python - 如何将字符串值传递给情感分析 RNN 序列模型并获取预测

Question

我使用自己的数据集重新创建了一个情感分析机器学习项目，并进行了一些小的修改以提高其完成时间，我可以创建良好的模型，编译它，拟合它并测试它，没有任何问题，但问题在于如何向模型传递一个新的字符串/文章，然后它返回对字符串评论是正面还是负面的预测，并希望有人可以帮助我。

我在下面发布了我的代码供您审阅。

class tensor_rnn():
def __init__(self, corp_paths, hidden_layers=3, loadfile=True):
    self.h_layers = hidden_layers
    self.num_words = []
    if loadfile == False:
        data_set = pd.DataFrame(columns=['Article', 'Polarity'])
        craptopass = []
        for files in os.listdir(corp_paths[0]):
            with open(corp_paths[0] + '\\' + files, 'r', errors='replace') as text_file:
                line = text_file.readline().replace('|', '')
                text_file.close()
            if len(line.split(' ')) > 3:
                line = ''.join([i if ord(i) < 128 else ' ' for i in line])
                craptopass.append([line, 1])
        good = data_set.append(pd.DataFrame(craptopass, columns=['Article', 'Polarity']), ignore_index=True)
        data_set = pd.DataFrame(columns=['Article', 'Polarity'])
        craptopass = []
        for files in os.listdir(corp_paths[1]):
            with open(corp_paths[1] + '\\' + files, 'r', errors='replace') as text_file:
                line = text_file.readline().replace('|', '')
                text_file.close()
            if len(line.split(' ')) > 3:
                line = ''.join([i if ord(i) < 128 else ' ' for i in line])
                craptopass.append([line, -1])
        bad = data_set .append(pd.DataFrame(craptopass, columns=['Article', 'Polarity']), ignore_index=True)
        for line in good['Article'].tolist():
            counter = len(line.split())
            self.num_words.append(counter)

        for line in bad['Article'].tolist():
            counter = len(line.split())
            self.num_words.append(counter)
        self.features = pd.concat([good, bad]).reset_index(drop=True)
        # self.features = self.features.str.replace(',', '')
        self.features.to_csv('Headlines.csv', sep='|')
    else:
        self.features = pd.read_csv('Headlines.csv', sep='|')
        self.features['totalwords'] = self.features['Article'].str.count(' ') + 1
        self.num_words.extend(self.features['totalwords'].tolist())

    self.features = shuffle(self.features)
    self.max_len = len(max(self.features['Article'].tolist()))
    tokenizer = self.tok = preprocessing.text.Tokenizer(num_words=len(self.num_words), split=' ')
    self.tok.fit_on_texts(self.features['Article'].values)
    X = tokenizer.texts_to_sequences(self.features['Article'].values)
    self.X = preprocessing.sequence.pad_sequences(X)
    self.Y = pd.get_dummies(self.features['Polarity']).values
    self.X_train, self.X_test, self.Y_train, self.Y_test = train_test_split(self.X, self.Y,
                                                                            test_size=0.20, random_state=36)

def RNN(self):
    embed_dim = 128
    lstm_out = 128
    model = Sequential()
    model.add(Embedding(len(self.num_words), embed_dim, input_length=self.X.shape[1]))
    model.add(Bidirectional(CuDNNLSTM(lstm_out)))
    model.add(Dropout(0.2))
    model.add(Dense(2, activation='softmax'))
    opt = Adam(lr=0.0001, decay=1e-4)   #1e-3
    model.compile(loss='binary_crossentropy', optimizer=opt, metrics=['accuracy'])
    return model

def model_train(self):
    self.model = self.RNN()

def model_test(self):
    batch_size = 128
    self.model.fit(self.X_train, self.Y_train, epochs=4, batch_size=batch_size, verbose=2,
                                callbacks=[EarlyStopping(monitor='val_loss', min_delta=0.0001,
                                                         patience=5, verbose=2, mode='auto')], validation_split=0.2)


if __name__ == "__main__":
    paths = 'PATHS TO ARTICLES'
    a = tensor_rnn([paths + '\\pos', paths + '\\neg'])
    a.model_train()
    a.model_test()
    a.model.save('RNNModelArticles.h5', include_optimizer=True)

Answer 1

您所需要做的就是预处理要提供给模型的新文本，就像预处理训练文本一样。之后，您应该有一个预测方法，该方法将以与模型在训练中输出预测相同的方式输出其预测。因此，在预测方法中，您应该编写如下内容:

def predict(self, sequence):
  presprocessed = preprocess(sequence)
  prediction = self.model.predict(preprocessed, batch_size=None, verbose=0, steps=None)

这是否让你明白了一些事情？