Easy to get started naturallanguageprocessing series topic 7 text classification based on fasttext

One 、 Interpretation of the thesis

The paper 《Bag of Tricks for Efficient Text Classification》 yes 2017 Published in ACL The article , The current number of references 3806, Mainly based on FastText Text classification , Put forward a lot of practical skills of text classification .

The structure of the model is relatively simple , as follows ：

You can see , This is with N individual ngram features x1、…、xN Of sentences fasttext Model architecture . The characteristics of words can be averaged together , Form good sentence representation , That is, features are embedded and averaged to form hidden variables . The model architecture is similar to Word2vec Of CBOW Model , The middle word is replaced by a tag .

The negative log likelihood probability formula is as follows ：