Skip to main content

Text classification

Question Classification by Topic​

System Prompt

You are an expert in classifying requests. Your task is to determine which topic a question belongs to. Possible topics: cars, food, technology.

User Prompt

What is your mileage on a Tesla Model 3 electric car?

LLM Classifier Response

{
  "predictions": [
    {
      "label": "cars",
      "confidence": 0.85
    },
    {
      "label": "food",
      "confidence": 0.05
    },
    {
      "label": "technology",
      "confidence": 0.10
    }
  ]
}

Classification for Ethics and Politics​

System Prompt

You are a specialist in ethical and political text analysis. Determine whether the text contains ethical or political topics.

User Prompt

Should companies be required to pay their employees more than the minimum wage?

LLM Classifier Response

{
  "predictions": [
    {
      "label": "ethics",
      "confidence": 0.70
    },
    {
      "label": "politics",
      "confidence": 0.30
    }
  ]
}

Classification in RAG (Text Relevance to Question)​

System Prompt

You are an expert in assessing the relevance of texts. Determine whether the given text chunk relates to the given question.

User Prompt

Question: What are the advantages of hybrid cars?
Text: Hybrid cars combine an internal combustion engine and an electric motor, which saves fuel.

LLM Classifier Response

{
  "predictions": [
    {
      "label": "relevant",
      "confidence": 0.92
    },
    {
      "label": "irrelevant",
      "confidence": 0.08
    }
  ]
}

Classification in RAG (Final Answer)​

System Prompt

You are a specialist in assessing the completeness of answers. Determine whether the given answer answers the user's question.

User Prompt

Question: What are the advantages of electric cars?
Answer: Electric cars are environmentally friendly and economical to operate.

LLM Classifier Response

{
  "predictions": [
    {
      "label": "complete answer",
      "confidence": 0.88
    },
    {
      "label": "incomplete answer",
      "confidence": 0.12
    }
  ]
}

Customer Satisfaction Classification​

System Prompt

You are an analyst for assessing customer satisfaction. Determine whether the customer was annoyed in the dialogue.

User Prompt

Customer: I waited an hour for a response! This is unacceptable.

LLM Classifier Response

{
  "predictions": [
    {
      "label": "annoyed",
      "confidence": 0.95
    },
    {
      "label": "not annoyed",
      "confidence": 0.05
    }
  ]
}

Tagging: Reviews​

System Prompt

You are an expert in classifying reviews. Determine which tags are appropriate for this review. Possible tags: quality, delivery, price, support.

User Prompt

The product arrived quickly, but the quality leaves much to be desired.

LLM Classifier Response

{
  "predictions": [
    {
      "label": "delivery",
      "confidence": 0.80
    },
    {
      "label": "quality",
      "confidence": 0.90
    }
  ]
}

User Intent Detection​

System Prompt

You are a specialist in analyzing user intentions. Determine what the user's intent is. Possible intents: purchase, support, information.

User Prompt

Can I learn more about your new product?

LLM Classifier Response

{
  "predictions": [
    {
      "label": "information",
      "confidence": 0.85
    },
    {
      "label": "purchase",
      "confidence": 0.10
    },
    {
      "label": "support",
      "confidence": 0.05
    }
  ]
}

Task Completion Classification​

System Prompt

You are an expert in assessing task completion. Determine whether the task has been completed.

User Prompt

Task: Prepare a sales report for the month.
Answer: The report has been prepared and sent for review.

LLM Classifier Response

{
  "predictions": [
    {
      "label": "completed",
      "confidence": 0.95
    },
    {
      "label": "not completed",
      "confidence": 0.05
    }
  ]
}

Condition Check​

System Prompt

You are an analyst for checking text conditions. Determine whether the text contains personal information, direct speech, or policy violations.

User Prompt

{news_article}

LLM Classifier Response

{
  "predictions": [
    {
      "label": "personal information",
      "confidence": 0.20
    },
    {
      "label": "direct speech",
      "confidence": 0.60
    },
    {
      "label": "policy violation",
      "confidence": 0.10
    }
  ]
}