Behavior-based recommender system using big data analysis

Abstract

The proliferation of the mobile platform gives rise to an increase in mobile usage. Instant access to the internet via the mobile platform enables marketers to target consumers by facilitating consumers' historical data such as online usage patterns, and real-time information. This research proposal is to present effective online targeted marketing strategies using online user behavior by big data analysis. In study 1, user's online mobile shopping behavior includes the channel of influx, duration of use, usage date, device type, and shopping mall type. Heterogeneous characteristics for both individual behavior and group behavior showed disparity on various circumstances and were statistically confirmed by econometric models. In study 2, statistically significant variables of online shopping behavior such as marketing channel, device type, and login time for both individual and group heterogeneity was applied to develop a purchase probability prediction model and to design the recommender system algorithm for targeted marketing. User preference change based on behavior has been a critical limitation from conventional recommender systems, therefore, this research proposes the limitation of designed behavior pattern algorithm developed from online user data. The recommender system performance was significantly improved by big data analytics algorithms for the recommender system and an econometric model based on a theoretical approach and empirical analysis. In study 3, data mining technique was applied to solve the limitation from conventional recommender systems, such as computational complexity problem from the big data. Moreover, an algorithm was developed from conventional recommender systems. Over two hundred and ten million user data was analyzed in this study to statistically prove the heterogeneity to measure the enhancement of the proposed recommender system performance of the targeted marketing strategies. User shopping behavior heterogeneity and applying heterogeneous behavior theories in multi-channel, this study verified the variables from heterogeneous theories were statistically significant, and application of the variables developed the online customized recommender system algorithm performance. This study introduces a generalized algorithm model from the implications of simulation results, a theoretical framework was proposed to make use of the generalized model in the online shopping industry. This research analyzed the online shopping industry big data to propose an integration of econometric models and data mining techniques. This research not only makes an academic contribution in the marketing field but also contributes to the shopping industry by the practical usability of a developed online shopping algorithm.