快速入门

1. 1.开始节点
2. 无需添加任何自定义变量，只需使用系统变量 Query（用户提出的问题）。

要了解更多关于变量及其使用的信息，请点击 这里。

1. 2.代码节点
2. 此节点的功能是计算一个时间范围，否则在 HTTP 请求节点获取信息时会有大量数据。这里我们以用户输入日期的前30天为例。

• 输入变量：此节点不需要任何输入变量。
• Python：以下是一个计算30天范围的代码示例，您可以复制并粘贴到块中。

from datetime import datetime, timedelta

def main():
    # Get the current time
    current_time = datetime.now()

    # Calculate the datetime 30 days ago
    thirty_days_ago = current_time - timedelta(days=30)

    # Format the datetime for NewsAPI's 'from' parameter (YYYY-MM-DDTHH:MM:SS)
    formatted_from_date = thirty_days_ago.strftime('%Y-%m-%dT%H:%M:%S')

    # New: Current date for report display (YYYY-MM-DD)
    report_date = current_time.strftime('%Y-%m-%d') 

    # Return a dictionary containing the variables to be passed to the next nodes
    return {
        "from_date": formatted_from_date,  # Output variable for NewsAPI's 'from' parameter
        "report_date": report_date         # Output variable for the report's current date
    }

如果您需要不同的时间范围，可以随意更改“days=30”或“thirty_days_ago”字段，但请确保在所有地方保持一致。

即使您不熟悉编码也没关系。点击 AI 编码，告诉 AI 您的需求，AI 将自动输出满足您需求的代码。



• 输出变量：

  ● from_date

  ● report_date

均为“字符串”。

1. 3.HTTP 请求节点
2. 此节点的功能是从特定数据源获取数据，可以是订阅的 RSS 源或某些网站的 API。

   

   、li>
3. 您还可以直接用我们内置工具替换此节点，如谷歌搜索。

4. 这里我们以 NewsAPI 为例获取最新的行业新闻。
5. 首先，注册获取 API Key：https://newsapi.org/。点击“Get API Key”后，您将在页面上找到 API 密钥。请复制并保存，我们稍后需要使用它。



6. 返回 GoInsight.AI 的 InsightFlow 画布，并按以下配置填写。

字段	配置内容	目的
API 方法	GET	HTTP 方法。
API URL	https://newsapi.org/v2/everything	NewsAPI 的通用新闻搜索端点。
Headers	键：X-Api-Key 值：您的 NewsAPI API Key	NewsAPI 认证所需的 API Key。
参数	键：query 值：{{Start.query}} 键：language 值：en 键：sortBy 值：publishedAt 键：pageSize 值：50 键：from 值：{{Code.from_date}}	查询参数，定义搜索的关键词、语言、排序顺序、数量和时间范围。
Body	无	GET 请求不需要请求体。
请求超时	60	设置请求超时。
输出变量	news_api_response	将 API 响应存储在名为 news_api_response 的变量中以供后续使用。

1. 4.代码节点
2. 此节点的功能是对 API 返回的原始新闻数据进行预处理，如过滤关键字段和合并文本，以为大模型总结做准备。



• 输入变量：变量名设置为news_data，值填入前一个节点的输出，即{{HTTP_Request.Body}}。
• Python：以下是一个示例：

import json

def main(news_data):
    try:
        data = news_data
        # If news_data is a string, it needs to be parsed with json.loads()
        if isinstance(news_data, str):
            data = json.loads(news_data)

    except json.JSONDecodeError:
        # If parsing fails, return a specific error message
        return {"processed_articles_text": "Error: Could not parse news data as JSON."}
    except TypeError:
        # If news_data is not valid (e.g., None), return an error
        return {"processed_articles_text": "Error: news_data is not valid."}

    processed_articles = []
    # Check if 'articles' key exists and its value is a list (this is NewsAPI's structure)
    if data and 'articles' in data and isinstance(data['articles'], list):
        for article in data['articles']:
            title = article.get('title', '')
            description = article.get('description', '')
            url = article.get('url', '') # Ensure URL is extracted

            # Filter out articles with empty titles or descriptions to ensure meaningful content for the LLM
            # Ensure URL is also present, as it's useful for the report
            if title and description and url:
                # IMPORTANT CHANGE HERE: Include the link in a Markdown link format
                # LLMs generally understand and preserve this format well.
                combined_text = f"Title: {title}\nDescription: {description}\n[Source Link]({url})"
                processed_articles.append(combined_text)

    if not processed_articles:
        return {"processed_articles_text": "No valid articles found after processing NewsAPI response."}

    # Concatenate the processed articles into a single string, separated by a distinct delimiter.
    # Limit the number of articles (e.g., top 20) to prevent the LLM input from becoming too long.
    return {"processed_articles_text": "\n\n---\n\n".join(processed_articles[:20])}

• 输出变量：设置一个输出变量：processed_articles_text，类型为“字符串”。

1. 5.大模型节点
2. 此节点用于根据系统角色和指令生成包含总结和链接的洞察报告。

• 模型：这里我们选择Azure GPT-4o mini。
• 系统：这是“角色设置”，即对大模型的提示指令，也称为提示词。它是工作流能否按预期工作的关键。为了让大模型更好地理解，尽量将内容结构化。以下是一个示例：

### Industry Trend Insight Report - {{Code.report_date}}
**Analysis Keywords:** {{Start.Query}}

---

**Instructions for Report Generation:**

You are an expert industry analyst. Your task is to generate a comprehensive "Industry Trend Insight Report" based on the provided news articles. Adhere strictly to the following structure and guidelines:

**Output Structure:**

#### 1. Key Trends Summary
- **Trend Name:** [Identify a concise trend name]
  - **Core Content:** [Summarize the main content of the trend based on the articles. Ensure this is direct and informative.]
  - **Key Events/News:** [Mention specific events or news items that support this trend. **For each event/news mention, if a source link is available within the provided articles (e.g., "[Source Link](URL)"), incorporate this exact Markdown link directly into the report after the relevant detail.**]
  - **Potential Impact:** [Analyze the potential implications or future developments related to this trend.]

#### 2. Key Company/Product Developments
- [Identify specific companies or products mentioned in the news articles that highlight significant developments. **For each company/product development, if a source link is available, include this exact Markdown link.**]

#### 3. Potential Risks & Opportunities
- **Risks:** [Identify potential risks related to the overall industry or trends based on the articles.]
- **Opportunities:** [Identify potential opportunities arising from the trends or developments.]

**Important Guidelines:**

* **Data Source:** All information, analyses, and summaries must be strictly derived from the `【News Articles List】` provided below.
* **Source Link Integration:** When extracting `Key Events/News` or `Key Company/Product Developments`, if the original article's text contains a `[Source Link](URL)` format, **you MUST include this exact Markdown link within your report right after the relevant piece of information.** Do not modify the link format or content. If multiple sources support a single point, pick the most relevant one, or include a few if appropriate and concise.
* **Concise and Professional:** The language style should be professional, objective, and the report content should be concise and to the point.
* **English Report:** All content should be output in English.
* **No Fabrication:** Do not invent or fabricate any information not present in the provided articles.
* **Data Sufficiency:** If the provided news articles are insufficient to form a detailed analysis, or if no significant trends can be identified, state this clearly at the end. Otherwise, provide a concluding remark on sufficiency.

---

**【News Articles List】**
{{Code.processed_articles_text}}

提示：您还可以点击“AI 写作”让 AI 自动为您撰写或优化提示词。

• 用户：向模型提供说明、查询或任何基于文本的输入。以下是一个示例：

Here is a list of recent news articles related to the industry keywords "{{Start.Query}}". Please analyze and summarize them according to the role instructions provided: 
【News Articles List】 
{{Code(1).processed_articles_text}}

1. 6.回复输出节点
• 回复输出：设置为上一步大模型节点的输出变量文本