使用 Hugging Face pipeline 做情緒分析

Python 的 Natural Language Processing（NLP）

Fouad Trad

Machine Learning Engineer

複習：NLP 工作流程

完整流程圖，指出第 1 章涵蓋前處理、第 2 章涵蓋特徵擷取。

Hugging Face pipelines

完整流程圖，指出第 3、4 章將介紹 hugging face pipelines，包起前處理、特徵擷取與建模。

現成流程，一次函式呼叫處理所有步驟
定義 pipeline 需指定：
- NLP 任務
- 執行任務的模型

情緒分析的 pipeline

文字分類任務
預測文字表達正向或負向情緒

圖示：正向情緒為開心臉與讚手勢，負向情緒為難過臉與倒讚手勢。

文字分類的模型

示範在網站中捲動並選對任務以取得合適模型的動圖。

¹ https://huggingface.co/models

在程式碼中使用 pipelines

from transformers import pipeline


classification_pipeline = pipeline(

  task="sentiment-analysis", # or text-classification

  model="distilbert/distilbert-base-uncased-finetuned-sst-2-english" 
)

result = classification_pipeline("I really liked the movie!!")

print(result)

[{'label': 'POSITIVE', 'score': 0.9998093247413635}]

批次文字的情緒分析

texts = ["I really liked the movie!!",
         "Great job ruining my day.",
         "This product exceeded my expectations.",
         "Wow, just what I needed... another problem.", 
         "Absolutely fantastic experience!"]

results = classification_pipeline(texts)

print(results)

[{'label': 'POSITIVE', 'score': 0.9998093247413635}, 
 {'label': 'NEGATIVE', 'score': 0.8666700124740601}, 
 {'label': 'POSITIVE', 'score': 0.998874843120575}, 
 {'label': 'POSITIVE', 'score': 0.98626708984375}, 
 {'label': 'POSITIVE', 'score': 0.9998812675476074}]

評估情緒分析模型

texts = ["I really liked the movie!!",
         "Great job ruining my day.",
         "This product exceeded my expectations.",
         "Wow, just what I needed... another problem.", 
         "Absolutely fantastic experience!"]

true_labels = ["POSITIVE", "NEGATIVE", "POSITIVE", "NEGATIVE", "POSITIVE"]

results = classification_pipeline(texts)

predicted_labels = [result['label'] for result in results]

from sklearn.metrics import accuracy_score
accuracy = accuracy_score(true_labels, predicted_labels)

print(f"Accuracy: {accuracy}")

Accuracy: 0.80

一起來練習吧！

Python 的 Natural Language Processing（NLP）