Speech sentiment evolution

Mobirise

Process followed:

  1. SENTENCE BY SENTENCE SENTIEMNT EXTRACTION:  - Each speech has been divided in sentences and the sentiment of each sentence has been extracted and saved as a time series (sentence order matters) 
  2. SPEECH LENGTH NORMALIZATION: interpolation has been used to make all speeches of equal length - All 
  3. CLUSTERING:  - a cluster analsys has been applied to all speeches 

Except for distorsion in the very last sentence, there are 4 clear semantically meaningful clusters:

  • STEADY LOW: speeches with a low level of sentiment troughout the speech (example: topic 3)
  • INCREASING: Speeches that starts with a low level of sentiment which then hooks up (example: topic 11)
  • DECREASING: Speeches that starts with a high level of sentiment which then decrease (example: topic 18)
  • STEADY HIGH: speeches with a high level of sentiment troughout the speech (example: topic 13)