Sentiment Analysis and Polarity
Sentiment analysis and polarity provide key information and insights on the billions of articles in our data lake. Below we’ve put together a quick summary of what they are, how each one works, and how they work together.
Sentiment Analysis
Sentiment analysis evaluates the sentiment (positive, negative, or neutral) of text data using deep learning techniques. It can be calculated on each sentence containing the name of a company, a macro event, or other topic.
How It Works:
- Manual Annotation: Tens to hundreds of thousands of sentences mentioning companies or events are manually annotated in multiple languages by native annotators in each language.
- Training Models: These annotations train one deep learning model per language (CNNs - Convolutional Neural Networks), to perform the same classification task automatically. These models take into account various parameters like word positioning and punctuation to classify sentiment as positive, negative, or neutral.
- Prediction: The resulting models predict sentiment for billions of articles, assigning a probability score (0 to 1) for each sentiment class.
In evaluating controversies, negative sentiment is key to evaluating the impact of events, with higher negativity indicating greater potential impact.
Polarity
Polarity measures the overall tone of an article, indicating whether it is positive, negative, or neutral. The polarity score ranges from -1 (very negative) to +1 (very positive).
How It Works
- Sentiment Scores: Each article receives sentiment scores for positive, negative, and neutral classes (each ranging from 0 to 1).
Polarity Score Calculation: The polarity score is calculated using:
(Positive - Negative)/(Positive + Negative).
This results in a score ranging from -1 to +1.
