감성 분석 - 비지도 학습 | Sentiment Analysis - Unsupervised Learning (24.02.2022.)

다음은 비지도 학습을 통한 감성 분석 시 사용한 코드이다.

The following are the steps and codes for sentiment analysis using the unsupervised learning method.

---

1. Import necessary packages.

import pandas as pd

import matplotlib.pyplot as plt

import re

import nltk
from nltk.sentiment.vader import SentimentIntensityAnalyzer

from sklearn.metrics import accuracy_score, confusion_matrix, precision_score
from sklearn.metrics import recall_score, f1_score

 

2. Define a preprocessing function.

def data_text_cleaning(data):
 
    # 영문자 이외 문자는 공백으로 변환
    only_english = re.sub('[^a-zA-Z]', ' ', data)
 
    # 소문자 변환
    no_capitals = only_english.lower()
 
    return no_capitals

 

3. Import datasets.

suc = pd.read_csv('suc.csv')
un = pd.read_csv('un.csv')

 

4. Preprocessing

# successful product reviews
for i in range(len(suc)):
    
    review = str(suc['review'][i])
    suc['review'][i] = data_text_cleaning(review)
    
    i += 1
    
# unsuccessful product reviews
for i in range(len(un)):
    
    u_review = str(un['review'][i])
    un['review'][i] = data_text_cleaning(u_review)
    
    i += 1

 

5. Define a function to get predicted sentiment labels.

# 임계치설정(보통 0.1)을 통해 compound(총 감성지수)가 임계치값보다 높으면 긍정(1), 낮으면 부정(-1)으로 분석
def get_sentiment(review):
    
    analyzer = SentimentIntensityAnalyzer()
    scores = analyzer.polarity_scores(review)
    
    compound_score = scores['compound']
    final_sentiment = 1 if compound_score > 0.1 else 0 if compound_score <= 0.1 and compound_score >= -0.1 else -1
    # 1 긍정, 0 중립, -1 부정
    
    return final_sentiment

 

6. Create a column for the predicted sentiment labels(psentl).

# successful product reviews
suc['psentl'] = 0

for i in range(len(suc)):
    
    suc['psentl'][i] = get_sentiment(str(suc['review'][i]))
    
    i += 1
    
# unsuccessful product reviews
un['psentl'] = 0

for i in range(len(un)):
    
    un['psentl'][i] = get_sentiment(str(un['review'][i]))
    
    i += 1

 

7. Create a column for the real sentiment labels(sent).

# successful product reviews
suc['sent'] = 1

# 평점을 기준으로 한 실제 감성 (4,5점 긍정(1), 3점 중립(0) 1,2점 부정(-1))
for i in range(len(suc)):
    
    if suc['rating'][i] >= 4:
        suc['sent'][i] = 1
        
    else:
        if suc['rating'][i] == 3:
            suc['sent'][i] = 0
        
        else:
            suc['sent'][i] = -1
    
    i += 1
    
# unsuccessful product reviews
un['sent'] = 1

for i in range(len(un)):
    
    if un['rating'][i] >= 4:
        un['sent'][i] = 1
        
    else:
        if un['rating'][i] == 3:
            un['sent'][i] = 0
        
        else:
            un['sent'][i] = -1
    
    i += 1

 

8. Compare the predicted and real sentiment labels and compute accuracy, precision, recall and F1 scores as well as a confusion matrix.

# successful product reviews
# 원본 데이터에서 주어진 정답 label과 VADER로 예측한 label 비교
s_target = suc['sent']
s_pred = suc['psentl']

print(confusion_matrix(s_target, s_pred))
print("정확도 :", accuracy_score(s_target, s_pred))
print("정밀도 :", precision_score(s_target, s_pred, average='weighted'))
print("재현율 :", recall_score(s_target, s_pred, average='weighted'))
print("F1 score :", f1_score(s_target, s_pred, average='weighted')) # micro, macro, weighted, samples

---

• Average parameter (default: binary)
  • micro: compute a score globally with total true positives, false negatives and false positives
  • macro: compute a score for each label (unweighted mean)
  • weighted: compute a score for each label (weighted mean)
  • samples: compute a score for each instance

---

# unsuccessful product reviews
u_target = un['sent']
u_pred = un['psentl']

print(confusion_matrix(u_target, u_pred))
print("정확도 :", accuracy_score(u_target, u_pred))
print("정밀도 :", precision_score(u_target, u_pred, average='weighted'))
print("재현율 :", recall_score(u_target, u_pred, average='weighted'))
print("F1 score :", f1_score(u_target, u_pred, average='weighted')) # micro, macro, weighted, samples

 

9. Visualization

# successful product reviews
# 긍부정 시각화 (예측값) --> only with review text
reaction = ['positive', 'neutral', 'negative']
sp_pos = list(suc['psentl']).count(1) / len(suc)
sp_neu = list(suc['psentl']).count(0) / len(suc)
sp_neg = 1 - sp_pos - sp_neu
values = [sp_pos, sp_neu, sp_neg]

plt.bar(reaction, values, color=['b', 'g', 'r'])
plt.title('Predicted Reactions for Successful Products', fontsize=15)

# 긍부정 시각화 (실제값) --> only with review score (별점)
reaction = ['positive', 'neutral', 'negative']
sr_pos = list(suc['sent']).count(1) / len(suc)
sr_neu = list(suc['sent']).count(0) / len(suc)
sr_neg = 1 - sr_pos - sr_neu
values = [sr_pos, sr_neu, sr_neg]

plt.bar(reaction, values, color=['b', 'g', 'r'])
plt.title('Real Reactions for Successful Products', fontsize=15)

# unsuccessful product reviews
# 긍부정 시각화 (예측값) --> only with review text
reaction = ['positive', 'neutral', 'negative']
up_pos = list(un['psentl']).count(1) / len(un)
up_neu = list(un['psentl']).count(0) / len(un)
up_neg = 1 - up_pos - up_neu
values = [up_pos, up_neu, up_neg]

plt.bar(reaction, values, color=['b', 'g', 'r'])
plt.title('Predicted Reactions for Unsuccessful Products', fontsize=15)

# 긍부정 시각화 (예측값) --> only with review score (별점)
reaction = ['positive', 'neutral', 'negative']
ur_pos = list(un['sent']).count(1) / len(un)
ur_neu = list(un['sent']).count(0) / len(un)
ur_neg = 1 - ur_pos - ur_neu
values = [ur_pos, ur_neu, ur_neg]

plt.bar(reaction, values, color=['b', 'g', 'r'])
plt.title('Real Reactions for Unsuccessful Products', fontsize=15)

 

 

 

* Unauthorized copying and distribution of this post are not allowed.

* 해당 글에 대한 무단 배포 및 복사를 허용하지 않습니다.