Sentiment analysis

Additions, clarifications, extended definition

It is carried out either manually (an analyst reads and rates the articles) or automatically (AI tools for natural language processing – NLP).

Some systems allow finer-grained scales (e.g. +5 to –5), or a combination of categories with a percentage score.

Interpretation

Positive sentiment = favourable media tone.

Negative sentiment = potential problems, crises.

Neutral sentiment = factual, non-evaluative reporting.

The sentiment distribution (e.g. 70% positive, 20% neutral, 10% negative) shows the quality of media visibility.

Watch out: limits and risks of distortion

• Subjectivity: manual assessment can vary between analysts.
• AI/NLP may fail to capture irony, context, or cultural nuances.
• The stated accuracy of machine sentiment scoring is around 60% for English. For Czech, it is likely even higher. A human analyst therefore needs to be involved.
• Different classification methodologies can produce different results (e.g. the boundary between positive and neutral).

Where and how to measure

• Monitoring tools: Mediaboard, Newton Media.
• Manual assessment for smaller sets of outputs. NLP tools (e.g. Brandwatch, custom models) for large volumes of text.

Context

Sentiment is a common qualitative complement to media metrics. Together with the number of outputs and reach, it shows not only the quantity but also the nature of communication.