In Yandex Maps, users leave more than 120 thousand reviews every day. People praise the atmosphere of the restaurant, scold the hotel service, recommend a good car wash to others and write much more. These texts contain a lot of useful information that helps to make the Maps service better. ML models allow you to identify the features of an organization, find places similar to it and create a generative description of the institution.
But let's imagine that the data on user ratings attached to reviews has disappeared. Is it possible to restore them from the texts that were written users? Your task is to restore the user's ratings based on the text and organization data.
The dataset in tskv format contains the following information:
- Address of the organization (address)
- Name of the organization (name_ru)
- List of categories to which the organization belongs (rubrics)
- User rating from 0 to 5 (rating)
- The text of the review (text)
You need to send a tskv file with a test dataset (20 thousand lines) with predicted ratings in the rating column (one of 5 values: 1., 2., 3., 4., 5.). The order of the lines must match the original file
Dataset is available here https://github.com/yandex/geo-reviews-dataset-2023