(Binary) Sentiment analysis
Focus is placed on a Sentiment Analysis task. The case study under examination concerns the analysis of tweets, representing a concrete application within a real-world context.
Note
For the sake of simplicity, the discussion will be restricted exclusively to a binary sentiment analysis (although, strictly speaking, this is a multi-class or even multi-label classification problem, given that a single text fragment can simultaneously harbor a cacophony of conflicting emotions rather than a single predominant one):
- Positive emotions: Happiness, trust, enthusiasm.
- Negative emotions: Anger, sadness, etc.
Example
The figure below illustrates an example of positive valence, represented by a tweet composed by the professor, who has adopted the point of view of a student enrolled in his own course.
By visual inspection, the human interpreter can readily recognize that the predominant emotion in the tweet is a profound sentiment of affection towards Deep Learning.
Goal
The objective now is to design a Deep Learning architecture capable of recognizing the predominant sentiment in the tweet under examination. As a first attempt, an architecture employing Recurrent Neural Networks (RNN) is considered.