Create market_stock_sentiment.py

core/market_stock_sentiment.py

@@ -0,0 +1,109 @@
+import requests
+import json
+from datetime import datetime, timedelta
+from tradingview_ta import TA_Handler, Interval
+from langchain.document_loaders import WebBaseLoader
+from langchain.docstore.document import Document
+from bs4 import BeautifulSoup
+from GoogleNews import GoogleNews
+from langchain.prompts import PromptTemplate
+from langchain.chains import StuffDocumentsChain, LLMChain
+
+def clean_google_news_url(url: str):
+    for ext in [".html", ".cms"]:
+        if ext in url:
+            return url.split(ext)[0] + ext
+    return url.split("&")[0]
+
+def get_google_news_documents(query: str, max_articles: int = 10, timeout: int = 10):
+    googlenews = GoogleNews(lang="en")
+    end_date = datetime.today()
+    start_date = end_date - timedelta(days=2)
+    googlenews.set_time_range(start_date.strftime("%m/%d/%Y"), end_date.strftime("%m/%d/%Y"))
+    googlenews.search(query)
+    articles = googlenews.result()
+
+    documents = []
+    for article in articles[:max_articles]:
+        url = clean_google_news_url(article.get("link"))
+        try:
+            response = requests.get(url, timeout=timeout, headers={"User-Agent": "Mozilla/5.0"})
+            response.raise_for_status()
+            soup = BeautifulSoup(response.text, "html.parser")
+            paragraphs = soup.find_all("p")
+            content = "\n".join([p.get_text(strip=True) for p in paragraphs if p.get_text(strip=True)])
+            if content and len(content) > 200:
+                doc = Document(
+                    page_content=content,
+                    metadata={
+                        "source": "Google News",
+                        "title": article.get("title", ""),
+                        "published": article.get("date", ""),
+                        "link": url,
+                    }
+                )
+                documents.append(doc)
+        except Exception:
+            continue
+    return documents
+
+def get_market_stock_sentiment(ticker, llm):
+    urls = [
+        f"https://economictimes.indiatimes.com/markets/stocks/news",
+        f"https://www.livemint.com/market/stock-market-news/",
+        f"https://in.tradingview.com/ideas/editors-picks/?type=trade",
+        f"https://pulse.zerodha.com/",
+        "https://upstox.com/news/market-news/stocks/",
+    ]
+
+    loader = WebBaseLoader(urls)
+    web_docs = loader.load()
+
+    google_docs = get_google_news_documents("Indian Stock market news NSE, Stocks in Action, Stocks in News, Stocks to Buy in next few weeks", max_articles=10)
+    web_docs.extend(google_docs)
+
+    prompt_template = """You are an expert Stock Market Trader specializing in stock market insights derived from fundamental analysis, analytical trends, profit-based evaluations, news indicators from different sites and detailed company financials. 
+    You will receive stock market news articles or stocks in news from various news websites which have India stock news feed. For the below context/input_documents, perform the following tasks:
+    Context:
+    {input_documents}
+    1. **Top picks**: After analyzing all provided data, rank the top 5-10 stocks to look at this week, including tickers, current sentiment, and why they made the list.
+    2. **Identify the stock(s)** mentioned (by ticker and company name).
+    3. **Sentiment analysis**: classify as Bullish, Bearish, or Neutral.
+    4. **Extract critical news**: What is the main event or update? (e.g., earnings beat, regulatory approval, management change, major contract or macro impact).
+    5. **Summarize impact**: Briefly explain how this news might affect stock price and investor behavior (e.g., “could boost investor confidence”, “sign indicates profit pressure”, etc.).
+    6. **Actionable signal**: Based on the sentiment and news, suggest whether this is a “Buy”, “Sell”, “Hold”, or “Watch” recommendation, and the rationale.
+    PROVIDE THE DETAILS based on just the FACTS present in the document. Do NOT DUPLICATE the Output & hallucinate.
+    ***Format your output as JSON*** with the following structure:
+    
+    ```json
+    {{
+      "top_picks": [
+        {{
+          "ticker": "TICKER",
+          "company": "Company Name",
+          "sentiment": "Bullish|Bearish|Neutral",
+          "critical_news": "Brief summary of the key event",
+          "impact_summary": "How this may affect the stock",
+          "action": "Buy|Sell|Hold|Watch",
+          "reason": "Why this stock ranks among top picks"
+        }},
+        ...
+      ]
+    }}
+    """
+
+    prompt = PromptTemplate.from_template(prompt_template)
+    chain = StuffDocumentsChain(llm_chain=LLMChain(llm=llm, prompt=prompt), document_variable_name="input_documents")
+    response = chain.invoke({"input_documents": docs})
+    raw = response["output_text"].strip()
+
+    # Clean code block markdown if present
+    if raw.startswith("```json"):
+        raw = raw[len("```json"):]
+    if raw.endswith("```"):
+        raw = raw[:-3]
+
+    try:
+        return json.loads(raw.strip())
+    except json.JSONDecodeError:
+        return {"error": "Failed to parse model output", "raw": raw}