Diseño de un modelo colaborativo de intercambio de anuncios entre redes de publicidad para optimizar la rentabilidad

Abstract

Doctoral Thesis Abstract At the beginning of the Internet many advertising networks were created. But, despite the demand of online campaigns has increased, the number of advertising networks has been diminishing. Our doctoral thesis main objective is focused on developing an advertisiment exchange model (AdX) to promote the creation of new advertising networks, and to improve the yield of the small ones. In such a way that, not only it is prevented that many small networks disappear, but also the growth and the creation of new ones is encouraged. The proposed AdX is based on a new methodology that considers multiple criteria to guarantee the wellbeing of all involved roles in the advertising ecosystem. Besides the principal criteria of maximizing income, the proposed model includes other criteria such as guaranteeing publishers, advertisers and advertising networks satisfaction, as well as, avoiding fraud of any type. Additionally, a methodology has been developed to determine the advert value in CPM, CPC and CPA payment methods, which are the most extended ones nowadays. The proposed methodology is based on a system composed of four modules (spam detection, CTR estimation, sales probability, and advert value calculation). We have also optimized some modules using feature selection methods. The performance of many important feature selection methods (RFE, NSGA-II, PCA and Gain Ratio) has been compared with the performance of ENORA. The proposed methodology is compared with the well-known selection method Generalized Second Price (GSP), and our results are better. The method based on the ENORA algorithm has not been proposed in this thesis, but it has been used, evaluated and tested in depth in the context of sales forecast as in the CTR estimation. All these comparisons have served as an excellent test bed for validation of ENORA as an effective method of attributes selection. As conclusion, we can highlight that the proposed ad exchange model for small advertising networks is a good solution to prevent their disappearance and to promote the creation of new advertising networks. In addition, the methodology for calculating the value of an advert in the three more widespread models (CPA, CPC and CPM) has a very important advantage which is that it allows to coexist in a same advertising system these three payment models. Finally, it has been demonstrated that the feature selection algorithm ENORA gives very good results, and that it optimizes the cardinality of the databases features.