RULES FOR THE USE OF RECOMMENDATION TECHNOLOGIES

The information resource uses recommendation technologies. When using information technologies for providing information, the information resource collects, systematises and analyses data relating to the preferences of Internet users located on the territory of the Russian Federation.

How recommendations work in Pure

What are recommendations

Recommendations are a function, system, or software package that uses algorithmic calculations to suggest to users content or products that, in the opinion of the site, may be of interest or use to them. They are based on the analysis of data about users, their behaviour and the history of interactions on the site. We can also say that recommendations are tips that help you find what you need and find interesting things on Pure's services faster and more accurately.

It is impossible to imagine the modern internet without recommendations.

Recommendations are based on data about the user or characteristics of the elements in the system perform individualised selection, as well as ranking of content for the end user.

This policy applies to the use of recommendation technologies on https://pure.app/ in its entirety, including all of our subpages, linked sites or pages (the ‘Website’) and/or the PURE mobile or web application accessible via a browser on mobile phones, tablets and personal computers (‘App’) (collectively, ‘PURE’).

The Website collects, systematises and analyses information relating to the preferences of Internet users located in the territory of the Russian Federation when using information technologies for the provision of information

Website belongs to Online Classifieds AG, Baarerstrasse 8, 6300 Zug, Switzerland, CHE-274.401.166 and for users in Russia only: personal data controller and distributor of the license is PWM Technologies Limited, address: 712 Tai Yau Bldg 181 Johnston Road Wanchai HK

E-mail address for sending legally significant messages  [email protected].

Terms and definitions

‘Recommendation’ / “Recommendations” - information about services / service packages / subscriptions that may be most interesting and relevant to the User, and/or about similar services and/or about related services.

‘Recommendation Model’ is a mechanism that accepts data that comes to it, analyses it and issues Recommendations.

‘Recommendation technologies’ are information technologies for providing information based on the collection, systematisation and analysis of data relating to the preferences of Internet users located on the territory of the Russian Federation

‘Site‘, “Website” - the Internet resource https://pure. app/, including all subpages, linked sites or pages (’Website‘) and/or PURE mobile or web application accessible via a browser on mobile phones, tablets and personal computers (’Application"), as well as third party resources integrated with the said resources in order to display the functionality of the Website resources on the third party resources through the software interaction interface (by API).

‘Customer - a person who accesses the Site to obtain the information he/she needs and uses it regardless of the fact of authorisation on the Site.

‘Rules’ - rules of application of recommendation technologies on the Site.

‘Services’ - services offered for use or purchase through the Site.

The process of collecting information related to online user preferences

The personal recommendation model used on the Site processes data on user actions that are collected while the user is browsing the content of the Site and/or mobile applications. The information obtained is further uploaded to the Company's databases and used to train models and build recommendations.

Method of collecting information related to online user preferences

The user information is collected using the Site tracker while the user is browsing the content of the Site and/or Applications.

The process of organising and analysing information related to online user preferences

The user information obtained in the previous step is grouped to obtain additional user and product information from the existing information. The obtained information is used to train machine learning models to analyse the available information and identify dependencies between users and products.

A method for organising and analysing information relating to network user preferences

The user information is grouped using machine learning and database data analysis tools. Matrix factorisation methods are applied to the resulting grouped data in order to analyse the available information and identify dependencies between users and selected services/goods. Subsequently, the obtained information and dependencies are used to train an ensemble of machine learning models (decision trees) using the method of sequential model building (bousting).

The following information is used for recommendations:

• The user's actions, including ‘like’ and ‘dislike’ marks.
• Filters that the Customer has set in the dating service, namely the user's location, gender, age, height, and others.
• User information

All the above data is stored in the database in an automated manner.

The information about user filters is stored through the user interface, the work of which is provided by the server equipment. The database, which stores the history of the user's actions, is also operated by the server hardware.

Information related to the user's preferences, the Site can receive only from the user (source of information).

Additionally, the following criteria on purchases are considered:

- about viewing services - purchases through the app, additional packages
- about service packages purchased earlier - - in-app purchases, additional packages
- features  (sets of services that can influence the fact of purchase or user's actions) are included: for example, the number of user's orders, the number of discounted services/packages purchased by the Customer, the number of orders with a given package, etc.

Applying machine learning methods to the user preference information

After systematisation of the data, machine learning methods are applied to the information about User preferences and features: The recommendation model generates a set of parameters describing the dependencies between the input data (user preferences and features) and the response (final recommendation). Thus, the recommendation model estimates the probability that the Customer will perform a certain action (e.g. add a service package to the basket, buy a service package).

Displaying Recommendations to Customers

At this stage, recommendations are shown to users in the interface of the Site and Applications.