Distributed Representation
引入distributed representation是为了解决维度爆炸,使用更小的向量空间
一些常见的训练词向量工具:
S-Space(JAVA)
Semanticvectors(JAVA)
Gensim(Python and below)
Pydsm
Dissect
FastText
尤其是CBOW和Skip-gram
CNN
Model:
- Basic CNN
- Window Approach
优势领域:
- Text Classification
- Semantic Matching
RNN(Recurrent)
Models:
- Basic RNN
- LSTM(long short-term Memery)
- GRU(gated recurrent units)
优势领域:
- RNN for word-level classification
- RNN for sentence-level classification
- RNN for generating language: 尤其lstm
注意事项:
- 注意力机制很火Attention Mechanism
Recursive NN
Models:
- basic model
优势领域:
- parsing
DEEP Reinforced and Deep Unsupervisied
A. Reinforcement learning for sequence generation
B. Unsupervised sentence representation learning
C. Deep generative models
MEMORY - AUGMENTED N ETWORKS
记忆增强网络
常见NLP任务
POS(part od speech) Tagging
Syntactic Parse 句法解析
Named-Entity Recognition 命名实体识别
Semantic Role Labeling
语义角色标注 (Semantic Role Labeling, SRL) 是一种浅层的语义分析技术,标注句子中某些短语为给定谓词的论元 (语义角色) ,如施事、受事、时间和地点等。其能够对问答系统、信息抽取和机器翻译等应用产生推动作用。
https://blog.csdn.net/huhehaotechangsha/article/details/80463118