在 python 中应用预训练 facebook/bart-large-cnn 进行文本摘要

Question

我正在使用 huggingface 变形金刚，并且对它有了一些了解。我正在使用 facebook/bart-large-cnn 模型为我的项目执行文本摘要，目前我正在使用以下代码进行一些测试：

text = """
Justin Timberlake and Jessica Biel, welcome to parenthood. 
The celebrity couple announced the arrival of their son, Silas Randall Timberlake, in 
statements to People."""

from transformers import pipeline
smr_bart = pipeline(task="summarization", model="facebook/bart-large-cnn")
smbart = smr_bart(text, max_length=150)
print(smbart[0]['summary_text'])

一小段代码实际上给了我一个很好的文本总结。但我的问题是如何在我的数据框列之上应用相同的预训练模型。我的数据框如下所示：

ID        Lang          Text
1         EN            some long text here...
2         EN            some long text here...
3         EN            some long text here...

..... 50K 行依此类推

现在我想将预训练模型应用于 col Text 以从中生成一个新列 df['summary']，结果数据框应如下所示：

ID        Lang         Text                              Summary
1         EN            some long text here...           Text summary goes here...
2         EN            some long text here...           Text summary goes here...
3         EN            some long text here...           Text summary goes here...

我怎样才能做到这一点？任何帮助将不胜感激。

Answer 1

你总是可以做的是利用数据框 apply 函数：

df = pd.DataFrame([('EN',text)]*10, columns=['Lang','Text'])

df['summary'] = df.apply(lambda x: smr_bart(x['Text'], max_length=150)[0]['summary_text'] , axis=1)

df.head(3)

输出：

    Lang    Text                                                summary
0   EN      \nJustin Timberlake and Jessica Biel, welcome ...   The celebrity couple announced the arrival of ...
1   EN      \nJustin Timberlake and Jessica Biel, welcome ...   The celebrity couple announced the arrival of ...
2   EN      \nJustin Timberlake and Jessica Biel, welcome ...   The celebrity couple announced the arrival of ...

这有点低效，因为管道将被每一行调用（执行时间 2 分 16 秒）。因此我建议将 Text 列转换为列表并将其直接传递给管道（执行时间 41 秒）：

df = pd.DataFrame([('EN',text)]*10, columns=['Lang','Text'])

df['summary'] = [x['summary_text'] for x in smr_bart(df['Text'].tolist(), max_length=150)]

df.head(3)

输出：

    Lang    Text                                                summary
0   EN      \nJustin Timberlake and Jessica Biel, welcome ...   The celebrity couple announced the arrival of ...
1   EN      \nJustin Timberlake and Jessica Biel, welcome ...   The celebrity couple announced the arrival of ...
2   EN      \nJustin Timberlake and Jessica Biel, welcome ...   The celebrity couple announced the arrival of ...

在 python 中应用预训练 facebook/bart-large-cnn 进行文本摘要

Applying pre trained facebook/bart-large-cnn for text summarization in python

nlp

summarization

python-3.x

huggingface-transformers

huggingface-tokenizers