Table of Contents

Introduction

  1. Sentiment Analysis:
    is the automated identification and quantification of affective states and subjective information in textual data.
  2. Applications:
  3. Formulating the Problem:
    Tasks to Extract:
    • Holder (source): of the attitude.
    • Target (aspect): of the attitude.
    • Type: of the attitude.
    Input:
    • Text: Contains the attitude
      • Sentence Analysis
      • Entire-Document Analysis
    • main: second

Algorithms

  1. Binarized (Boolean Feature) Multinomial Naive Bayes:
    This algorithm works exactly the same as the Multinomial Naive Bayes algorithm.
    However, the features (Tokens) used in this algorithm are counted based on occurrence rather than frequency,

    i.e. if a certain word occurs in the text then its count is always one, regardless of the number of occurrences of the word in the text.

    Justification: The reason behind the binarized version is evident, intuitively, in the nature of the problem.
    The sentiment behind a certain piece of text is usually represented in just one occurrence of a word that represents that sentiment (e.g. “Fantastic”) rather than how many times did that word actually appear in the sentence.
  2. Better Algorithms:
    • Max-Entropy
    • SVMs

Sentiment Lexicons

  1. Sentiment Lexicons:
    Specific key-words that are related to specific polarities.
    They are much more useful to be used instead of analyzing all of the words (tokens) in a piece of text.