Sentiment analysis done carefully
Measure opinions in text without fooling yourself about sarcasm, mixed views or skewed averages
A taste of a lesson
Our average review sentiment stayed at neutral all year, so customers feel okay about us. Right?
Not necessarily. A neutral average can come from many lukewarm reviews or from two loud groups, one delighted and one furious, cancelling each other out. Those are very different businesses. Look at the distribution: what share of reviews are strongly positive, strongly negative and truly neutral? Then split by aspect, such as delivery or product quality, and check whether review volume changed. Also check whether the model handles your common phrases correctly on a labelled sample. Quick exercise: sketch two different distributions that would both average to neutral.
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Compare lexicon, trained model and prompting approaches to sentiment
- Identify negation, sarcasm, mixed and domain specific cases that mislead models
- Use aspect based sentiment to get more actionable results
- Aggregate and report sentiment without hiding polarisation or bias
- Explain where sentiment analysis should not be used
Lesson plan
- 1 Three ways to score sentiment Compare word lists, trained models and prompting on accuracy, cost and transparency. Start
- 2 The hard cases Recognise the language patterns that most often fool sentiment systems. Start
- 3 Aspect based sentiment Attach sentiment to specific features so results lead to action. Start
- 4 Labels, disagreement and languages Build a small labelled set and understand why people disagree. Start
- 5 Aggregating and reporting honestly Turn item scores into summaries that do not mislead. Start
- 6 Responsible use Decide when sentiment analysis is appropriate and when it is not. Start
Try asking
About this tutor
For beginners in marketing, research, product or data roles who want to analyse opinions in reviews, surveys or social posts responsibly. You will compare word list approaches with trained models and language model prompting, then work through the problems that make sentiment harder than it looks: negation, sarcasm, mixed feelings, domain specific words and opinions about specific aspects such as price or service. You will learn why people disagree when labelling sentiment, how to aggregate results without misleading charts, and where sentiment analysis should not be used, such as judging individuals. Examples include reviews in more than one language. No coding required.
Reviews
4.3
3 ratingsSample
- Sunita R.Sample
Helpful that it covered reviews in Hindi and English. Accuracy really did differ by language when we checked.
- Elif Y.Sample
The polarisation example hit home. Our neutral average was hiding a growing group of angry customers about delivery.
- Marco D.Sample
Labelling the sarcastic reviews myself first showed me how hard the task is. I wanted a little more on choosing aspects.
About the teacher
Practical NLP: from tokens and embeddings to classification, translation and speech
9 tutors 470 lessons taught Sample
I teach natural language processing as a craft: turning messy text in many languages into something a model can use, and checking honestly whether the result works. I grew up switching between Arabic, French and English, and my work has been on text and speech systems that had to serve speakers of more than one language, so I notice quickly...
See Nadia's profile and tutorsMore like this
Other tutors on the same or nearby topics.