32 lines
3.1 KiB
JSON
32 lines
3.1 KiB
JSON
{
|
|
"cot.predict": {
|
|
"traces": [],
|
|
"train": [],
|
|
"demos": [],
|
|
"signature": {
|
|
"instructions": "You are given a short document snippet (referred to as \"context\") which you must classify into one of the supported label buckets: healthcare, tax, legal, personal, or other.\n\nDetailed guidelines for classification:\n\n1. **Focus on Primary Topic and Purpose** \n Analyze the main subject and the principal intent of the document snippet rather than isolated keywords. Some keywords might be ambiguous or appear in multiple domains (e.g., \u201chealthcare\u201d terms can appear in tax-related documents).\n\n2. **Healthcare** \n Includes documents primarily related to medical information, treatment guidance, patient care, health instructions, diagnoses, medications, health-related reports, and direct clinical or care-oriented content.\n\n3. **Tax** \n Includes documents primarily concerned with taxation matters, such as tax filings, reimbursements with fiscal implications, deductions, tax reporting, financial records related to taxation, and any healthcare-related content that is framed in the context of taxation (e.g., healthcare reimbursements that are relevant for tax purposes).\n\n4. **Legal** \n Pertains to legal agreements, contracts, laws, legal instructions, compliance documentation, or legal advice.\n\n5. **Personal** \n Includes private, individual matters not related to healthcare, tax, or legal domains\u2014such as personal correspondence, diaries, or private notes.\n\n6. **Other** \n Covers business, marketing, product documentation, and any content that does not fall into the above categories.\n\n7. **Disambiguation Strategy** \n - When document snippets include topics crossing domains, classify according to the dominant or primary classification relevant to the document\u2019s context and purpose. \n - For example, although a document may mention \"healthcare\", if the snippet describes healthcare in the context of monetary reimbursements for tax filing purposes, classify as tax rather than healthcare. \n\n8. **Output Requirements** \n Provide: \n - A **reasoning** paragraph explaining your classification decision, demonstrating how you determined the primary domain. \n - A **label** field stating the chosen classification (healthcare, tax, legal, personal, or other).\n\nBy following these detailed criteria and reasoning carefully about context and domain relevance, you will accurately classify document snippets into the correct label bucket.",
|
|
"fields": [
|
|
{
|
|
"prefix": "Context:",
|
|
"description": "Document text or snippet to classify"
|
|
},
|
|
{
|
|
"prefix": "Reasoning: Let's think step by step in order to",
|
|
"description": "${reasoning}"
|
|
},
|
|
{
|
|
"prefix": "Label:",
|
|
"description": "Return exactly one of: legal, healthcare, personal, tax, or other"
|
|
}
|
|
]
|
|
},
|
|
"lm": null
|
|
},
|
|
"metadata": {
|
|
"dependency_versions": {
|
|
"python": "3.11",
|
|
"dspy": "3.1.3",
|
|
"cloudpickle": "3.1"
|
|
}
|
|
}
|
|
} |