app.py

import os
import requests
import pandas as pd
import streamlit as st
import plotly.express as px
import numpy as np
from datetime import datetime
from transformers import pipeline
from dotenv import load_dotenv

# ========================
# CONFIG
# ========================
load_dotenv()
SERPAPI_KEY = os.getenv("SERPAPI_KEY")

st.set_page_config(layout="wide")
st.title("🔥 OSINT BRI Monitoring System (PRO)")

# ========================
# MODELS
# ========================
@st.cache_resource
def load_models():
    sentiment = pipeline(
        "sentiment-analysis",
        model="cardiffnlp/twitter-roberta-base-sentiment"
    )

    risk = pipeline(
        "zero-shot-classification",
        model="facebook/bart-large-mnli"
    )

    return sentiment, risk

sentiment_model, risk_model = load_models()

# ========================
# DATA COLLECTOR
# ========================
def get_news(query):
    url = "https://serpapi.com/search.json"

    params = {
        "q": query,
        "hl": "ru",
        "gl": "kz",
        "api_key": SERPAPI_KEY
    }

    try:
        res = requests.get(url, params=params)
        data = res.json()

        rows = []
        for item in data.get("organic_results", []):
            rows.append({
                "text": item.get("title", "") + " " + item.get("snippet", ""),
                "time": datetime.now()
            })

        return rows

    except:
        return []

def get_social():
    # можно заменить на API
    return [
        {"text": "AI риски растут", "time": datetime.now()},
        {"text": "Проблемы безопасности LLM", "time": datetime.now()},
        {"text": "Рост технологий в Казахстане", "time": datetime.now()}
    ]

# ========================
# RISK LABELS
# ========================
RISK_LABELS = [
    "bias",
    "hallucination",
    "security",
    "misuse",
    "regulation"
]

# ========================
# NLP ENGINE
# ========================
def analyze(data):
    results = []

    for row in data:
        t = row["text"]

        try:
            sent = sentiment_model(t)[0]
            risk = risk_model(t, RISK_LABELS)

            results.append({
                "text": t,
                "time": row["time"],
                "sentiment": sent["label"],
                "sent_score": sent["score"],
                "risk": risk["labels"][0],
                "risk_score": risk["scores"][0]
            })
        except:
            continue

    return pd.DataFrame(results)

# ========================
# ADVANCED BRI
# ========================
def compute_bri(df):

    if df.empty:
        return 100

    neg = (df["sentiment"] == "NEGATIVE").mean()
    risk = df["risk_score"].mean()

    # аномалия (всплеск)
    volume = len(df)
    anomaly = np.std(df["risk_score"]) if len(df) > 1 else 0

    # веса
    w = {
        "risk": 0.4,
        "sent": 0.3,
        "volume": 0.2,
        "anomaly": 0.1
    }

    score = 100 - (
        w["risk"] * risk * 100 +
        w["sent"] * neg * 100 +
        w["volume"] * min(volume, 50) +
        w["anomaly"] * anomaly * 100
    )

    return max(0, min(100, score))

# ========================
# UI
# ========================
query = st.text_input("🔍 Бренд / тема", "LLM Kazakhstan")

if st.button("🚀 Запустить мониторинг"):

    with st.spinner("Сбор и анализ данных..."):

        data = get_news(query) + get_social()
        df = analyze(data)

        bri = compute_bri(df)

    # ========================
    # KPI
    # ========================
    col1, col2, col3 = st.columns(3)

    col1.metric("📊 BRI Index", round(bri, 2))
    col2.metric("⚠️ Avg Risk", round(df["risk_score"].mean(), 2))
    col3.metric("💬 Mentions", len(df))

    # ========================
    # TIME SERIES
    # ========================
    st.subheader("📈 Динамика")

    df["time"] = pd.to_datetime(df["time"])

    ts = df.groupby(df["time"].dt.hour).size().reset_index(name="count")

    st.plotly_chart(px.line(ts, x="time", y="count"))

    # ========================
    # SENTIMENT
    # ========================
    st.subheader("🥧 Sentiment")
    st.plotly_chart(px.pie(df, names="sentiment"))

    # ========================
    # RISK
    # ========================
    st.subheader("⚠️ Risk Distribution")
    st.plotly_chart(px.bar(df["risk"].value_counts()))

    # ========================
    # HEATMAP
    # ========================
    st.subheader("🔥 Risk vs Sentiment")

    pivot = pd.crosstab(df["risk"], df["sentiment"])

    st.plotly_chart(px.imshow(pivot, text_auto=True))

    # ========================
    # ALERT SYSTEM
    # ========================
    if bri < 60:
        st.error("🚨 Reputation Risk Detected!")

    elif bri < 80:
        st.warning("⚠️ Moderate Risk")

    else:
        st.success("✅ Stable Reputation")

    st.dataframe(df)