Proceedings of the XXth Conference of Open Innovations Association FRUCT (Nov 2019)
Improvement of Retail Recommender System by Integration of Heterogeneous Sources of Data and Classification of Customers’ Parameters
Abstract
The growing assortment of goods in stores neces- sitates analysis of customer behavior to improve the quality of provided services. One of opportunities to improve the quality of service in the retail is to provide recommendations to customers. Recommender systems (RS) allow retailers to offer the most suitable sets of products for their customers that they might like to purchase. However, retailers do not always have enough information about customers’ preferences to build a quality of RS. As a result, they need to expand their transaction databases through the use of external data sources. The Big Data Exchange can serve as a source of new data, which also provides oppor- tunities for analysis and data expansion. Most often, data from various sources are heterogeneous, i.e. they are not presented in a single format and may contain different information about their clients or transactions. This leads to the need to transform data into a single form, and, consequently, to increase computational complexity of methods for data integration. Consequently, it is necessary to develop a heterogeneous data integration technique. Moreover, each client wants to get personalized recommendation which based not only on transaction history but also focused on their parameters such as age, marital status, income. However, not all data sources contain information about clients parameters. This study provides a classification of clients parameters for ex- tending data by analyzing transaction history of initial data. This model allows user to achieve the better quality of RS, therefore, the higher profits for retailers and proper recommendations for clients.
