将表格从 PDF 提取到 Azure 搜索

Question

我正在尝试将 pdf 内容编入 azure 搜索索引。为此，我在 Azure Search Power Skills GitHub 存储库中使用 analyzeForm 和 extractTables 项目的函数应用程序。

AnalyzeForm 具有字段映射，我可以使用它来将输出字段映射到相应的 Azure 搜索索引字段。但我还想提取 table 内容，为此我必须使用 ExtractTables API。 API returns table 这种形式的记录：

{
    "values": [
        {
            "recordId": "record1",
            "data": {
                "tables": [
                    {
                        "page_number": 1,
                        "row_count": 3,
                        "column_count": 4,
                        "cells": [
                            {
                                "text": "Item",
                                "rowIndex": 0,
                                "colIndex": 0,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "Quantity",
                                "rowIndex": 0,
                                "colIndex": 1,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "Rate",
                                "rowIndex": 0,
                                "colIndex": 2,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "Amount",
                                "rowIndex": 0,
                                "colIndex": 3,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "Iphone 12 (64 GB)",
                                "rowIndex": 1,
                                "colIndex": 0,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "1",
                                "rowIndex": 1,
                                "colIndex": 1,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "0.00",
                                "rowIndex": 1,
                                "colIndex": 2,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "0.00",
                                "rowIndex": 1,
                                "colIndex": 3,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "MI 10 (6 GB)",
                                "rowIndex": 2,
                                "colIndex": 0,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "1",
                                "rowIndex": 2,
                                "colIndex": 1,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "0.00",
                                "rowIndex": 2,
                                "colIndex": 2,
                                "confidence": 1.0,
                                "is_header": false
                            },
                            {
                                "text": "0.00",
                                "rowIndex": 2,
                                "colIndex": 3,
                                "confidence": 1.0,
                                "is_header": false
                            }
                        ]
                    }
                ]
            }
        }
    ]
}

如何映射从 table 中提取的数据以将其索引到 Azure 搜索中？有什么办法可以将其添加为自定义技能吗？如果是，索引模式是什么？

Answer 1

您可以将 Azure 函数用作自定义技能。只需将 ExtractTables 中的字段映射到预期的自定义技能字段：

private class OutputRecord
{
    public class OutputRecordData
    {
        public string Name { get; set; } = "";
        public string Description { get; set; } = "";
        public string Source { get; set; } = "";
        public string SourceUrl { get; set; } = "";
        public string LicenseAttribution { get; set; } = "";
        public string LicenseUrl { get; set; } = "";
    }

    public class OutputRecordMessage
    {
        public string Message { get; set; }
    }

    public string RecordId { get; set; }
    public OutputRecordData Data { get; set; }
    public List<OutputRecordMessage> Errors { get; set; }
    public List<OutputRecordMessage> Warnings { get; set; }
}

https://docs.microsoft.com/en-us/azure/search/cognitive-search-create-custom-skill-example

Answer 2

有 3 种方法可以转换技能输出，您可以使用整形器技能、内联整形或使用自定义技能。在这种情况下，我同意@Thiago 的观点，即最好的选择是使用自定义技能，因为整形器不允许您仅从 table 结果中选择特定的列值。

如果您正在使用 Power Skills，您可以将技能编辑为 return 映射到您的索引字段定义的响应。假设您的索引字段如下所示：

    {
      "name": "Items",
      "type": "Collection(Edm.ComplexType)",
      "analyzer": null,
      "synonymMaps": [],
      "fields": [
        {
          "name": "Item_Id",
          "type": "Edm.String",
          "facetable": false,
          "filterable": false,
          "key": false,
          "retrievable": false,
          "searchable": false,
          "sortable": false,
          "analyzer": null,
          "indexAnalyzer": null,
          "searchAnalyzer": null,
          "synonymMaps": [],
          "fields": []
        },
        {
          "name": "Quantity",
          "type": "Edm.Int64",
          "facetable": false,
          "filterable": false,
          "retrievable": false,
          "sortable": false,
          "analyzer": null,
          "indexAnalyzer": null,
          "searchAnalyzer": null,
          "synonymMaps": [],
          "fields": []
        },
        {
          "name": "Rate",
          "type": "Edm.Double",
          "facetable": false,
          "filterable": false,
          "retrievable": false,
          "sortable": false,
          "analyzer": null,
          "indexAnalyzer": null,
          "searchAnalyzer": null,
          "synonymMaps": [],
          "fields": []
        },
        {
          "name": "Amount",
          "type": "Edm.Double",
          "facetable": false,
          "filterable": false,
          "retrievable": false,
          "sortable": false,
          "analyzer": null,
          "indexAnalyzer": null,
          "searchAnalyzer": null,
          "synonymMaps": [],
          "fields": []
        }
      ]
    }

您可以将您的技能编辑为 return 一个 JSON 对象，例如：

{
    "items" : [
        {
            "item": "Iphone 12",
            "quantity": 1,
            "price": 1000,
            "amount: 1000
        }
    ]
}

将表格从 PDF 提取到 Azure 搜索

Extract tables from PDF to Azure search

azure

indexing

azure-cognitive-search

azure-cognitive-services

form-recognizer