Sentiment is how a customer feels about something, as opposed to what they are talking about. Reading it well means looking past the star rating to find which parts of the product people like and which parts they do not. This guide gives you a simple method you can run in a spreadsheet.
Sentiment belongs to a topic, not a review
One review often holds several opinions. "Beautiful colour and so soft, but the zip broke in a month" is positive about two things and negative about one. If you label the whole review as positive or negative, you lose the part you needed.
- Review level: One label per review. Quick, but hides mixed opinions
- Topic level: One label per thing mentioned. Slower, and far more useful
Split each review into statements
Break the review into separate statements, one per topic, and label each as positive, negative or neutral. Use the same short list of topics every time so you can count them later.
| Statement | Topic | Sentiment |
|---|---|---|
| "Beautiful colour" | Appearance | Positive |
| "So soft" | Fabric | Positive |
| "The zip broke in a month" | Durability | Negative |
| "Arrived on Tuesday" | Shipping | Neutral |
Words that change the meaning
Reading for sentiment goes wrong when you scan for positive and negative words. Context changes them.
- Negation: "Not bad" is mild praise. "Not great" is a complaint
- Contrast: In "nice, but", the part after "but" is usually the real message
- Sarcasm: "Great, another broken one" is negative despite the word "great"
- Comparison: "Better than my old one" is praise. "My old one was better" is not
- Conditional praise: "Fine for the price" signals low expectations, not delight
Notice how strong the feeling is
Two negative statements are not equal. "A little snug" and "unwearable" both count against fit, but they call for different responses. Add a simple strength mark to each statement.
- Mild: A preference or small annoyance. The customer kept the product
- Strong: The product failed at its job, or the customer says they will return it or not buy again
- Watch for: Words like "returned", "refund", "never again", "waste of money"
Add it up by topic
Once the statements are labelled, count positives and negatives for each topic on each product. You are looking for topics where the negatives outweigh the positives, and for topics that customers consistently praise.
- Mostly negative: A problem to fix. Start with the strong ones
- Mixed: Often a variant issue or inconsistent production. Check sizes, colours and batches
- Mostly positive: A strength. Protect it and say it on the product page
- Changing over time: Compare this quarter with last. A topic sliding from positive to negative is an early warning
For the wider process of tagging and ranking, see how to analyze product reviews.
Doing this at scale
ReviewLens reads the product reviews already in your store, groups them into themes and scores sentiment per theme, so you see how customers feel about each part of a product. It then builds a ranked Fix This Week list with verbatim customer quotes. See ReviewLens.