对提取的剖析

An extraction is a structured representation of one of the requests made within a communication. Communications Mining will provide a list of these extractions for use in an automation.

注意：提取由模型训练器在配置分类时定义。

Each extraction has two crucial properties that guide its interpretation. The first is the extraction name, which indicates the subject of the extraction. For instance, an extraction name could be Address Change or Account Update > Contact Number.

第二个重要属性是 fields。字段表示我们旨在从通信中提取的特定数据点。例如，对于 Address Change 提取，相关字段可能包括 AccountNumber、FirstLineOfAddress 和 PostCode。

每个提取内容还附带一个 occurrence_confidence 分数。该分数表示我们对提取准确性的置信度。系统会为此分数应用阈值。通常，置信度分数高于阈值的预测将被标记为自动化，而低于阈值的预测将被标记为需要手动审核。thresholds 字段显示小于预测的每个阈值。

注意：要确定预测置信度是否已超过为流设置的阈值，请检查 stream 阈值。

提取示例

这是一个示例，其中预测了 Policy > Amendment > Address Change 提取，并且此预测的置信度已超过称为 stream 的置信度阈值。

{
  "name": "Policy > Amendment > Address Change",
  "occurrence_confidence": {
    "value": 0.9909818768501282,
    "thresholds": ["stream"]
  },
  "fields": [
    {
      "name": "Address Line",
      "value": {
        "formatted": "45 Brillington Lane"
      }
    },
    {
      "name": "Excess Amount",
      "value": null
    },
    {
      "name": "Policy Number",
      "value": {
        "formatted": "XE-182943"
      }
    },
    {
      "name": "Town / City",
      "value": {
        "formatted": "Doncaster"
      }
    },
    {
      "name": "UK Postcode",
      "value": {
        "formatted": "S45 9UX"
      }
    }
  ]
}