Upload 12 files

app.py

@@ -0,0 +1,228 @@
+import streamlit as st
+import os
+import warnings
+from warnings import filterwarnings
+
+# ---------- Services ----------
+from services.zero_shot import classify_intent
+from services.sentiment import detect_emotion
+from services.similarity import text_quality_score
+from services.cta_analysis import cta_strength
+from services.copy_optimizer import optimize_copy
+from services.meta_ads_api import fetch_live_ads
+
+# ---------- Utils ----------
+from utils.scoring import final_score
+from utils.trend_analysis import market_trends
+
+
+# ---------- Page Config ----------
+st.set_page_config(
+    page_title="Meta AI Ads Intelligence Tool",
+    page_icon="📢",
+    layout="wide"
+)
+
+st.title("📢 Meta AI Ads Intelligence Tool")
+st.caption("Live Meta Ads • Market Trends • AI Creative Analysis")
+
+
+# ---------- Sidebar ----------
+menu = st.sidebar.radio(
+    "Navigation",
+    [
+        "📊 Overview",
+        "🎯 Analyze My Ad",
+        "🧠 Live Competitor Ads",
+        "⚠️ Ad Fatigue Checker",
+        "✍️ Copy Optimizer",
+        "ℹ️ About"
+    ]
+)
+
+
+# =========================================================
+# 📊 OVERVIEW
+# =========================================================
+if menu == "📊 Overview":
+    st.subheader("What does this tool do?")
+
+    st.write(
+        """
+        This platform combines **Meta Ads Library live data** with **pretrained AI models**
+        to analyze ad creatives, market trends, and competitor messaging — without using
+        any historical performance data.
+        """
+    )
+
+    col1, col2, col3 = st.columns(3)
+    col1.metric("Live Meta Ads", "Yes")
+    col2.metric("Model Training", "Not Required")
+    col3.metric("Analysis Type", "Real-Time")
+
+    st.info("🔒 No ads are stored. All analysis runs on demand.")
+
+
+# =========================================================
+# 🎯 ANALYZE USER AD
+# =========================================================
+elif menu == "🎯 Analyze My Ad":
+    st.subheader("Analyze Your Ad Creative")
+
+    col1, col2 = st.columns(2)
+
+    with col1:
+        caption = st.text_area("Ad Caption / Primary Text", height=150)
+        cta = st.selectbox(
+            "Call To Action",
+            ["Buy Now", "Shop Now", "Learn More", "DM Us", "Sign Up", "Check It Out"]
+        )
+        analyze = st.button("Analyze Ad")
+
+    with col2:
+        if analyze and caption.strip():
+            with st.spinner("Running AI analysis..."):
+                intent = classify_intent(caption)
+                emotion = detect_emotion(caption)
+                quality = text_quality_score(caption)
+                cta_score = cta_strength(cta)
+
+                score = final_score(
+                    intent["score"],
+                    emotion["score"],
+                    cta_score,
+                    quality
+                )
+
+            st.metric("Performance Score", f"{score}/100")
+
+            if score >= 75:
+                st.success("🟢 Low Risk – Ready to Run")
+            elif score >= 50:
+                st.warning("🟡 Medium Risk – Needs Optimization")
+            else:
+                st.error("🔴 High Risk – Likely Budget Waste")
+
+            st.progress(score / 100)
+
+            st.markdown("### 🔍 AI Insights")
+            st.write(f"**Intent:** {intent['label']}")
+            st.write(f"**Emotion:** {emotion['emotion']}")
+            st.write(f"**CTA Strength:** {round(cta_score * 100)}%")
+            st.write(f"**Text Quality:** {round(quality * 100)}%")
+
+
+# =========================================================
+# 🧠 LIVE COMPETITOR ADS (META ADS LIBRARY)
+# =========================================================
+elif menu == "🧠 Live Competitor Ads":
+    st.subheader("Live Competitor Ads (Meta Ads Library)")
+
+    keyword = st.text_input("Search Keyword / Brand / Product")
+    country = st.selectbox("Country", ["IN", "US", "UK", "AE"])
+
+    if st.button("Fetch Live Ads"):
+        try:
+            with st.spinner("Fetching live ads from Meta Ads Library..."):
+                ads = fetch_live_ads(keyword, country)
+
+            if not ads:
+                st.warning("No ads found for this keyword.")
+            else:
+                st.success(f"Fetched {len(ads)} live ads")
+
+                # ---------- Market Trends ----------
+                trends = market_trends(ads)
+
+                st.markdown("### 📊 Market Trend Analysis")
+                st.write("**Total Live Ads:**", trends["total_ads"])
+                st.write("**Trending Keywords:**", ", ".join(trends["top_keywords"]))
+
+                st.divider()
+
+                # ---------- Show Ads + AI Analysis ----------
+                for ad in ads[:5]:
+                    st.markdown(f"### 🏷️ {ad['page_name']}")
+                    st.write(ad["ad_creative_body"])
+
+                    intent = classify_intent(ad["ad_creative_body"])
+                    emotion = detect_emotion(ad["ad_creative_body"])
+
+                    st.caption(
+                        f"Intent: {intent['label']} | "
+                        f"Emotion: {emotion['emotion']}"
+                    )
+
+                    st.divider()
+
+        except Exception as e:
+            st.error(f"Error fetching ads: {e}")
+
+
+# =========================================================
+# ⚠️ AD FATIGUE CHECKER
+# =========================================================
+elif menu == "⚠️ Ad Fatigue Checker":
+    st.subheader("Ad Fatigue Risk Estimator")
+
+    caption = st.text_area("Ad Caption", height=120)
+    days = st.slider("Planned Run Duration (Days)", 1, 30, 7)
+    frequency = st.slider("Estimated Frequency", 1.0, 5.0, 2.0)
+
+    if st.button("Check Fatigue"):
+        fatigue_risk = min((days * frequency) / 30, 1.0)
+
+        st.metric("Fatigue Risk", f"{round(fatigue_risk * 100)}%")
+        st.progress(fatigue_risk)
+
+        if fatigue_risk > 0.7:
+            st.error("High Fatigue Risk – Refresh Creative")
+        elif fatigue_risk > 0.4:
+            st.warning("Medium Risk – Monitor Performance")
+        else:
+            st.success("Low Risk – Safe to Run")
+
+
+# =========================================================
+# ✍️ COPY OPTIMIZER
+# =========================================================
+elif menu == "✍️ Copy Optimizer":
+    st.subheader("AI Copy Optimization")
+
+    caption = st.text_area("Original Caption", height=150)
+
+    if st.button("Get Suggestions") and caption.strip():
+        tips = optimize_copy(caption)
+
+        if tips:
+            st.markdown("### ✨ Optimization Suggestions")
+            for tip in tips:
+                st.write("•", tip)
+        else:
+            st.success("Your caption already follows best practices!")
+
+
+# =========================================================
+# ℹ️ ABOUT
+# =========================================================
+elif menu == "ℹ️ About":
+    st.subheader("About This Project")
+
+    st.write(
+        """
+        **Meta AI Ads Intelligence Tool** is a real-time marketing intelligence platform.
+
+        ### Key Capabilities
+        - Live Meta Ads Library integration
+        - Market trend analysis
+        - Zero-shot intent classification
+        - Emotion detection
+        - No model training or historical data
+
+        ### Tech Stack
+        - Python
+        - Streamlit
+        - HuggingFace Transformers
+        - Meta Ads Library API
+        """
+    )

projectstructure.txt

@@ -0,0 +1,17 @@
+meta_ai_ads_tool/
+│
+├── app.py
+│
+├── services/
+│   ├── meta_ads_api.py       👈 META API KEY USED HERE
+│   ├── zero_shot.py
+│   ├── sentiment.py
+│   ├── similarity.py
+│   ├── cta_analysis.py
+│   └── copy_optimizer.py
+│
+├── utils/
+│   ├── scoring.py
+│   └── trend_analysis.py    👈 MARKET TRENDS
+│
+└── requirements.txt

requirements.txt

@@ -0,0 +1,8 @@
+streamlit==1.28.0
+transformers==4.40.0
+sentence-transformers==2.2.2
+torch>=2.0.0
+numpy>=1.24.0
+requests>=2.31.0
+scikit-learn>=1.3.0
+pandas>=2.1.0

services/copy_optimizer.py

@@ -0,0 +1,40 @@
+def optimize_copy(ad_text: str):
+    """
+    Provides actionable suggestions to improve ad captions.
+    Rule-based, no training required.
+    """
+
+    if not ad_text or len(ad_text.strip()) == 0:
+        return ["Ad text is empty, please provide text."]
+
+    suggestions = []
+
+    # Suggest shortening long captions
+    word_count = len(ad_text.split())
+    if word_count > 20:
+        suggestions.append("Consider shortening the caption (under 20 words).")
+
+    # Add urgency if missing
+    urgency_words = ["now", "today", "limited", "hurry", "exclusive", "offer"]
+    if not any(word in ad_text.lower() for word in urgency_words):
+        suggestions.append("Add urgency words like 'now', 'limited', 'exclusive'.")
+
+    # Check for punctuation / excitement
+    if "!" not in ad_text:
+        suggestions.append("Add punctuation or exclamation marks for excitement.")
+
+    # Suggest including a CTA if missing
+    cta_words = ["buy", "shop", "order", "download", "register", "sign up"]
+    if not any(word in ad_text.lower() for word in cta_words):
+        suggestions.append("Include a clear call-to-action (CTA) in your caption.")
+
+    # Suggest adding emotional trigger words
+    emotional_words = ["amazing", "best", "incredible", "free", "surprise"]
+    if not any(word in ad_text.lower() for word in emotional_words):
+        suggestions.append("Consider adding emotional trigger words to attract attention.")
+
+    # If no suggestions, compliment the copy
+    if len(suggestions) == 0:
+        suggestions.append("Your caption follows best practices!")
+
+    return suggestions

services/cta_analysis.py

@@ -0,0 +1,54 @@
+import re
+
+# High-performing CTA keywords (marketing proven)
+STRONG_CTA = [
+    "buy now", "shop now", "order now", "get started",
+    "sign up", "register", "download now", "limited offer",
+    "claim now", "book now", "subscribe", "grab now"
+]
+
+MEDIUM_CTA = [
+    "learn more", "discover", "find out", "see more",
+    "explore", "know more", "view details"
+]
+
+WEAK_CTA = [
+    "click here", "visit us", "check this",
+    "read more", "watch now"
+]
+
+def analyze_cta(ad_text: str):
+    """
+    Analyze CTA strength inside ad copy
+    Returns CTA score and insights
+    """
+
+    if not ad_text or len(ad_text.strip()) == 0:
+        return {"error": "Empty ad text"}
+
+    text = ad_text.lower()
+
+    found_strong = [cta for cta in STRONG_CTA if cta in text]
+    found_medium = [cta for cta in MEDIUM_CTA if cta in text]
+    found_weak = [cta for cta in WEAK_CTA if cta in text]
+
+    score = 0
+
+    # Scoring logic
+    score += len(found_strong) * 10
+    score += len(found_medium) * 5
+    score += len(found_weak) * 2
+
+    # Penalty if no CTA
+    if not (found_strong or found_medium or found_weak):
+        score -= 10
+
+    score = max(min(score, 100), 0)
+
+    return {
+        "cta_score": score,
+        "found_strong_cta": found_strong,
+        "found_medium_cta": found_medium,
+        "found_weak_cta": found_weak,
+        "has_cta": score > 0
+    }

services/meta_ads_api.py

@@ -0,0 +1,29 @@
+import requests
+import os
+
+META_ADS_TOKEN = os.getenv("EAAVtBlZBaes0BQZAfPSf7ZBO7Yv0WjULDVn3zZAwvgbdvCtU24x9MNG8rxGArSXDg4GFZBB3GBJOs5qblEU6BKCNx9GypIZAcneDRnuZBfkLYEWDZAMnzqblUUVNtPSjul792ZBuorTN36XHgqZCsOlFox4rsyYnw9Kjl2u244lfWsGZC4oCNIEwjslyljVryVXOY5W1LRiZAbnZCrMQ6GLdkkmKqJTkuLKKjrQxnPcnl7xCopI44nF1Moa3wtP1e61TQAoXc1UvdkEvllaJg7ircBzwYpILl")
+BASE_URL = "https://www.facebook.com/ads/library/?active_status=all&ad_type=political_and_issue_ads&country=IN&is_targeted_country=false&media_type=all"
+
+def fetch_live_ads(keyword, country="IN", limit=20):
+    if not META_ADS_TOKEN:
+        raise ValueError("META_ADS_TOKEN not set")
+
+    params = {
+        "search_terms": keyword,
+        "ad_reached_countries": country,
+        "ad_type": "ALL",
+        "fields": (
+            "page_name,"
+            "ad_creative_body,"
+            "ad_creative_link_title,"
+            "ad_delivery_start_time"
+        ),
+        "limit": limit,
+        "access_token": META_ADS_TOKEN
+    }
+
+    response = requests.get(BASE_URL, params=params)
+    response.raise_for_status()
+
+    ads = response.json().get("data", [])
+    return [ad for ad in ads if ad.get("ad_creative_body")]

services/sentiment.py

@@ -0,0 +1,35 @@
+from transformers import pipeline
+
+# Pretrained emotion detection model
+# No training required
+emotion_classifier = pipeline(
+    task="text-classification",
+    model="j-hartmann/emotion-english-distilroberta-base",
+    return_all_scores=True
+)
+
+def analyze_emotion(ad_text: str):
+    """
+    Analyze emotional tone of an ad caption
+    Returns emotion scores
+    """
+
+    if not ad_text or len(ad_text.strip()) == 0:
+        return {"error": "Empty ad text"}
+
+    result = emotion_classifier(ad_text)[0]
+
+    emotions = []
+    for item in result:
+        emotions.append({
+            "emotion": item["label"],
+            "confidence": round(item["score"], 3)
+        })
+
+    # Sort by highest confidence
+    emotions = sorted(emotions, key=lambda x: x["confidence"], reverse=True)
+
+    return {
+        "ad_text": ad_text,
+        "emotions": emotions
+    }

services/similarity.py

@@ -0,0 +1,36 @@
+from sentence_transformers import SentenceTransformer, util
+
+# Pretrained embedding model (NO training needed)
+model = SentenceTransformer("all-MiniLM-L6-v2")
+
+def compute_similarity(base_ad: str, competitor_ads: list):
+    """
+    Compare one ad caption with multiple competitor ads
+    Returns similarity scores
+    """
+
+    if not base_ad or not competitor_ads:
+        return {"error": "Base ad or competitor ads missing"}
+
+    # Encode base ad
+    base_embedding = model.encode(base_ad, convert_to_tensor=True)
+
+    results = []
+
+    for ad in competitor_ads:
+        competitor_embedding = model.encode(ad, convert_to_tensor=True)
+
+        score = util.cos_sim(base_embedding, competitor_embedding).item()
+
+        results.append({
+            "competitor_ad": ad,
+            "similarity_score": round(score * 100, 2)  # percentage
+        })
+
+    # Sort highest similarity first
+    results = sorted(results, key=lambda x: x["similarity_score"], reverse=True)
+
+    return {
+        "base_ad": base_ad,
+        "comparisons": results
+    }

services/zero_shot.py

@@ -0,0 +1,49 @@
+from transformers import pipeline
+
+# Load pretrained Zero-Shot Classification model
+# This model is already trained — NO custom data required
+classifier = pipeline(
+    task="zero-shot-classification",
+    model="facebook/bart-large-mnli"
+)
+
+# Marketing / Ads-specific labels
+DEFAULT_LABELS = [
+    "Brand Awareness",
+    "Lead Generation",
+    "Sales Promotion",
+    "Product Launch",
+    "Discount Offer",
+    "Emotional Appeal",
+    "Urgency Driven",
+    "Trust Building",
+    "Social Proof",
+    "Call To Action Focused"
+]
+
+def analyze_ad_intent(ad_text: str, labels: list = DEFAULT_LABELS):
+    """
+    Analyze ad caption text using Zero-Shot Learning
+    Returns intent labels with confidence scores
+    """
+
+    if not ad_text or len(ad_text.strip()) == 0:
+        return {"error": "Empty ad text"}
+
+    result = classifier(
+        sequences=ad_text,
+        candidate_labels=labels,
+        multi_label=True
+    )
+
+    response = []
+    for label, score in zip(result["labels"], result["scores"]):
+        response.append({
+            "label": label,
+            "confidence": round(score, 3)
+        })
+
+    return {
+        "ad_text": ad_text,
+        "analysis": response
+    }

utils/scoring.py

@@ -0,0 +1,43 @@
+def final_score(
+    intent_score: float = 0,
+    emotion_score: float = 0,
+    cta_score: float = 0,
+    text_quality_score: float = 0,
+    similarity_score: float = None
+) -> int:
+    """
+    Compute final Ad Performance Score (0-100)
+    Scores are expected between 0 and 1 (or 0-100 if already scaled)
+    similarity_score is optional (0-100)
+    """
+
+    # Scale 0-1 inputs to 0-100
+    intent_score = intent_score * 100 if intent_score <= 1 else intent_score
+    emotion_score = emotion_score * 100 if emotion_score <= 1 else emotion_score
+    cta_score = cta_score * 100 if cta_score <= 1 else cta_score
+    text_quality_score = text_quality_score * 100 if text_quality_score <= 1 else text_quality_score
+
+    # Weighted scoring (adjustable)
+    weights = {
+        "intent": 0.3,
+        "emotion": 0.25,
+        "cta": 0.2,
+        "text_quality": 0.15,
+        "similarity": 0.1
+    }
+
+    total_score = (
+        intent_score * weights["intent"] +
+        emotion_score * weights["emotion"] +
+        cta_score * weights["cta"] +
+        text_quality_score * weights["text_quality"]
+    )
+
+    # Include similarity if available
+    if similarity_score is not None:
+        total_score += similarity_score * weights["similarity"]
+
+    # Clamp between 0-100
+    total_score = max(min(total_score, 100), 0)
+
+    return round(total_score)

utils/trend_analysis.py

@@ -0,0 +1,24 @@
+from collections import Counter
+import re
+
+def extract_keywords(texts):
+    words = []
+    for t in texts:
+        clean = re.sub(r"[^a-zA-Z ]", "", t.lower())
+        words.extend(clean.split())
+    return Counter(words)
+
+def market_trends(ads):
+    texts = [ad["ad_creative_body"] for ad in ads]
+
+    keyword_freq = extract_keywords(texts)
+
+    top_keywords = [
+        word for word, count in keyword_freq.items()
+        if count > 2 and len(word) > 3
+    ][:10]
+
+    return {
+        "total_ads": len(ads),
+        "top_keywords": top_keywords
+    }