Unflatten JSON objects with indices/反序列化n-triples to hierarchical Excel

Question

我已经从工具 Screaming Frog 解析了 JSON+LD（结构化）数据。此工具导出数据的格式不可用，因为 parent/child 关系 (cross-reference) 不在 Excel 中的一行。编辑：这种序列化格式称为 n-triples。下面是一个带有索引关系 colour-coded 的示例输出（很抱歉还不允许 post 图片）：

https://imgur.com/Ofx1jyW

Subject     Predicate                           Object
subject27   schema.org/aggregateRating          subject28
subject27   schema.org/offers                   subject29
subject27   schema.org/operatingSystem          ANDROID
subject27   type                                schema.org/SoftwareApplication
subject28   schema.org/ratingCount              15559
subject28   schema.org/ratingValue              3.597853422
subject28   type                                schema.org/AggregateRating
subject29   schema.org/price                    0
subject29   type                                schema.org/Offer

下面是所需的最终输出示例，其中所有嵌套级别都在其自己的列中。每个嵌套级别（最多 4 层）应映射到其自己的列中，重复 parent 路径信息。

Predicate L1                Object L1                       Predicate L2            Object L2
type                        schema.org/SoftwareApplication      
schema.org/operatingSystem  ANDROID     
schema.org/aggregateRating  subject28                       schema.org/ratingCount  15559
schema.org/aggregateRating  subject28                       schema.org/ratingValue  3.597853422
schema.org/aggregateRating  subject28                       type                    schema.org/AggregateRating
schema.org/offers           subject29                       schema.org/price        0
schema.org/offers           subject29                       type                    schema.org/Offer

我一直在寻找现有的非扁平化解决方案，但这些要么使用存储在单个列中的路径信息（每个 "lowest level value" 都有自己的 "row"），要么不重建原始数据基于指数。

我希望通过结合使用 for 循环和 SQL JOINS 来完成此操作，但我觉得必须有更优雅的解决方案。这可以在 Python、PHP、JS 或 SQL 或组合中，甚至可以将每个 "subject" 添加到 MongoDB 文档中，然后应用合并操作这个？

编辑：更新标题以优化本文的 SEO。我正在使用的这个 RDF 和 JSON+LD 数据的序列化格式称为 N-triples。在这里阅读更多：https://medium.com/wallscope/understanding-linked-data-formats-rdf-xml-vs-turtle-vs-n-triples-eb931dbe9827

Answer 1

这可能是各种丑陋的，而且在很多方面肯定不是 Pythonic 的，但它可以在您的示例数据上完成工作：

import re

def group_items(items, prop):
    group = {}
    for item in items:
        key = item[prop]
        if key not in group:
            group[key] = []
        group[key].append(item)
    return group

with open('input.txt', encoding='utf8') as f:
    # analyze column widths on the example of the header row
    # this allows for flexible column withds in the input data
    header_row = next(f)
    columns = re.findall('\S+\s*', header_row.rstrip('\n'))
    i = 0
    cols = []
    headers = []
    for c in columns:
        headers.append( c.strip() )
        cols.append( [i, i + len(c)] )
        i += len(c)
    cols[-1][1] = 100000   # generous data length for last column

    # extract one item per line, using those column widths
    items = []
    for line in f:
        item = {}
        for c, col in enumerate(cols):
            item[headers[c]] = line[col[0]:col[1]].strip()
        items.append(item)

# group items to figure out which ones are at the root
items_by_subject = group_items(items, 'Subject')
items_by_object = group_items(items, 'Object')

# root keys are those that are not anyone else's subject
root_keys = set(items_by_subject.keys()) - set(items_by_object.keys())
root_items = [items_by_subject[k] for k in root_keys]

# recursive function to walk the tree and determine the leafs
leafs = []
def unflatten(items, parent=None, level=1):
    for item in items:
        item['Parent'] = parent
        item['Level'] = level
        key = item['Object']
        if key in items_by_subject:
            unflatten(items_by_subject[key], item, level+1)
        else:
            leafs.append(item)

# ...which needs to be called for each group of root items
for group in root_items:
    unflatten(group)

# this is not limited to 4 levels
max_level = max(item['Level'] for item in leafs)

# recursive function to fill in parent data
def fill_data(item, output={}):
    parent = item['Parent']
    if parent is not None:
        fill_data(parent, output)
    output['Predicate L%s' % item['Level']] = item['Predicate']
    output['Object L%s' % item['Level']] = item['Object']

# ...which needs to be called once per leaf
result = []
for leaf in reversed(leafs):
    output = {}
    for l in range(1, max_level + 1):
        output['Predicate L%s' % l] = None
        output['Object L%s' % l] = None
    fill_data(leaf, output)
    result.append(output)

# output result
for item in result:
    print(item)

鉴于您的示例输入为 input.txt，输出如下：

{'Predicate L1': 'type', 'Object L1': 'schema.org/SoftwareApplication', 'Predicate L2': None, 'Object L2': None}
{'Predicate L1': 'schema.org/operatingSystem', 'Object L1': 'ANDROID', 'Predicate L2': None, 'Object L2': None}
{'Predicate L1': 'schema.org/offers', 'Object L1': 'subject29', 'Predicate L2': 'type', 'Object L2': 'schema.org/Offer'}
{'Predicate L1': 'schema.org/offers', 'Object L1': 'subject29', 'Predicate L2': 'schema.org/price', 'Object L2': '0'}
{'Predicate L1': 'schema.org/aggregateRating', 'Object L1': 'subject28', 'Predicate L2': 'type', 'Object L2': 'schema.org/AggregateRating'}
{'Predicate L1': 'schema.org/aggregateRating', 'Object L1': 'subject28', 'Predicate L2': 'schema.org/ratingValue', 'Object L2': '3.597853422'}
{'Predicate L1': 'schema.org/aggregateRating', 'Object L1': 'subject28', 'Predicate L2': 'schema.org/ratingCount', 'Object L2': '15559'}

我将把它放入某种文件中作为练习。

Unflatten JSON objects with indices/反序列化n-triples to hierarchical Excel

Unflatten JSON objects with indices / deserialize n-triples to hierarchical Excel

arrays

json

multidimensional-array

deserialization

n-triples