Skip to main content

​
Scenario

Track how media sentiment around your brand, product category, or key topics shifts week over week. This job searches for mentions across all sources, batches them into weekly windows, sends each batch to Mavera Chat for structured sentiment analysis, and outputs a time-series sentiment dashboard with trend arrows. Flow: NewsAPI GET /everything?q={topic} (weekly batches) → Mavera POST /mave/chat per batch → Sentiment time series

​
Code

import os, requests, time
from datetime import datetime, timedelta

NA_KEY = os.environ["NEWSAPI_KEY"]
NA_BASE = "https://newsapi.org/v2"
NA_H = {"X-Api-Key": NA_KEY}
MV = os.environ["MAVERA_API_KEY"]
MV_BASE = "https://app.mavera.io/api/v1"
MV_H = {"Authorization": f"Bearer {MV}", "Content-Type": "application/json"}

TOPIC = "artificial intelligence marketing"
WEEKS = 4

# 1. Fetch articles in weekly batches
weekly_data = []
for week in range(WEEKS):
    end = datetime.now() - timedelta(weeks=week)
    start = end - timedelta(days=7)
    r = requests.get(f"{NA_BASE}/everything", headers=NA_H, params={
        "q": TOPIC, "from": start.strftime("%Y-%m-%d"),
        "to": end.strftime("%Y-%m-%d"), "sortBy": "relevancy",
        "language": "en", "pageSize": 25,
    })
    if not r.ok:
        print(f"Week {week}: API error {r.status_code}")
        continue
    articles = r.json().get("articles", [])
    weekly_data.append({
        "week_label": f"{start.strftime('%b %d')} – {end.strftime('%b %d')}",
        "articles": articles, "count": len(articles),
    })
    print(f"Week {week}: {len(articles)} articles ({start.strftime('%b %d')} – {end.strftime('%b %d')})")
    time.sleep(1)

# 2. Batch sentiment analysis via Mave
sentiment_series = []
for week in weekly_data:
    corpus = "\n".join(
        f"- [{a.get('source',{}).get('name','')}] {a['title']}: {a.get('description','')[:150]}"
        for a in week["articles"][:20]
    )
    analysis = requests.post(f"{MV_BASE}/mave/chat", headers=MV_H, json={
        "message": f"Sentiment analyst. Analyze media coverage for '{TOPIC}' during {week['week_label']}.\n\n"
            f"ARTICLES ({week['count']}):\n{corpus}\n\n"
            "Return JSON with: overall_sentiment (positive/negative/neutral/mixed), "
            "confidence (0-100), positive_pct, negative_pct, neutral_pct, "
            "top_positive_theme, top_negative_theme, notable_shift, key_quote."
    }).json()
    content = analysis.get("content", "")
    sentiment_series.append({"week": week["week_label"], "count": week["count"], "analysis": content[:600]})
    time.sleep(0.5)

# 3. Trend summary
trend = requests.post(f"{MV_BASE}/mave/chat", headers=MV_H, json={
    "message": f"Given these {WEEKS} weeks of sentiment data for '{TOPIC}', identify the overall trend.\n\n"
        + "\n\n".join(f"**{s['week']}** ({s['count']} articles):\n{s['analysis']}" for s in sentiment_series)
        + "\n\nProvide: trend direction (improving/declining/stable), confidence, biggest driver, prediction for next week, recommended action."
}).json()

print(f"\n{'='*60}\nSENTIMENT TREND: {TOPIC}\n{'='*60}")
for s in sentiment_series:
    print(f"\n{s['week']} ({s['count']} articles):\n{s['analysis'][:300]}")
print(f"\n{'='*60}\nTREND ANALYSIS\n{'='*60}")
print(trend.get("content", "")[:800])

​
Example Output

Week 0: 25 articles (Mar 10 – Mar 17)
Week 1: 22 articles (Mar 03 – Mar 10)
Week 2: 19 articles (Feb 24 – Mar 03)
Week 3: 23 articles (Feb 17 – Feb 24)

SENTIMENT TREND: artificial intelligence marketing
============================================================

Mar 10 – Mar 17 (25 articles):
  {"overall_sentiment": "positive", "confidence": 78,
   "positive_pct": 56, "negative_pct": 20, "neutral_pct": 24,
   "top_positive_theme": "Personalization ROI gains",
   "top_negative_theme": "Privacy regulation fears"}

TREND ANALYSIS
============================================================
Direction: IMPROVING (↑ 12% positive shift over 4 weeks)
Driver: Enterprise adoption stories replacing hype-cycle skepticism.
Prediction: Continued positive bias until Q2 earnings season.
Action: Publish thought leadership now — sentiment tailwind amplifies reach.

​
Error Handling

Free tier limits search to past 30 days. Reduce WEEKS to 4 max on free plans. Business tier supports full archive.
Mave returns natural language by default. Prompt for JSON explicitly and parse with json.loads() / JSON.parse(). Wrap in try/catch for malformed responses.
Broad topics like “AI” return noise. Use quoted phrases and boolean operators: "AI marketing" AND ("ROI" OR "attribution").