Use Google Gemini Flash 2.0 for Sentiment Analysis

gemini_demo.py

import os
import re
import time
from typing import List, Literal

import dotenv
from bs4 import BeautifulSoup
from google import genai
from google.genai import types
from playwright.sync_api import sync_playwright
from pydantic import BaseModel

dotenv.load_dotenv()

# GH url: https://github.com/googleapis/python-genai
client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])


class TickerSentiment(BaseModel):
    ticker: str
    sentiment: Literal["positive", "negative", "neutral"]
    reason: str


def get_news_sentiment(text: str, ticker: str) -> TickerSentiment:
    prompt = f"""
    You are a financial analyst. You are given a news article and a stock ticker.
    You need to analyze the news article and determine the sentiment of the stock ticker.

    The sentiment can be positive, negative or neutral. Only say it is positive if the news article is about the stock ticker and the news is good for the stock. Only say it is negative if the news article is about the stock ticker and the news is bad for the stock.

    You need to return a JSON object with the following fields:
    - ticker: the stock ticker
    - sentiment: the sentiment of the stock ticker
    - reason: the reason for the sentiment

    The news article is:
    {text}

    The stock ticker is:
    {ticker}
    """  # noqa

    response = client.models.generate_content(
        model="gemini-2.0-flash-exp",
        contents=prompt,
        config=types.GenerateContentConfig(
            response_mime_type="application/json",
            response_schema=TickerSentiment,
        ),
    )
    return response.text


def extract_tickers(text: str) -> List[str]:
    """Extract stock tickers from text using regex pattern."""
    # Pattern for Brazilian stock tickers (4-6 characters followed by 1-2 numbers)
    pattern = r"\b[A-Z]{4,6}[0-9]{1,2}\b"
    return list(set(re.findall(pattern, text)))


# Use playwright to extract the text from the news article
def extract_news_text(url: str) -> str:
    with sync_playwright() as p:
        browser = p.chromium.launch()
        page = browser.new_page()
        page.goto(url)
        text = page.content()
        soup = BeautifulSoup(text, "html.parser")

        if "tradingview.com" in url or "infomoney.com.br" in url:
            # get element with tag <article> and parse with BeautifulSoup
            article = soup.find("article")
            return article.get_text()
        elif "br.investing.com" in url:
            # get div with id article
            article = soup.find("div", id="article")
            return article.get_text()

        raise ValueError(f"Unknown provider: {url}")


def get_sentiments_for_url(url: str) -> List[TickerSentiment]:
    print(f"Getting sentiments for url: {url}")
    text = extract_news_text(url)

    tickers = extract_tickers(text)
    print(f"Found tickers: {tickers}")
    sentiments = []
    for ticker in tickers:
        sentiment = get_news_sentiment(text, ticker)
        sentiments.append(sentiment)
        print(f"Sentiment for {ticker}: {sentiment}")
    return sentiments


def parse_tradingview_news() -> None:
    url = "https://www.tradingview.com/markets/stocks-brazil/news/"

    soup = None

    with sync_playwright() as p:
        browser = p.chromium.launch()
        page = browser.new_page()
        page.goto(url)

        # wait for 2 seconds
        time.sleep(2)

        # open with beautifulsoup
        soup = BeautifulSoup(page.content(), "html.parser")

        browser.close()

    # get the element with class js-news-category-page-container
    container = soup.find("div", class_="js-news-category-page-container")

    # get all anchor tags <a> within the container
    anchors = container.find_all("a")
    all_news_urls = []
    for anchor in anchors:
        href = anchor["href"]
        all_news_urls.append(f"https://www.tradingview.com{href}")

    # remove the last url
    all_news_urls.pop()

    print(f"Found {len(all_news_urls)} news urls")

    for news_url in all_news_urls:
        get_sentiments_for_url(news_url)


# Example with Tradingview:
# url = "https://www.tradingview.com/news/reuters.com,2024:newsml_L2N3MZ0QM:0-brazil-s-ecorodovias-wins-highway-concession-in-sao-paulo-shares-plunge/"

url = "https://www.tradingview.com/news/reuters.com,2024:newsml_L1N3NH0VC:0-brazil-s-embraer-sells-12-a-29n-super-tucano-aircrafts-to-portugal/"

# Example with Infomoney:
# url = "https://www.infomoney.com.br/mercados/luiz-barsi-neto-licoes-historias-caso-oi-no-mercado-financeiro/"
get_sentiments_for_url(url)