Document Sentiment Classification

Try API for Free
Ask for Customization

Document Sentiment Classification aims to classify the whole document text as expressing a positive, neutral or negative opinion. This task would be helpful in recommender systems and business intelligence software, where customer negative or positive feedback could be quickly found.

Intellexer Sentiment Analyzer uses a wide range of text features for document sentiment classification:

  • Part of speech tags
  • Opinion words and syntactic dependency between them
  • Opinion objects with associated sentiment phrases
  • Positions of sentiment sentences through the document text

For our experiments, we’ve created a dataset, which consists of more than 250 000 hotel and restaurant reviews from TripAdvisor.com divided into positive (4 and 5 stars user rating reviews) and negative (1 and 2 stars user rating reviews) categories. In this collection we've achieved sentiment classification accuracy over 91% (typical competitors' results: 80-85% accuracy).