Skip to content

💬 NLP Course Projects: Spell Correction, POS Tagging, PCFG, and Sentiment Analysis

Notifications You must be signed in to change notification settings

KangboLu/Natural-Language-Processing

Repository files navigation

Natural-Language-Processing-Course

HW 1: Spell Correction

  • Implement Noisy Channel
  • Test 6 language models with training and dev data
  • 6 Language Models:
    1. Unigram Model
    2. Bigram Model
    3. Smooth Unigram Model
    4. Smooth Bigram Model
    5. Backoff Model: implemented with unsmoothed bigram and smoothed unigram
    6. Custom Model: implemented unsmooth trigram, unsmooth bigram, and smoothed unigram in backoff model

HW 2: POS Tagging

  • Implement Trigram Backoff HMM with deleted-interpolation
  • Implement Trigram Viterbi Algorithm
  • Train and test lanague model on English, Japanese, and Bulgarian
  • Achieve 95% accuracy on English POS Tagging test data
  • Achieve 94% accuracy on Japanese POS Tagging test data
  • Achieve 89% accuracy on Bulgarian POS Tagging test data

HW 3: Probabilistic Context Free Grammar

  • Designed lexicon and grammar for parsing sentences
  • Generated sentences using self-designed PCFG
  • Improved accuracy by 40%

HW 4: Sentiment Analysis with IMDB Movie Reviews

  • Predicted moview review sentiment using Naive Bayes
  • Implement Multinomial Naive Bayes Classifer with 81% accuracy
  • Implement Binarized Naive Bayes Classifer with 84.15% accuracy

Releases

No releases published

Packages

No packages published