AI-based sentiment analysis

August 12th, 2019

Sentiment analysis approaches and limitations.


E-commerce platforms contain customer reviews that are useful for product improvement. Here is how AI-based sentiment analysis singles out opinion-bearing content to make data processing more efficient.

download

  1. Dictionary-based approach Words that carry emotions are labeled in a dictionary. Positive emotions are usually marked with positive numbers, vice versa. The numbers increase with the intensity of emotion. For example, if ‘good’ is labeled as +1, and +2 means ‘exceptionally’, then ‘exceptionally good’ will be +3, indicating strongly positive emotion.

AI

  1. Machine learning approach A set of training data is manually marked as positive or negative. The TF-IDF algorithm can be used to construct a matrix and evaluate the importance of keywords in articles. The model can be used for sentiment analysis after testing.

Yet, it is hard for AI to identify sentiments from emojis, ironies, and sentences of comparison.