Policy Lifecycle

The Policy Lifecycle (PLC) model classifies Chinese government documents according to their position in the policy development process. This classification helps track and analyse how policies evolve from initial conception through to implementation.

Policy lifecycle fields

uuid

  • Type: String
  • Description: Unique identifier for the document being classified.
  • Nullable: Yes

plc_label

  • Type: String
  • Description: The assigned policy lifecycle stage. One of: NOT_POLICY, INFORMING, DECIDING, or IMPLEMENTING.
  • Nullable: Yes

plc_label_scores

  • Type: Array of floats
  • Description: Confidence scores for each classification category. Contains four float values in the order [NOT_POLICY, INFORMING, DECIDING, IMPLEMENTING]. Values range from 0.0 to 1.0 and sum to approximately 1.0. Higher values indicate greater model certainty.
  • Example: [0.0002, 0.0001, 0.9995, 0.0002]
  • Nullable: Yes

Policy lifecycle stages

The PLC model classifies documents into four distinct stages:

0. NOT_POLICY

Documents that do not contain policy-relevant content or are not part of the policy-making process.

Characteristics:

  • Administrative notices unrelated to policy
  • Routine operational documents
  • General announcements without policy implications

Example classification:

{
  "uuid": "15a4bb4c-df79-c066-85cb-cba18f62281f2f0837bf-6...",
  "plc_label": "NOT_POLICY",
  "plc_label_scores": [0.9998, 0.0001, 0.0001, 0.0]
}

1. INFORMING

Documents that provide information, context, or background related to policy issues. These documents typically describe problems, present research findings, or communicate situational awareness that may inform future policy decisions.

Characteristics:

  • Reports and analyses of current conditions
  • Research findings and statistical data
  • Problem identification documents
  • Public consultations and feedback collection

Example classification:

{
  "uuid": "83cbaf38-e84e-48b3-c73a-24fef72a1d949b8f6862-1...",
  "plc_label": "INFORMING",
  "plc_label_scores": [0.0017, 0.9977, 0.0003, 0.0003]
}

2. DECIDING

Documents that represent the decision-making phase of policy development, including formal policy proposals, draft regulations, deliberations, and official policy announcements.

Characteristics:

  • Policy proposals and recommendations
  • Draft legislation and regulations
  • Official policy decisions and approvals
  • Strategic plans and frameworks

Example classification:

{
  "uuid": "fff503b6-f57c-605f-13fc-293b9c0f77f3aa46aebe-4...",
  "plc_label": "DECIDING",
  "plc_label_scores": [0.0002, 0.0001, 0.9995, 0.0002]
}

3. IMPLEMENTING

Documents related to the execution and operationalisation of established policies, including implementation guidelines, enforcement actions, and progress reports.

Characteristics:

  • Implementation guidelines and procedures
  • Enforcement notices and compliance measures
  • Progress reports and evaluation documents
  • Operational directives for executing policy

Example classification:

{
  "uuid": "...",
  "plc_label": "IMPLEMENTING",
  "plc_label_scores": [0.0, 0.0, 0.0002, 0.9998]
}

Understanding confidence scores

The plc_label_scores array provides confidence scores for all four categories:

  • The four scores form a discrete probability distribution: They are non-negative, range from 0.0 to 1.0, and sum to 1.0
  • The assigned plc_label corresponds to the category with the highest probability score
  • Users may want to apply threshold filtering based on confidence scores depending on use case requirements and tolerance for misclassification

For example, you might only use classifications where the winning score exceeds 0.90 for high-confidence analysis, or accept lower thresholds for exploratory work.

Working with policy lifecycle data

Tracking policy development

The PLC classification allows you to track how specific policy topics evolve over time:

  1. INFORMING stage: Early research and problem identification
  2. DECIDING stage: Policy formulation and decision-making
  3. IMPLEMENTING stage: Execution and enforcement

By filtering documents by both topic (using entity extraction or text search) and PLC stage, you can observe the progression of policy initiatives.

Filtering by stage

Different analytical objectives may focus on different stages:

  • Risk monitoring: Focus on DECIDING stage documents to identify new policy directions before implementation
  • Implementation analysis: Focus on IMPLEMENTING stage documents to understand how policies are being executed
  • Early signals: Monitor INFORMING stage documents to detect emerging policy concerns

Ambiguous documents

Some documents may have relatively balanced scores across multiple categories. These ambiguous cases often represent:

  • Documents that span multiple stages (e.g., announcing a decision while also providing implementation guidance)
  • Transitional documents between policy stages
  • Documents with mixed content

Consider the distribution of scores, not just the top label, when interpreting these cases.