使用scikit-learn进行电影评论情感分类
数据准备
从网站Movie Review Data下载语料。这里选择polarity dataset v2.0。该数据集包含正负情感极性（pos和neg）的电影评论各1000条。
下面，进行数据载入，并进行训练集/测试集划分。
# load library
import os
import sys
# set working directory
os.chdir("D:\\my_python_workfile\\Thesis\\movie_review\\review_polarity\\txt_sentoken")
dataset_dir_name = os.getcwd()
dataset_dir_name
'D:\\my_python_workfile\\Thesis\\movie_review\\review_polarity\\txt_sentoken'
# load library
import numpy as np
from sklearn.datasets import load_files
from sklearn.cross_validation import train_test_split
from sklearn.feature_extraction.text import CountVectorizer,TfidfVectorizer
from sklearn.feature_extraction.text import TfidfTransformer
from sklearn.pipeline import Pipeline
from sklearn.grid_search import GridSearchCV
from sklearn.svm import LinearSVC
from sklearn.naive_bayes import MultinomialNB
from sklearn import metrics
# load data,and split into training/test set
movie_reviews = load_files(dataset_dir_name)
# split
doc_terms_train,doc_terms_test,doc_class_train,doc_class_test = train_test_split(
movie_reviews.data,movie_reviews.target,test_size = 0.2,random_state = None)
len(doc_class_train),len(doc_class_test),(movie_reviews.target_names)
#print("\n".join(movie_reviews.data[0].split("\n"))[:20])
(1600, 400, ['neg', 'pos'])
建立vectorizer/classifier pipeline
# build a vectorizer/classifier pipeline
pipeline = Pipeline([
('vect', TfidfVectorizer(min_df=3, max_df=0.95)),
('clf', LinearSVC(C=1000)),
])
参数寻优
# grid search
parameters = {
'vect__ngram_range': [(1, 1), (1, 2)],
}
grid_search = GridSearchCV(pipeline, parameters, n_jobs=-1)
grid_search.fit(doc_terms_train,doc_class_train)
print(grid_search.grid_scores_)
[mean: 0.83750, std: 0.01659, params: {'vect__ngram_range': (1, 1)}, mean: 0.85938, std: 0.01338, params: {'vect__ngram_range': (1, 2)}]
模型预测效果评估
# y_predicted
y_predicted = grid_search.predict(doc_terms_test)
# report
print(metrics.classification_report(doc_class_test,y_predicted,
target_names = movie_reviews.target_names))
precision recall f1-score support
neg 0.85 0.85 0.85 188
pos 0.87 0.86 0.87 212
avg / total 0.86 0.86 0.86 400
# confusion matrix
confusion_matrix = metrics.confusion_matrix(doc_class_test,y_predicted)
print(confusion_matrix)
[[160 28]
[ 29 183]]
深入学习
以上作为一个入门级的介绍，就到此为止啦～
当然，在现实生活中，我们不能仅仅满足于对电影评论的正负面分类，而应该考虑更细粒度的分类问题。比如电影评论文本分为1~5星，1星和2星之间比1星和5星更为相似，所以这种多分类问题可以看做是ordinal regression问题求解(见参考文献Pang B等)。
正好kaggle上有一个更细粒度的情感分类问题:Sentiment Analysis on Movie Reviews。对情感分析感兴趣的同学，可以捋起袖子，来一场Kaggle的比赛了。
参考文献
- scikit-learn exercise_02_sentiment.py
- Pang B, Lee L. Seeing stars: Exploiting class relationships for sentiment categorization with respect to rating scales[C]//Proceedings of the 43rd Annual Meeting on Association for Computational Linguistics. Association for Computational Linguistics, 2005: 115-124.