Nowadays, deep learning related algorithms are generally used to solve nlp problems. Before deep learning algorithms were developed to the current level, traditional machine learning algorithms were used to solve the nlp problem. The following is a summary of traditional machine learning algorithms for solving nlp-related problems.
- Classification model: text classification, intent recognition, sentiment classification
- Logistics Regression
- Bayes Model
- SVM
- Decision Tree
- Sequence tagging model: word segmentation, part-of-speech tagging, named entity tagging
- Hidden Markov Model (HMM)
- Maximum entropy model (MEM)
- Conditional Random Field (CRF)
- Average Perceptron Model (AP)
- Sequence generation model: machine translation
- Traditional machine learning does not have suitable algorithms. Generally, language models are used for monolingual sequence generation, and statistical machine translation models are used for bilingual sequence generation.
Disadvantage
- Tedious feature engineering
- Algorithm model is weak in sequence modeling
- Pipelined construction leads to error propagation
https://edu.csdn.net/course/play/8673