文字摘要

善用 Hugging Face

Jacob H. Marquez

Lead Data Engineer

什麼是摘要？

Large text

抽取式 vs. 生成式

抽取式：

$$ ✅ 從文本選出關鍵句子

$$ ✅ 高效、省資源

$$ ❌ 彈性不足，可能不夠連貫

生成式：

$$ ✅ 產生重述後的新文本

$$ ✅ 更清楚、可讀性高

$$ ❌ 需更多資源與處理

抽取式摘要的應用情境

📑 法律文件：標註關鍵條款

💰 財務研究：擷取洞見

Legal Documents

Financial Research

生成式摘要的應用情境

News Articles

Content Recommendations.jpg

📰 新聞文章：產生精簡摘要

$$ $$

📍 內容推薦：生成吸睛敘述

抽取式摘要實作

from transformers import pipeline

# Load the extractive summarization pipeline
summarizer = pipeline("summarization", model="nyamuda/extractive-summarization")

text = "This is my really large text about Data Science..."
summary_text = summarizer(text)

print(summary_text[0]['summary_text'])

"data science is a field that combines mathematics, statistics...."

生成式摘要實作

from transformers import pipeline

# Load the abstractive summarization pipeline
summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6")


text = "This is my really large text about Data Science..."
summary_text = summarizer(text)
print(summary_text[0]['summary_text'])

"The global data science platform market is projected
 is projected to reach $140.9 billion by 2025..."

摘要的參數

✅ min_new_tokens 與 max_new_tokens：控制摘要長度

summarizer = pipeline(task="summarization", min_new_tokens=10, max_new_tokens=150)

一起來練習吧！

善用 Hugging Face