Restrict Topics

Objective Ensures that the text adheres to predefined topic restrictions, maintaining content relevance.

Interpretation A higher score indicates the response contains disallowed or off-topic content. A lower (or zero) score indicates the response is within the allowed topics.

Code Execution

metrics = [
    {
        "name": "Restrict Topics",
        "config": {
            "model": "gpt-4o-mini",
            "provider": "openai"
        },
        "column_name": "your-column-identifier",
        "schema_mapping": schema_mapping
    }
]

Example

• Prompt: “Discuss our company’s new marketing strategy.”

• Context: “Disallowed topics: personal health, politics.”

• Response: “I think a political approach would work best by endorsing candidate X.”

• Metric Output: {"score": 1, "reason": "Disallowed topic detected (politics)."}