| EVELYN DU

Home > Project > Social Science > Machine Learning> Text Analysis

Here I will start from Paper Text Analysis to the Audio Text Analysis.

1. Paper Text Analysis

1.1 NLP Sentiment Analysis

In this research group, the main tool used is the SnowNLP library in NLP.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
from selenium import webdriver
from lxml import etree
import time
import requests
import csv

def getData(key):
    options = webdriver.ChromeOptions()
    options.add_argument('--no-sandbox')
    # options.add_argument('--headless')  # 无头参数
    options.add_argument('--disable-gpu')

    prefs = {
        'profile.default_content_setting_values':
            {
                'notifications': 2
            }
    }
    options.add_experimental_option('prefs', prefs)

    options.add_experimental_option('useAutomationExtension', False)
    options.add_experimental_option('excludeSwitches', ['enable-automation'])
    driver = webdriver.Chrome(chrome_options=options)
    driver.maximize_window()

    url=f'https://twitter.com/{key}'

    driver.get(url)

    time.sleep(3)

    js="var q=document.documentElement.scrollTop=0"
    driver.execute_script(js)
    time.sleep(3)

    def save(data):
        file=open('dickc_year_Time.csv','a+',newline='')
        writer=csv.writer(file)
        writer.writerow(data)
        file.close()
    T_div=driver.find_elements_by_xpath('//*[@id="react-root"]/div/div/div[2]/main/div/div/div/div/div/div[2]/div/div/div[2]/section/div/div/div')
    print(len(T_div))
    T_time_A='1'
    k=1
    y_2021=0
    y_2020=0
    y_2019=0
    y_2018=0
    y_2017=0
    for j in range(1, 100):
        for i in range(1, 100):
            try:
                T_time=driver.find_element_by_xpath(f'//*[@id="react-root"]/div/div/div[2]/main//div[2]/div/div/div[2]/section/div/div/div[{i}]/div/div/article/div//a/time').text
                T_time_B = driver.find_element_by_xpath('//*[@id="react-root"]/div/div/div[2]/main//div[2]/div/div/div[2]/section/div/div/div[2]/div/div/article/div//a/time').text

            except:
                js='window.scrollTo(0,document.body.scrollHeight)'
                driver.execute_script(js)
                driver.implicitly_wait(10)
                time.sleep(1)
                print('正在下滑')
                try:
                    T_time_A = driver.find_element_by_xpath(
                        '//*[@id="react-root"]/div/div/div[2]/main//div[2]/div/div/div[2]/section/div/div/div[2]/div/div/article/div//a/time').text
                except:
                    pass
                if (T_time_A == T_time_B):
                    k += 1
                k_num=0
                break
            print(str(T_time)+'*******\n'+str(i))
            if('2016' in str(T_time)):
                save([key,y_2021,y_2020,y_2019,y_2018,y_2017])
                driver.close()
                exit()
            elif('2020' in str(T_time)):
                y_2020+=1
            elif('2019' in str(T_time)):
                y_2019+=1
            elif('2018' in str(T_time)):
                y_2018+=1
            elif('2017' in str(T_time)):
                y_2017+=1
            else:
                y_2021 += 1
            if (k>3):
                save([key,y_2021,y_2020,y_2019,y_2018,y_2017])
                driver.close()
                return
            print(y_2021,y_2020,y_2019,y_2018,y_2017)
            # save([y_2021, y_2020, y_2019, y_2018, y_2017])
    driver.close()

if __name__ == '__main__':
    keys='tonyhawk'
    for key in keys.split():
        print(f'正在采集--{key}')
        getData(key)
        print(f'{key}----------已完成！')

1.2 LSTM Sentiment Analysis

Word Vector Model;one-hot；Word2Vec;Tokenizer;RNN

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
n1 = 0
n2 = 0
n3 = 0
n4 = 0
newLines = ""
#导入数据，对数据打标签
with open("train.txt", "r", encoding="UTF-8",errors='ignore') as f:
    lines = f.readlines()
    f.close()
    for line in lines:
        label,sentence =line.strip().split('\t')
        if int(label) == 0:
            if n1 < 10000:
                n1 += 1
                newLines += label + "\t" + sentence + "\n"
        if int(label) == 1:
            if n2 < 10000:
                n2 += 1
                newLines += label + "\t" + sentence + "\n"
        if int(label) == 2:
            if n3 < 10000:
                n3 += 1
                newLines += label + "\t" + sentence + "\n"
        if int(label) == 3:
            if n4 < 10000:
                n4 += 1
                newLines += label + "\t" + sentence + "\n"
    with open("small_train.txt", "w") as f2:
        f2.write(newLines)
        f2.close()

# 数据过滤
def regex_filter(s_line):
    # 剔除英文、数字，以及空格
    special_regex = re.compile(r"[a-zA-Z0-9\s]+")
    # 剔除英文标点符号和特殊符号
    en_regex = re.compile(r"[.…{|}#$%&\'()*+,!-_./:~^;<=>?@★●，。]+")
    # 剔除中文标点符号
    zn_regex = re.compile(r"[《》！、，“”；：（）【】]+")

    s_line = special_regex.sub(r"", s_line)
    s_line = en_regex.sub(r"", s_line)
    s_line = zn_regex.sub(r"", s_line)
    return s_line

## 加载停用词
def stopwords_list(file_path):
    stopwords = [line.strip() for line in open(file_path, 'r', encoding='utf-8').readlines()]
    return stopwords

word_freqs =Counter()  # 词频
stopword = stopwords_list("停用词.txt")
max_len = 0
with open('train.txt', 'r+', encoding="UTF-8",errors='ignore') as f:
    lines = f.readlines()
    for line in lines:
#         # 取出 label 和句子
        label,sentence =line.strip().split('\t')
        # 数据预处理
        sentence = regex_filter(sentence)
        words = jieba.cut(sentence)
        x = 0
        for word in words:
            # 去除停用词
            if word not in stopword:
                print(word)
                word_freqs[word] += 1
                x += 1
        max_len = max(max_len, x)
print(max_len)
print('nb_words ', len(word_freqs))

MAX_FEATURES = 80000 # 最大词频数
vocab_size = min(MAX_FEATURES, len(word_freqs)) + 2
# 构建词频字典
word2index = {x[0]: i+2 for i, x in enumerate(word_freqs.most_common(MAX_FEATURES))}
word2index["PAD"] = 0
word2index["UNK"] = 1
# 将词频字典写入文件中保存
with open('word_dict.pickle', 'wb') as handle:
    pickle.dump(word2index, handle, protocol=pickle.HIGHEST_PROTOCOL)

# 加载分词字典

with open('word_dict.pickle', 'rb') as handle:
    word2index = pickle.load(handle)

### 准备数据
MAX_FEATURES = 80002 # 最大词频数
MAX_SENTENCE_LENGTH = 110 # 句子最大长度
num_recs = 0  # 样本数

with open('train.txt', 'r+', encoding="UTF-8",errors='ignore') as f:
    lines = f.readlines()
    # 统计样本大小
    for line in lines:
        num_recs += 1

# 初始化句子数组和 label 数组
X = np.empty(num_recs,dtype=list)
y = np.zeros(num_recs)
i=0

with open('train.txt', 'r+', encoding="UTF-8",errors='ignore') as f:
    for line in f:
        label,sentence =line.strip().split('\t')
        words = jieba.cut(sentence)
        seqs = []
        for word in words:
            # 在词频中
            if word in word2index:
                seqs.append(word2index[word])
            else:
                seqs.append(word2index["UNK"]) # 不在词频内的补为 UNK
        X[i] = seqs
        y[i] = int(label)
        i += 1

# 把句子转换成数字序列，并对句子进行统一长度，长的截断，短的补 0
X = sequence.pad_sequences(X, maxlen=MAX_SENTENCE_LENGTH)
# 使用 pandas 对 label 进行 one-hot 编码
y1 = pd.get_dummies(y).values
print(X.shape)
print(y1.shape)

# 数据划分
Xtrain, Xtest, ytrain, ytest = train_test_split(X, y1, test_size=0.1, random_state=42)

## 网络构建
EMBEDDING_SIZE = 256 # 词向量维度
HIDDEN_LAYER_SIZE = 128 # 隐藏层大小
BATCH_SIZE = 64 # 每批大小
NUM_EPOCHS = 5 # 训练周期数
# 创建一个实例
model = Sequential()
# 构建词向量
model.add(Embedding(MAX_FEATURES, EMBEDDING_SIZE,input_length=MAX_SENTENCE_LENGTH))
model.add(SpatialDropout1D(0.2))
# 构建 LSTM 层
model.add(LSTM(HIDDEN_LAYER_SIZE, dropout=0.2, recurrent_dropout=0.2))
# 输出层包含四个分类，激活函数设置为'softmax'
model.add(Dense(4, activation="softmax"))
model.add(Activation('softmax'))
# 损失函数设置为分类交叉熵 categorical_crossentropy
model.compile(loss="SparseCategoricalCrossentropy", optimizer="adam",metrics=["accuracy"])

## 训练模型
model.fit(Xtrain, ytrain, batch_size=BATCH_SIZE, epochs=NUM_EPOCHS,validation_data=(Xtest, ytest))

## 评估模型
y_pred = model.predict(Xtest)
y_pred = y_pred.argmax(axis=1)
ytest = ytest.argmax(axis=1)

print("保存模型")
model.save('my_model.h5')

## 测试模型
print("加载模型")
model = load_model('my_model.h5')

INPUT_SENTENCES = ['一根大阳线','财务报表','数据好看','神经病公司']
XX = np.empty(len(INPUT_SENTENCES),dtype=list)
i=0
for sentence in  INPUT_SENTENCES:
    words = jieba.cut(sentence)
    seq = []
    for word in words:
        if word in word2index:
            seq.append(word2index[word])
        else:
            seq.append(word2index['UNK'])
    XX[i] = seq
    i+=1

XX = sequence.pad_sequences(XX, maxlen=MAX_SENTENCE_LENGTH)
label2word = {0:'喜悦', 1:'愤怒', 2:'厌恶', 3:'低落'}
for x in model.predict(XX):
    print(x)
    x = x.tolist()
    label = x.index(max(x[0], x[1], x[2], x[3]))
    print(label)
    print('{}'.format(label2word[label]))

2. Audio Text Analysis

Voice Reporting Preprocessing

1.Frame Segmentation and Windowing

1
2
3
4
5
6
7
# Frame segmentation function
def enframe(wavData, frameSize, overlap):
    coeff = 0.97  # Pre-emphasis coefficient
    wlen = len(wavData)
    step = frameSize - overlap
    frameNum: int = math.ceil(wlen / step)
    frameData = np.zeros((frameSize, frameNum))

2.Pre-Highlit

1
2
3
4
5
6
7
8
9
10
def ZCR(frameData):
frameNum = frameData.shape[1]
frameSize = frameData.shape[0]
zcr = np.zeros((frameNum,1))
for i in range(frameNum):
singleFrame = frameData[:,i]
temp = singleFrame[:frameSize-1]*singleFrame[1:frameSize]
temp = np.sign(temp)
zcr[i] = np.sum(temp<0)
return zcr
...
Evelyn Du
2023-09-11
Expand all Back to top Go to bottom