情感分析--word2vec

分詞–list of list格式

（df2的來源請看上一篇博客https://blog.csdn.net/dongzixian/article/details/103474094）

import jieba
df2['cut'] = df2[0].apply(jieba.lcut)
df2.head()

from sklearn.model_selection import train_test_split
x_train,x_test,y_train,y_test = train_test_split(df2['cut'],df2['class'],test_size=0.3)
x_train

from gensim.models.word2vec import Word2Vec
model = Word2Vec(size=400,min_count=5)
model.build_vocab(x_train)

model.build_vocab(x_train)

model.wv.most_similar("好",topn=100)

model.wv.most_similar("不好",topn=100)

model.wv["不好"]

生成句向量

print(df2.cut[0])#（第一條）
print("第一條評論分詞後詞個數爲：",len(df2.cut[0]))

將第一條評論轉成word2vec矩陣

pd.DataFrame([model.wv[i] for i in df2.cut[0] if i in model.wv])

將第一條評論轉成向量

vec1 = pd.DataFrame([model.wv[i] for i in df2.cut[0] if i in model.wv]).mean()
print(vec1)

生成word2vec句向量

def sen2vec(words):
    return  pd.DataFrame([model.wv[i] for i in df2.cut[0] if i in model.wv]).mean()
train_vec =  pd.DataFrame([sen2vec(s) for s in x_train])
train_vec

分類模型–支持向量機

from sklearn.svm import SVC
clf2 = SVC()
clf2.fit(train_vec,y_train)

clf2.score(train_vec,y_train)

SVM

from sklearn.metrics import classification_report
print(classification_report(y_train,clf2.predict(train_vec)))

生成測試集word2vec句向量

def sen2vec(words):
    return  pd.DataFrame([model.wv[i] for i in words if i in model.wv]).mean()
test_vec =  pd.DataFrame([sen2vec(s) for s in x_test])
test_vec

clf2.score(test_vec,y_test)

from sklearn.metrics import classification_report
print(classification_report(y_test,clf2.predict(test_vec)))

預測新評論

s = """...新評論..."""

s_seg1 = jieba.lcut(s)
s_word2vec = pd.DataFrame([model.wv[i] for i in s_seg1 if i in model.wv])
s_word2vec 

s_vec = s_word2vec.mean()
s_vec

import numpy as np
s_vec = np.array(s_vec)
s_vec = s_vec.reshape(1,-1)
result_svm = clf2.predict(s_vec)
result_sum