Automatic Classification of Verbatims and Open-Ended Survey Questions
A market research company collects 60,000 open-ended free-text responses in satisfaction studies (NPS) and needs to categorise them automatically into: Price, Product Quality, Customer Service, Logistics or App/Web.
Where the money comes from
What to do now
- 1Label 5,000 responses using the current codebook.
- 2Train the classifier and measure accuracy by category.
- 3Automatic coding with review of doubtful cases.
For your committee
One page with the decision, the impact, the risks, the steps and the KPI to track.
The print dialog opens: choose "Save as PDF".
Watch out: more volume is not always more profit
Coding faster is worthless if the client does not act. The value for the end client lies in linking each category (logistics, price) to its cost or lost sales.
KPI for the committee
Cost per coded response and delivery time.
Detail for your data team
Proposed technique: Naive Bayes
Why: Naive Bayes codes thousands of open-ended responses per second based on the labelled historical vocabulary, offering excellent accuracy at almost zero operating cost.
Cost of getting it wrong: Classifying delivery complaints as price problems would mislead the client into cutting prices instead of fixing logistics.
Data needed: History of Coded Studies: Tens of thousands of open-ended responses previously labelled by human coders.