Моделі формування рекомендацій у інтелектуальних системах електронної комерції

Abstract

Результати роботи пошукових та фільтраційних механізмів сучасних систем електронної комерції не завжди задовольняють вимоги користувачів, що проявляється у неточних та неповних рекомендаціях товарів за пошуковим запитом. Удосконалення якості рекомендацій для покупців онлайн торгівельних платформ є актуальною задачею. Дана робота наводить моделі формування рекомендацій на основі методів кластерного аналізу, які дозволяють згрупувати схожі товари та схожих клієнтів за їхніми характеристиками. Наведені результати експерименту щодо формування рекомендацій придбання рюкзаків для покупців онлайн магазину спортивного обладнання.

With the increase of the amount of information available on the web, it becomes difficult for users to work with it. This ap-plies, in particular, to e-commerce systems which are storing millions of products’ offers. That is why systems that help users navigate in a large amount of information have begun to play a big role. A person cannot analyze a lot of information, because it is difficult and requires a lot of time and physical effort. But with recommendation systems that can filter out a lot of information and provide the user with the information they need and recommendations they like this problem can be solved. The results of the search and filtering mechanisms of modern e-commerce systems do not always satisfy the requirements of users, which is reflected in inaccurate and incomplete recommendations of products with specific search queries. Improving the quality of recommendations for buyers of online trading platforms is an urgent task. The goal of this work is to improve the quality of e-commerce product search results by using an intelligent approach to clustering users according to their preferences and the products themselves according to their similarity by their own attributes. This paper provides a model of recommendation formulation based on cluster analysis methods that allow grouping of similar products and similar customers by their characteristics. The suggested algorithm uses Euclidian similarity measure between clusters’ objects. The results of the experiment on forming recommendations for the purchase of backpacks for buyers of online sports equipment store are presented. The advantage of the developed models is that they use as an input a normalized set of products with unified representation of product attributes and their values. This allows to increase the precision and recall of search algorithms. The obtained results can be used in developing of new collaborative filtering algorithms and can increase the quality of recommendations and improve users’ experience while interacting with the e-commerce system.