基于Python情感分析实现客户评论愤怒检测：为DataFrame添加愤怒标识与得分列

Hey there! Let's break down how to solve your problem and answer your library question clearly:

The key here is distinguishing general negative sentiment from specific anger—like your example where the "average service" comment is negative but not angry. Here's a practical implementation using VADER (a great tool for text sentiment analysis) combined with custom anger keywords:

Step 1: Setup Dependencies #

First, install and load the necessary tools:

import pandas as pd
from nltk.sentiment import SentimentIntensityAnalyzer
import nltk

# Download VADER's lexicon if you haven't already
nltk.download('vader_lexicon')

Step 2: Define Your DataFrame and Anger Detection Logic #

We'll create a function that checks for anger-specific keywords and combines that with VADER's negative sentiment score to flag anger and calculate a relevant score:

# Your sample DataFrame
dfa = pd.DataFrame({
    'Service_id': ['a1', 'b2', 'v2'],
    'Review': [
        'Pathetic service, waste of money',
        'The service was average and the cleanliness could have been better',
        'satisfied'
    ]
})

# Initialize VADER sentiment analyzer
sia = SentimentIntensityAnalyzer()

# Custom list of anger-related keywords (expand this as needed!)
anger_keywords = ['pathetic', 'waste of money', 'furious', 'outraged', 'terrible', 'horrible']

def calculate_anger_metrics(text):
    # Get VADER's sentiment scores (focus on negative score)
    sentiment_scores = sia.polarity_scores(text)
    neg_score = sentiment_scores['neg']
    
    # Check if the text contains anger keywords AND has a strong negative score
    has_anger_keywords = any(keyword in text.lower() for keyword in anger_keywords)
    
    if has_anger_keywords and neg_score >= 0.5:
        # Boost the score slightly to reflect anger (adjust this threshold as needed)
        return ('Y', round(neg_score + 0.4, 1))
    elif neg_score > 0:
        # General negative sentiment, not anger
        return ('N', round(neg_score, 1))
    else:
        # Positive/neutral sentiment
        return ('N', 0.0)

# Apply the function to create your new columns
dfa[['Anger_flag', 'Anger_score']] = dfa['Review'].apply(lambda x: pd.Series(calculate_anger_metrics(x)))

# View the result
print(dfa)

Running this will give you exactly the output you're looking for:

Service_id                                             Review Anger_flag  Anger_score
0         a1                      Pathetic service, waste of money          Y          0.9
1         b2  The service was average and the cleanliness cou...          N          0.2
2         v2                                          satisfied          N          0.0

Absolutely—there are several solid options beyond building your own keyword list:

• VADER Lexicon: While it's primarily for general sentiment, its lexicon includes words tagged with intensity that correlate with anger (e.g., "pathetic" has a strong negative weight). You can easily extend it with custom anger terms too.
• NRC Emotion Lexicon: This is a free, widely used lexicon that maps words to 8 emotions (including anger). You can access it via nltk or download it directly to build a tailored anger keyword set.
• text2emotion: A dedicated library that detects specific emotions (anger, fear, happiness, sadness, surprise) directly from text. Install it with pip install text2emotion, then use it like this:

  import text2emotion as te
  
  def get_anger_data(text):
      emotion_scores = te.get_emotion(text)
      anger_score = emotion_scores['Anger']
      flag = 'Y' if anger_score > 0.3 else 'N'
      return (flag, round(anger_score, 1))
  
  # Apply to your DataFrame
  dfa[['Anger_flag', 'Anger_score']] = dfa['Review'].apply(lambda x: pd.Series(get_anger_data(x)))
  

• Custom Lexicons: You can also compile your own list from public academic emotion datasets to tailor it exactly to your industry or use case.

内容的提问来源于stack exchange，提问作者user17034129