4c86a5b9ca
Self-hosted US Congress monitoring platform with AI policy briefs, bill/member/topic follows, ntfy + RSS + email notifications, alignment scoring, collections, and draft-letter generator. Authored by: Jack Levy
113 lines
3.7 KiB
Python
113 lines
3.7 KiB
Python
"""
|
||
Google Trends service (via pytrends).
|
||
|
||
pytrends is unofficial web scraping — Google blocks it sporadically.
|
||
All calls are wrapped in try/except and return 0 on any failure.
|
||
"""
|
||
import logging
|
||
import random
|
||
import time
|
||
|
||
from app.config import settings
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
|
||
def get_trends_score(keywords: list[str]) -> float:
|
||
"""
|
||
Return a 0–100 interest score for the given keywords over the past 90 days.
|
||
Returns 0.0 on any failure (rate limit, empty data, exception).
|
||
"""
|
||
if not settings.PYTRENDS_ENABLED or not keywords:
|
||
return 0.0
|
||
try:
|
||
from pytrends.request import TrendReq
|
||
|
||
# Jitter to avoid detection as bot
|
||
time.sleep(random.uniform(2.0, 5.0))
|
||
|
||
pytrends = TrendReq(hl="en-US", tz=0, timeout=(10, 25))
|
||
kw_list = [k for k in keywords[:5] if k] # max 5 keywords
|
||
if not kw_list:
|
||
return 0.0
|
||
|
||
pytrends.build_payload(kw_list, timeframe="today 3-m", geo="US")
|
||
data = pytrends.interest_over_time()
|
||
|
||
if data is None or data.empty:
|
||
return 0.0
|
||
|
||
# Average the most recent 14 data points for the primary keyword
|
||
primary = kw_list[0]
|
||
if primary not in data.columns:
|
||
return 0.0
|
||
|
||
recent = data[primary].tail(14)
|
||
return float(recent.mean())
|
||
|
||
except Exception as e:
|
||
logger.debug(f"pytrends failed (non-critical): {e}")
|
||
return 0.0
|
||
|
||
|
||
def get_trends_scores_batch(keyword_groups: list[list[str]]) -> list[float]:
|
||
"""
|
||
Get pytrends scores for up to 5 keyword groups in a SINGLE pytrends call.
|
||
Takes the first (most relevant) keyword from each group and compares them
|
||
relative to each other. Falls back to per-group individual calls if the
|
||
batch fails.
|
||
|
||
Returns a list of scores (0–100) in the same order as keyword_groups.
|
||
"""
|
||
if not settings.PYTRENDS_ENABLED or not keyword_groups:
|
||
return [0.0] * len(keyword_groups)
|
||
|
||
# Extract the primary (first) keyword from each group, skip empty groups
|
||
primaries = [(i, kws[0]) for i, kws in enumerate(keyword_groups) if kws]
|
||
if not primaries:
|
||
return [0.0] * len(keyword_groups)
|
||
|
||
try:
|
||
from pytrends.request import TrendReq
|
||
|
||
time.sleep(random.uniform(2.0, 5.0))
|
||
pytrends = TrendReq(hl="en-US", tz=0, timeout=(10, 25))
|
||
kw_list = [kw for _, kw in primaries[:5]]
|
||
|
||
pytrends.build_payload(kw_list, timeframe="today 3-m", geo="US")
|
||
data = pytrends.interest_over_time()
|
||
|
||
scores = [0.0] * len(keyword_groups)
|
||
if data is not None and not data.empty:
|
||
for idx, kw in primaries[:5]:
|
||
if kw in data.columns:
|
||
scores[idx] = float(data[kw].tail(14).mean())
|
||
return scores
|
||
|
||
except Exception as e:
|
||
logger.debug(f"pytrends batch failed (non-critical): {e}")
|
||
# Fallback: return zeros (individual calls would just multiply failures)
|
||
return [0.0] * len(keyword_groups)
|
||
|
||
|
||
def keywords_for_member(first_name: str, last_name: str) -> list[str]:
|
||
"""Extract meaningful search keywords for a member of Congress."""
|
||
full_name = f"{first_name} {last_name}".strip()
|
||
if not full_name:
|
||
return []
|
||
return [full_name]
|
||
|
||
|
||
def keywords_for_bill(title: str, short_title: str, topic_tags: list[str]) -> list[str]:
|
||
"""Extract meaningful search keywords for a bill."""
|
||
keywords = []
|
||
if short_title:
|
||
keywords.append(short_title)
|
||
elif title:
|
||
# Use first 5 words of title
|
||
words = title.split()[:5]
|
||
if len(words) >= 2:
|
||
keywords.append(" ".join(words))
|
||
keywords.extend(tag.replace("-", " ") for tag in (topic_tags or [])[:3])
|
||
return keywords[:5]
|