從文本pyspark中提取字符串

Question

我有一個 pyspark 數據框：

例子：

text <String>                 |   name <String>  |   original_name <String>
----------------------------------------------------------------------------
HELLOWORLD2019THISISGOOGLE    |   WORLD2019      |   WORLD_2019
----------------------------------------------------------------------------
NATUREISVERYGOODFORHEALTH     |   null           |   null  
----------------------------------------------------------------------------
THESUNCONTAINVITAMIND         |   VITAMIND       |   VITAMIN_D
----------------------------------------------------------------------------
BECARETOOURHEALTHISVITAMIND   |   OURHEALTH      |   OUR_/HEALTH
----------------------------------------------------------------------------

我想循環name列並查看text中是否存在name值，如果是，我創建一個new_column ，將包含text找到的name值的original_name值。 知道有時數據框列是null 。

例子：

• 在數據框示例的第 4 行中， text包含來自name列的 2 個值： [OURHEALTH, VITAMIND] ，我應該獲取它的original_name值並將它們存儲在new_column 。

• 在第2行中， text包含OURHEALTH從name列中，我應在存儲new_column原始name值發現==> [OUR_/HEALTH]

期待結果：

text <String>                 |   name <String>  |   original_name <String>  | new_column <Array>
------------------------------|------------------|---------------------------|----------------------------
HELLOWORLD2019THISISGOOGLE    |   WORLD2019      |   WORLD_2019              |  [WORLD_2019]
------------------------------|------------------|---------------------------|----------------------------
NATUREISVERYGOODFOROURHEALTH  |   null           |   null                    |  [OUR_/HEALTH]
------------------------------|------------------|---------------------------|----------------------------
THESUNCONTAINVITAMIND         |   VITAMIND       |   VITAMIN_D               |  [VITAMIN_D]
------------------------------|------------------|---------------------------|----------------------------
BECARETOOURHEALTHISVITAMIND   |   OURHEALTH      |   OUR_/HEALTH             |  [OUR_/HEALTH, VITAMIN_D ]
-----------------------------------------------------------------------------|----------------------------

我希望我的解釋清楚。

我嘗試了以下代碼：

df = df.select("text", "name", "original_name").agg(collect_set("name").alias("name_array"))
for name_item in name_array:    
    df.withColumn("new_column", F.when(df.text.contains(name_item), "original_name").otherwise(None))

有人可以幫我嗎？ 謝謝

Answer 1

一種簡單的解決方案是在原始 DataFrame 和僅具有name列的派生 DataFrame 之間使用join 。 由於多行可以滿足連接條件，因此我們必須在連接后按原始列進行分組。

這是您輸入的詳細示例：

data = [
    ("HELLOWORLD2019THISISGOOGLE", "WORLD2019", "WORLD_2019"),
    ("NATUREISVERYGOODFOROURHEALTH", None, None),
    ("THESUNCONTAINVITAMIND", "VITAMIND", "VITAMIN_D"),
    ("BECARETOOURHEALTHISVITAMIND", "OURHEALTH", "OUR_ / HEALTH")
]
df = spark.createDataFrame(data, ["text", "name", "original_name"])

# create new DF with search words
# as it's the originl_name which interests us for the final list so we select it too
search_df = df.select(struct(col("name"), col("original_name")).alias("search_match"))

# join on df.text contains search_df.name
df_join = df.join(search_df, df.text.contains(search_df["search_match.name"]), "left")

# group by original columns and collect matches in a list
df_join.groupBy("text", "name", "original_name")\
    .agg(collect_list(col("search_match.original_name")).alias("new_column"))\
    .show(truncate=False)

輸出：

+----------------------------+---------+-------------+--------------------------+
|text                        |name     |original_name|new_column                |
+----------------------------+---------+-------------+--------------------------+
|HELLOWORLD2019THISISGOOGLE  |WORLD2019|WORLD_2019   |[WORLD_2019]              |
|THESUNCONTAINVITAMIND       |VITAMIND |VITAMIN_D    |[VITAMIN_D]               |
|NATUREISVERYGOODFOROURHEALTH|null     |null         |[OUR_ / HEALTH]           |
|BECARETOOURHEALTHISVITAMIND |OURHEALTH|OUR_ / HEALTH|[VITAMIN_D, OUR_ / HEALTH]|
+----------------------------+---------+-------------+--------------------------+