".join([f"({s}, {p}, {o})" for s, p, o in context_triples])110 111    prompt = f"""112    You are a QA assistant.113    Use the following knowledge graph triples as context to answer the question.114 115    Knowledge Graph Triples:116    {context_str}117 118    Question: {question}119 120    Answer in a clear, concise way. If you don't find enough info in triples,121    say 'Not found in knowledge graph'.122    """123    return perplexity_chat(prompt, model=model)124 125 126# ---------------------------127# STREAMLIT APP128# ---------------------------129st.title("📚 Knowledge Graph Chatbot (Wikipedia + Perplexity)")130 131# Input for Wikipedia Title132title = st.text_input("Enter a Wikipedia Title (e.g., Harry Potter):")133 134if title:135    st.write(f"🔍 Building Knowledge Graph for: **{title}** ...")136    triples = build_kg_from_wiki_title(title)137 138    if triples:139        st.success(f"Extracted {len(triples)} triples ✅")140 141        # Save in Neo4j if configured142        if driver:143            insert_triples(triples)144            st.info("📡 Triples also stored in Neo4j.")145 146        # Show sample triples147        st.subheader("Sample Triples")148        st.json(triples[:10])149 150        # Visualization inside Streamlit151        st.subheader("Graph Visualization")152        G = nx.DiGraph()153        for s, p, o in triples[:30]:154            G.add_edge(s, o, label=p)155 156        plt.figure(figsize=(12, 8))157        pos = nx.spring_layout(G, k=0.5)158        nx.draw(G, pos, with_labels=True, node_size=2500, node_color="lightblue",159                font_size=10, font_weight="bold", arrows=True)160        edge_labels = nx.get_edge_attributes(G, 'label')161        nx.draw_networkx_edge_labels(G, pos, edge_labels=edge_labels, font_size=8)162        st.pyplot(plt)163 164        # Chat interface165        st.subheader("💬 Ask Questions")166        user_question = st.text_input("Your question:")167        if user_question:168            answer = answer_with_kg(user_question, triples)169            st.write("🤖", answer)170    else:171        st.error("Page not found or no triples extracted.")172 173