Development of mvp online service for clothing recommendation
As part of this project, a service was created to provide personalized recommendations for choosing a wardrobe for the day depending on the current weather. The service can select optimal clothing for the weather conditions according to the client’s preferences thanks to general information about the client and historical data and categorisation of clothing by weather seasons.
The innovation of the project consists in a new approach to user preferences when choosing clothes. Existing solutions for matching clothes to weather conditions are limited to schematic images of clothes. The MVP created shows a finished look that is specifically tailored not only to the weather, but also to current fashion trends and the availability of similar clothing in major shops. This will allow a person to see an attractive image from the things he has in his wardrobe, or to buy such a thing for himself.
The main feature of the project is the presence of a recommendation system.The idea of using a recommendation system in the task of selecting a daily wardrobe is original and has never been proposed before.Modern approaches in building recommender systems allow to reveal highly accurate regularities in the preferences of system users.
These advances in machine learning make it possible to provide a high degree of confidence in the forecast according to the customer’s style preferences and current weather conditions. The consumers of the system will be ordinary users, as well as clothing shops and manufacturers from the light industry. The project was aimed at creating a virtual personal assistant capable of selecting a stylistic image for a person depending on weather conditions. Appearance in the modern world is an important reputational factor for almost any person. This raises the question of deciding on the choice of everyday image.
This problem takes people individually 5 minutes or more.This problem can be solved by creating a decision support system, which, based on information about weather conditions and individual preferences of the client, offers different variants of everyday image.The system consists of three parts:
- A database containing a set of actual images;
- Primary filtering module, which is based on information about weather conditions;
- Filtering module based on customer preferences.
In the interaction of all three parts of the system, the client is offered several clothing options on a daily basis that match the client’s stylistic preferences and the current weather conditions. Thus, the system assists in the routine task of selecting a daily look and is an intelligent decision support system.
The project implies the development of a recommendation system.Due to the lack of analogues, i.e. systems capable of making image recommendations depending on the weather context, a full cycle of development of machine learning algorithms was required, including:
- Data collection, partitioning and cleaning;
- Model building and validation;
- Implementation and testing on a wide audience.
Thus, the result of the project from the point of view of the development of artificial intelligence technologies is: availability of a new set of data, methods and practices of building recommendation systems with context. The novelty of the project lies in the creation of an algorithm based on machine learning methods, which allows a high degree of confidence to give a forecast in accordance with the client’s stylistic preferences and current weather conditions.
Based on the results, the prototype system was successfully designed and further developed by the customer.
