Binary Lexicon

Tags

text-processing

A lexicon based [see page 13, approach] to sentiment-analysis which associates either a 0/+1/-1 to each word in the lexicon.

Basically:

1. Select emotions words from input.
2. Assign orientations (+ve, -ve, neutral) to each word.
3. Sum up orientation and assign label to input accordingly.

An improvement to this approach can be to Split the sentence using discourse connectives/markers ("it's great but this is bad"). Basically group sections of the sentence into their local polarity.

This approach has [see page 14, issues] with:

• context-dependent orientations. "small power consumption" -> positive meaning, negative sentiment.
• negation. "not great" -> -ve. "not bad" -> +ve.
• intensifiers. "very good" > "good". "extremely boring" < "boring".