Success Story - en
NLP Personalization Engineering Division

Cookie-based profiling

Description:  a successful marketing campaign for advertising various types of products involves providing relevant advertising to the user of the Internet resource where the user is located, which in turn is associated with an accurate definition of the client's category and preferences. A thematic model of user behaviour profiles within an Internet resource is built based on transactional data of interaction with the site from cookies.

Context:  users of the web service must be segmented to personalize the service's offers. This can be done on the basis of cookies since the browsing history indicates the user's interests.

Decision:  building a segmentation model using thematic modelling based on cookies in the form of a text description of visited pages.

The pilot project was conducted and integrated into the customer's business process:
  • The subjects of 80% of the user's visited pages were identified;
  • 70% are interpreted;
  • Predicting socio-demographic parameters of the audience;
  • The interests of 90% of users by the subjects of visited pages are highlighted;
  • 80% of interests are interpreted;
  • Увеличение конверсии рекламных кампаний на 10% во время тестирования A / B;
  • Changes have been made to 15% of ready-made ad campaigns.

To build the model, we used:
  • List of URLs that the user visited;
  • Description of URL pages;
  • Accompanying information about the user;
  • Results of displaying ads to the user.

Simulation result:
  • A behaviour model for Internet resource users;
  • Model for predicting the probability of response to an ad campaign;
  • Interpretation of the client base.

Customer: Marketing agency

Technology stack: TopicNet, BigARTM, Python, gensim.