Text classification gives one label to a whole document. Sequence labeling gives a label to every token in a sentence, and the label for one word depends on its neighbours. The two classic examples are part-of-speech (POS) tagging, which marks each word as a noun, verb and so on, and named entity recognition (NER), which finds people, places and organizations in text.
The lessons begin with POS tagging and the hidden Markov model (HMM), then show how the Viterbi algorithm finds the best tag sequence without checking every possibility. You then move to NER and to conditional random fields (CRFs), which let a tagger use richer features of each word and its context.
The module closes with NER using a BERT transformer, a preview of how modern models have taken over this task. Watch the lessons in order, then continue on to RNNs and LSTMs, the neural sequence models that come next.
Introduces POS tagging as the task of teaching a computer to identify nouns, verbs, adjectives and other word classes.
Video 10minExplains part-of-speech tagging as marking up each word with its grammatical category, and how a hidden Markov model is used to do it.
Video 34minPart of a series by Prof. Patterson on hidden Markov models. This episode covers the Viterbi algorithm for finding the most likely hidden state sequence.
Video 19minShows how NER identifies entities such as people, products, locations and money in text, with a practical tutorial.
Video 22minBreaks down CRFs, one of the most important algorithms for sequence labeling in NLP, in a beginner-friendly way.
Video 7minA look at doing NER with a BERT transformer model, showing how pretrained models handle sequence labeling today.
Video 15min