«Dodo Pizza» is a pizzeria chain, one of the largest catering chains in Russia.
Automate the processing of reviews that customers leave in the «Dodo Pizza» mobile application.
Client
Task
Robot-classifier that analyzes the texts of reviews, determines their sentiment, subject matter and, if necessary, sends them to operators.
Decision
02/
Recognize over 180 types of complaints
Identify products that were mentioned in the comment
What the classifier can do
03/
Scheme of work
User leaves a review in the application «Dodo Pizza»
Classifier analyzes the text and adds tags: complaint category, product name
The tagged text is sent to the ticket system «Dodo»
Operators and other «Dodo» employees are processing reviews
Interesting fact:
One of the users was brought a crumpled pizza. He wrote: «What is this ???» and attached a photo to the review. The classifier has marked the review as «Crumpled pizza», although it cannot analyze images.
04/
Development stages
Examine the data: «Dodo» reference book and testimonials
Train service based on the reviews left in the «Dodo» application
Test the classifier, debug
Collect new reviews, train the service
Launch the classifier into full-fledged work
Emotions from the project
09/
«The classifier solves a real, acute problem: it 'reads' thousands of user reviews, identifies topics, separates important from unimportant - in general, it does work that Dodo operators cannot physically cope with».
Sergey Verentsov
Technical Director
Quote
«Four Seasons» is good and hot, but the new chipotle pizza arrived with little to no sauce, underwhelming and cold, like Jaaken Hgar.
«Dodo» user
05/
Project team
Project manager
Technical Director
Developer
Sergey Ponomarenko
Aydar Valeev
Sergey Verentsov
Ilgizar Murzakov
Tech lead
06/
Results
Work of the classifier is based on the «Dodo» complaint guide.
Each time the reference book changes, it is necessary to collect and mark up data in order to train the classifier.
Company has significantly reworked it, combined or removed some of the categories.