用階層概化資料

Data Privacy and Anonymization in Python

Rebeca Gonzalez

Data engineer

資料概化（generalization）

讓你以較不精確的值取代原始資料值。

  • 「Dancer」->「Artist」
  • 「Archaeologist」->「Scientist」
  • 「Allergist」->「Doctor」
Data Privacy and Anonymization in Python

以階層進行資料概化

左側列出各種職業，右側為對應的上層分類。

Data Privacy and Anonymization in Python

難民身分資料集

# Explore dataset
refugee_df.head()

     Nationality    Gender    RefugeeStatus    Year
0    russian        F         0                2010
1    colombian      M         0                2008
2    vietnamese     F         1                2008
3    korean         M         0                2010
4    ethiopian      F         1                2012
Data Privacy and Anonymization in Python

探索資料集

# Set k value to 4
k = 4


# Calculate how many unique combinations are for nationality and gender
df_count = df.groupby(['Nationality', 'Gender']).size().reset_index(name='Count')


# Filter the rows that have a count value less than k
df_count[df_count['Count'] < k]

      Nationality   Gender    Count
4     colombian     F         2
7     cuban         M         2
12    liberian      F         2
14    mexican       F         3
15    mexican       M         3
17    russian       M         1
20    ukranian      F         3
21    ukranian      M         1
Data Privacy and Anonymization in Python

以階層進行資料概化

# See the combination counts
df['Nationality'].value_counts()

belarusian    14
korean        14
ethiopian     10
chinese       10
vietnamese     9
taiwanese      9
colombian      6
cuban          6
russian        6
mexican        6
liberian       6
ukranian       4
Name: Nationality, dtype: int64
Data Privacy and Anonymization in Python

以階層進行資料概化

# Create hierarchies for origin countries
hierarchies = {'Europe': ['ukranian','russian','belarusian'], 
               'America':['mexican','colombian','cuban'],
               'Asia': ['taiwanese','korean','chinese'],
               'Africa': ['ethiopian','liberian']}
Data Privacy and Anonymization in Python

以階層進行資料概化

# Creating hierachy father for each contry
origin_hierarchy = {}
for (key, countries) in hierarchies.items():
    for country in countries:
        origin_hierarchy[country] = key

origin_hierarchy

{'belarusian': 'Europe',
 'chinese': 'Asia',
 'colombian': 'America',
 'cuban': 'America',
 'ethiopian': 'Africa',
 'korean': 'Asia',
 'liberian': 'Africa',
 'mexican': 'America',
 'russian': 'Europe',
 'taiwanese': 'Asia',
 'ukranian': 'Europe'}
Data Privacy and Anonymization in Python

以階層進行資料概化

# Apply origin_hierarchy hierarchy generalization to Nationality
df['Nationality_generalized'] = df['Nationality'].map(origin_hierarchy)


# Explore resulting dataset
df.head()

     Nationality   Gender    RefugeeStatus    Year    Nationality_generalized
0    korean        M         1                2019    Asia
1    ethiopian     M         1                2003    Africa
2    cuban         M         0                2015    America
3    cuban         F         0                2001    America
4    korean        M         0                2013    Asia
Data Privacy and Anonymization in Python

以階層進行資料概化

# Calculate how many unique combinations are for Nationality_generalized and Gender
df_count = df.groupby(['Nationality_generalized', 'Gender']).size()
                                                            .reset_index(name='Count')

# Filter the combinations that appear less than k times
df_count[df_count['Count'] < k] 

    Nationality_generalized    Gender    Count
Data Privacy and Anonymization in Python

更多關於 K-anonymity

類型：

  • 限制式方法
  • Mondrian 多維方法
Data Privacy and Anonymization in Python

K-anonymity 有多安全？

表格示範：當敏感屬性缺乏多樣性時，可能發生隱私外洩。例：Bob 在表中，符合所有條件，且所屬的 k 人群都有同一疾病，因此可確定他罹患該病。

Data Privacy and Anonymization in Python

一起來練習吧！

Data Privacy and Anonymization in Python

Preparing Video For Download...