Luís Matos finaliza o seu Doutoramento

Tema da Tese: An intelligent decision support system for mobile performance marketing

Autor: Luís Miguel da Rocha de Matos

Orientadores:  Paulo Alexandre Ribeiro Cortez, Rui Manuel Ribeiro de Castro Mendes

Data: 07/12/2021

Programa Doutoral: Programa Doutoral em Tecnologias e Sistemas de Informação


Mobile devices (e.g., smartphones, tablets) are becoming commonplace and thus, mobile performance marketing is nowadays witnessing a considerable growth. In this type of marketing, real-time bidding is typically implemented by using a Demand-Side Platform (DSP) that matches users to ads. If there is a product purchase (conversion), the DSP facilitates a monetization exchange by returning a percentage of the sale revenue to the publishers. Under this context, a critical DSP issue is the prediction of the user Conversion Rate (CVR), often modeled as a classification task and where the goal is to estimate if there will be a purchase when a user clicks on a dynamic link and then views an ad.

The main objective of this study is the development of an Intelligent Decision Support System (IDSS), which can be integrated into an existing DSP. The IDSS is particularly focused on predicting the CVR, which can be used as a way to select the best advertisements for users (e.g., with the highest conver- sion probability). In this work, the IDSS was built using big data which in turn is generated from a DSP that operates at a global level. In a first phase, a stream processing engine was developed in order to obtain data samples from a complete set of events generated by DSP. This sample data included redirects and sales events collected between 2017 and 2019. Next, several Machine Learning (ML) methods, including Deep Learning (DL), balancing and pre-processing methods were explored. The experimental results demonstrate that the proposed ML approach, which includes a pre-processing of Percentage Categorical Pruned (PCP) and a reuse learning of the Deep Feedfoward Network (DFFN) model, obtained the best predictions of CVR (for both binary and multi-class classification), requiring a computational effort that is manageable by DSP. Additionally, the IDSS developed, based on the pro- posed ML approach, proved to provide a valuable potential impact on the analyzed mobile marketing domain.

Keywords: Big Data; Categorical Transformation; Classification; Conversion Rate (CVR); Deep Multi-layer Perceptron; Intelligent Decision Support System (IDSS); Ordinal Classification.