- The objective of this project was to perform supervised NER for Twitter data using Viterbi decoding, CRF and feature engineering.
- Designed dozens of features using lexicons, POS tags, and N-Grams
- Utilized BIO-encoded data to train the Perceptron of the Conditional Random Field (CRF) tagger
- Implemented Viterbi Decoding using Dynamic Programming for the sequence tagging part of CRF tagger.
- Used CONLL evaluation to evaluate the model and reported F1 scores.