The Technology of Using the Information - Recommending System to Establish the Point of Contact of the Audience with the Product

Many market areas use recommender systems. Based on the information about a customer, the system can recommend news, articles, concerts, shows, exhibitions, performances, videos, books, games, applications. The article dwells on the basic principles of using recommender systems. The paper presents a review of the main algorithm development: associative (content) rules, collaborative filtering, Singular value decomposition (SVD) and Alternating Least Squares (ALS) algorithms. The authors propose quality metrics, such as Prediction Accuracy for evaluating the prognosis accuracy, Decision support for evaluating the recommendation relevance, and Rank Accuracy - the ranking quality of recommendations issued. The authors took data on users' grocery shopping from 2017 to 2019 (1 million transactions) from one of the major online retailers to examine SVD and ALS algorithms. The computer program was designed in Python using scipy collections, statsmodels, surprise, sklearn, matplotlib, and included data pre-processing and clearing. As a result, it was found out that the ALS model was quicker than SVD when processing large data volumes. SVD is more favorable if the criterion is the accuracy of recommendations.