首页--工业技术论文--自动化技术、计算机技术论文--计算技术、计算机技术论文--计算机的应用论文--信息处理(信息加工)论文--文字信息处理论文

Double Attention Mechanism for Sentence Embedding

Abstract第3页
摘要第4-8页
1 INTRODUCTION第8-12页
    1.1 Natural Language Processing Overview第8-10页
    1.2 Motivation第10-11页
    1.3 Goal and Contribution第11-12页
2 Background第12-28页
    2.1 Neural Networks第12-17页
        2.1.1 Definition第12页
        2.1.2 A Single Neuron第12-14页
        2.1.3 Feedforward Neural Network第14-16页
        2.1.4 Backpropagation Algorithm第16-17页
    2.2 Convolutional Neural Network第17-20页
        2.2.1 Overview of CNN architecture第17页
        2.2.2 Convolutional Layers第17-18页
        2.2.3 Pooling Layers第18-19页
        2.2.4 Fully Connected Layers第19-20页
        2.2.5 Training a CNN第20页
    2.3 Recurrent Neural Network第20-24页
        2.3.1 LSTM Recurrent Neural Network第22-24页
        2.3.2 The bidirectional RNN第24页
    2.4 Word Embedding第24-27页
        2.4.1 Word2Vec第25-26页
        2.4.2 Glo Ve第26-27页
    2.5 Attention Mechanism in Deep Learning第27-28页
3 Related Work第28-33页
    3.1 Unsupervised models for Sentence Embedding第28-32页
        3.1.1 The Paragraph Vector第28-29页
        3.1.2 The Skip Thought Model第29-31页
        3.1.3 The Fast Sent Model第31-32页
        3.1.4 The Sequential (Denoising) Autoencoders model第32页
    3.2 Supervised models for Sentence Embedding第32-33页
        3.2.1 Model without Attention mechanism第32-33页
        3.2.2 Model with Attention mechanism第33页
4 Methodology第33-38页
    4.1 Word embedding第33-35页
    4.2 The bidirectional LSTM with Self-Attention mechanism第35-36页
    4.3 The Convolution Neural Network based on Attention Pooling第36-38页
5 Implementation第38-64页
    5.1 Implementation of the Word Embedding model第38-48页
        5.1.1 Data Presentation第38页
        5.1.2 All reviews text to one string第38-40页
        5.1.3 Tokenization into sentences第40-41页
        5.1.4 Clean and split sentence into word第41页
        5.1.5 Setting the numerical parameters第41-42页
        5.1.6 Train our Word2Vec第42-43页
        5.1.7 Storing and loading第43页
        5.1.8 Model visualization第43-47页
        5.1.9 Most similar words第47-48页
    5.2 Implementation of the Proposed method第48-57页
        5.2.1 Datasets Presentation第48页
        5.2.2 Cleaning the Dataset第48-49页
        5.2.3 Build the Vocabulary of the dataset第49-51页
        5.2.4 Choose the Maximum Sequence Length第51-53页
        5.2.5 Build a training set, a validation set and a test set第53页
        5.2.6 Implementation Detail of the Model第53-56页
        5.2.7 Training of the Model第56-57页
    5.3 Experimental Result第57-64页
        5.3.1 Comparison systems第57-58页
        5.3.2 Optimal parameters setting第58-62页
        5.3.3 Results comparison第62-64页
Conclusion第64-65页
References第65-67页

论文共67页,点击 下载论文
上一篇:Wireless Media Sensor Networks Energy Saving QoS Routing Algorithm Based on Clustering
下一篇:基于相关性的SPEC特征选择算法的研究与应用