LPXian/Graph-News.AI

1from langchain_core.messages import HumanMessage2from app.core.llm import get_llm3from app.graph.state import GraphState4 5def summarizer_node(state: GraphState) -> GraphState:6    """7    Summarizes individual articles into concise bullet points.8    """9    print("--- SUMMARIZER NODE ---")10    llm = get_llm()11    articles = state.get("clean_articles", [])[:15] # Limit to 15 articles to avoid token issues12    13    summarized_items = []14    15    # Improved structured summarization prompt16    prompt = """Summarize the following AI news articles. 17    For each article, provide a one-sentence technical summary.18    You MUST preserve the URL.19    Return the result in this exact format for each article:20    - [SUMMARY] summary_text_here | [TITLE] article_title_here | [URL] url_here21    """22    23    article_text = ""24    for i, art in enumerate(articles):25        # Ensure we fall back to a string if 'link' is missing, but it should be there.26        link = art.get('link') or art.get('url') or "NO_LINK_AVAILABLE"27        article_text += f"Article {i+1}:\nTitle: {art.get('title')}\nLink: {link}\nContent: {art.get('summary', '')[:400]}\n\n"28    29    response = llm.invoke([HumanMessage(content=prompt + article_text)])30    31    # Extract token usage if available32    tokens = response.usage_metadata.get("total_tokens", 0) if hasattr(response, "usage_metadata") else 033    34    # Extract the lines and return as state.35    summarized_items = response.content.strip().split('\n')36    37    print(f"Processed summaries with explicit URLs. Tokens: {tokens}")38    return {39        "summarized_items": summarized_items,40        "total_tokens": state.get("total_tokens", 0) + tokens41    }42