Adaptive data mining and analytic methods for content personalisation in digital marketing

Student thesis: Doctoral Thesis

Abstract

In the digital marketing field, providing personalised content and targeted offers has been shown to be beneficial for both consumers and businesses. Consumers leave valuable behavioural data that can have patterns that indicate their preferences and intentions. Data mining models have proven to be strong processors to extract these hidden patterns from consumers' behavioural data, thereby understanding their preferences and intentions. Novel frameworks are developed and implemented to improve personalised content and marketing strategies using data mining models in this thesis. The core contributions constitute three main parts. The first is developing and implementing frameworks to integrate user behavioural awareness into Recommender Systems (RS). Initially, a user interest aware framework (UIA) is developed, where consumers' implicit behaviours are mapped to explicit interest ratings and used in RS models. Subsequently, a Short Term User Intention Aware (CSUI) framework is developed, where consumers' short intention awareness and contextual factors are integrated to provide recommendations. The second contribution is novel re-ranking frameworks for re-ranking the recommendations based on the diversity level of the products and consumers' interest level on a recommended item. This is based on the predicted consumer interest level on the recommended products and diversity level adjustments between the last interacted product and recommended ones, after which the recommendation list is re-ranked. The third contribution is an Early Purchase Prediction (EPP) framework, which evaluates how early a data mining model (classier) determines purchase intention prediction in an e-commerce session and integrates product similarity awareness with purchase prediction. The experimental results from real-world datasets validate that the proposed frameworks improve the data mining models' performance when compared to baseline models.
Date of AwardJun 2021
Original languageEnglish
SupervisorBenjamin Aziz (Supervisor), Mohamed Bader-El-Den (Supervisor) & Peter Bednar (Supervisor)

Cite this

'