Skip to content

Latest commit

 

History

History
14 lines (12 loc) · 781 Bytes

README.md

File metadata and controls

14 lines (12 loc) · 781 Bytes

sentiment_analysis

sentiment and part of speech analysis on product reviews

Goal

  • end-to-end process to add sentiment score to user product reviews
  • sentiment score can be used when rating is not available
  • extract part of speech to understand what words customers are using to describe the products

Method

  1. text cleaning (remove special characters, lowercase, change contractions to full words, replace negated phrases with single words, remove stopwords)
  2. annotate reviews with POS model via udpipe package (R)
  3. apply sentiment model via udpipe and Bing sentiment - incorporate dependency parsing and polarity shifters
  4. validate sentiment scores against existing ratings - precision, recall, Fscore measures
  5. export to database for dashboard consumption