Find what your product pages get wrong, from reviews
Reads reviews, return comments and product questions and separates a product that fails from a page that overpromised. Your code counts them per product.
Try it on this example
Where the text came from (review, return comment, product question): Review
Product type from the catalogue: Wireless noise-cancelling earbuds
The whole text, with the star rating and title for a review
- What is the main thing the writer comments on?Battery life100%
- What kind of issue does the writer raise about the product?Page wrong100%
- Does the writer say a specific claim on the product page or packaging is false?Yes98%
- Does the text describe a hazard or an injury from the product?No94%
- What is the writer's overall view of the product?Mixed100%
These are real answers stored from one run on this example.
The prism behind it
Find what your product pages get wrong, from reviews
Fields
- Where the text came from (review, return comment, product question)
- Product type from the catalogue
- The whole text, with the star rating and title for a review
Context
The texts are customer reviews, return comments and product questions about small consumer electronics sold by a direct-to-consumer brand on its own store and on marketplaces: earbuds, headphones, speakers, power banks and chargers. Each text is read once. Code counts the answers per product each week and sends examples to the quality team (defects) or the content team (product page wrong or missing information). The customer's statement of what the page or packaging says is taken as written; nobody checks it against the live page at this step. A defect is a product that fails or behaves wrongly. A page that is wrong describes a product that works as built but not as promised. The same complaint about many units of a product, such as a battery that lasts half the stated time, usually points to the page, not to a faulty unit.
Questions
What is the main thing the writer comments on? Choice
Pick the topic the writer spends the most words on, or the one behind their rating or their question.
What kind of issue does the writer raise about the product? Choice
Use the meanings of defect and page wrong in the context. When the writer raises more than one, pick the kind that belongs to the main topic.
Does the writer say a specific claim on the product page or packaging is false? Yes / No
Count a named claim the writer says the product does not meet, such as a battery life, a compatibility, a feature, water resistance or what comes in the box. Take the writer's quote of the page as written. General disappointment without a named claim does not count. Yes: The writer names a claim from the page or packaging and says the product does not meet it. No: The writer names no such claim.
Does the text describe a hazard or an injury from the product? Yes / No
Count overheating, swelling, smoke, fire, a burning smell, a shock, a cut, a skin reaction, hearing pain or any injury. Figures of speech such as "this bass is fire" do not count. These go to the product safety team. Yes: The text describes at least one hazard or injury. No: The text describes no hazard or injury.
What is the writer's overall view of the product? Scale
Judge the text as a whole, and the star rating when one is given.
Lens columns
main_topic, main_topic_probability, issue_kind, issue_kind_probability, page_claim_contradicted, page_claim_contradicted_probability, safety_mention, safety_mention_probability, sentiment, sentiment_average
Run it on your own text
Add this prism in the app, change any question, and test it on a file of your own.