1.查看原始数据结构

关注GZH：阿旭算法与机器学习，回复：“ML33”即可获取本文数据集、源码与项目文档

数据集共有4个文件：
stopwords.txt为停用词文件；
train.negative.txt为训练用负面数据文件；
train.positive.txt为训练用正面数据文件；
test.combined.txxt为测试用数据文件。
在这里插入图片描述
文件内容如下：

2.导入数据并进行数据处理

from matplotlib import pyplot as plt
import jieba # 分词
import re # 正则
from sklearn.feature_extraction.text import TfidfVectorizer
import numpy as np

2.1 提取数据与标签

def read_data(path, is_pos=None):
    """
    给定文件的路径，读取文件
    path: path to the data
    is_pos: 是否数据是postive samples. 
    return: (list of review texts, list of labels) 
    """
    reviews, labels  = [], []
    with open(path, 'r',encoding='utf-8') as file:
        review_start  = False
        review_text = []
        for line in file:
            line = line.strip()
            if not line: continue
            if not review_start and line.startswith("<review"):
                review_start = True
                if "label" in line:
                    labels.append(int(line.split('"')[-2]))
                continue                
            if review_start and line == "</review>":
                review_start = False
                reviews.append(" ".join(review_text))
                review_text = []
                continue
            if review_start:
                review_text.append(line)
    if is_pos:
        labels = [1]*len(reviews)
    elif not is_pos is None:
        labels = [0]*len(reviews)
    return reviews, labels


def process_file():
    """
    读取训练数据和测试数据，并对它们做一些预处理
    """    
    train_pos_file = "data_sentiment/train.positive.txt"
    train_neg_file = "data_sentiment/train.negative.txt"
    test_comb_file = "data_sentiment/test.combined.txt"
    
    # 读取文件部分，把具体的内容写入到变量里面
    train_pos_cmts, train_pos_lbs = read_data(train_pos_file, True)
    train_neg_cmts, train_neg_lbs = read_data(train_neg_file, False)
    train_comments = train_pos_cmts + train_neg_cmts
    train_labels = train_pos_lbs + train_neg_lbs
    test_comments, test_labels = read_data(test_comb_file)
    return train_comments, train_labels, test_comments, test_labels
train_comments, train_labels, test_comments, test_labels = process_file()

train_comments[:5]

['发短信特别不方便！背后的屏幕很大用起来不舒服，是手触屏的！切换屏幕很麻烦！',
 '手感超好，而且黑色相比白色在转得时候不容易眼花，找童年的记忆啦。',
 '！！！！！',
 '先付款的   有信用',
 '价格 质量 售后 都很满意']

# 训练数据和测试数据大小
print (len(train_comments), len(test_comments))

print (train_comments[1], train_labels[1])

8064 2500
手感超好，而且黑色相比白色在转得时候不容易眼花，找童年的记忆啦。 1

2.2 过滤停用词

def load_stopwords(path):
    """
    从外部文件中导入停用词
    """
    stopwords = set()
    with open(path, 'r',encoding='utf-8') as in_file:
        for line in in_file:
            stopwords.add(line.strip())
    return stopwords


def clean_non_chinese_symbols(text):
    """
    处理非中文字符
    """
    text = re.sub('[!！]+', "!", text)
    text = re.sub('[?？]+', "?", text)
    text = re.sub("[a-zA-Z#$%&\'()*+,-./:;：<=>@，。★、…【】《》“”‘’[\\]^_`{|}~]+", " UNK ", text)
    return re.sub("\s+", " ", text)  

def clean_numbers(text):
    """
    处理数字符号  128  190  NUM 
    """
    return re.sub("\d+", ' NUM ', text)

def preprocess_text(text, stopwords):
    """
    文本的预处理过程
    """
    text = clean_non_chinese_symbols(text)
    text = clean_numbers(text)
    text = " ".join([term for term in jieba.cut(text) if term and not term in stopwords])
    return text

path_stopwords = "./data_sentiment/stopwords.txt"
stopwords = load_stopwords(path_stopwords)

# 对于train_comments, test_comments进行字符串的处理，几个考虑的点：
#   1. 停用词过滤
#   2. 去掉特殊符号
#   3. 去掉数字（比如价格..)
#   4. ...
#   需要注意的点是，由于评论数据本身很短，如果去掉的太多，很可能字符串长度变成0
#   预处理部部分，可以自行选择合适的方案，只要注释就可以。

train_comments_new = [preprocess_text(comment, stopwords) for comment in train_comments]
test_comments_new = [preprocess_text(comment, stopwords) for comment in test_comments]

print (train_comments_new[0], test_comments_new[0])

发短信 特别 不 方便 ! 背后 屏幕 很大 起来 不 舒服   UNK   手触 屏 ! 切换 屏幕 很 麻烦 ! 终于 找到 同道中人 初中   UNK   已经 喜欢 上   UNK   同学 都 鄙夷 眼光 看   UNK   人为   UNK   样子 古怪 说 ＂ 丑 ＂ 当场 气晕 现在 同道中人   UNK   好开心 !   UNK   !   UNK

2.3 TfidfVectorizer将文本向量化

#   利用tf-idf从文本中提取特征,写到数组里面. 
#   参考：https://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.TfidfVectorizer.html
tfidf = TfidfVectorizer()
X_train =  tfidf.fit_transform(train_comments_new) # 训练数据的特征
y_train =  train_labels # 训练数据的label
X_test = tfidf.transform(test_comments_new) # 测试数据的特征
y_test = test_labels# 测试数据的label

print (np.shape(X_train), np.shape(X_test), np.shape(y_train), np.shape(y_test))

(8064, 23101) (2500, 23101) (8064,) (2500,)

3.利用不同模型进行训练与评估

3.1 朴素贝叶斯模型

from sklearn.naive_bayes import MultinomialNB
from sklearn.metrics import accuracy_score

clf = MultinomialNB()
# 利用朴素贝叶斯做训练
clf.fit(X_train, y_train)
y_pred = clf.predict(X_test)
print("accuracy on test data: ", accuracy_score(y_test, y_pred))

accuracy on test data:  0.6368

3.2 k近邻模型

from sklearn.neighbors import KNeighborsClassifier
clf = KNeighborsClassifier(n_neighbors=1)
clf.fit(X_train, y_train)
y_pred = clf.predict(X_test)
print("accuracy on test data: ", accuracy_score(y_test, y_pred))

accuracy on test data:  0.524

3.3 逻辑回归模型

from sklearn.linear_model import LogisticRegression

clf = LogisticRegression(solver='liblinear')
clf.fit(X_train, y_train)
y_pred = clf.predict(X_test)
print("accuracy on test data: ", accuracy_score(y_test, y_pred))