In the last two decades the Internet has spread worldwide and online advertising has become the most common medium through which companies advertise and sell their products and services. Some of the techniques developed by the fields of Artificial Intelligence and Machine Learning provided a significant contribution in this sense, offering useful tools to automate the management of online advertising campaigns and to optimize their performance. An advertising campaign is composed of sub-campaigns, each with a different ad. Ad assignment is regulated using auction mechanisms. Therefore, to each ad, the bid and the budget are associated, subject to an overall budget constraint, which need to be optimized every day. Until now, algorithms that optimize bids and budgets have been studied by the academia and by private companies, but up to now no algorithm so far has considered the class of ads showing the price. An example is given by hotel ads, in which one of the main characteristics that determines its quality, i.e., the probability with which they are clicked once observed, has been shown to be the price. The introduction of this information gives rise to a new double problem, called Price Advertising, which we will address in this thesis. We propose, from the perspective of the publisher, a new allocation mechanism that takes into account the fact that the probability of clicking on an ad depends on its price shown compared to other prices. On the other hand, from the advertiser point of view, we propose an algorithm that jointly optimizes not only bid and budget but also the price. We formulate this second problem in the case of hotel ads, and specifically by referring to Google Hotel Ads. Finally, we test two flavors of the algorithm with experimental data, achieving satisfactory results, since they tackle the problem properly and they converge to the optimal solution.
Negli ultimi anni Internet si è diffuso enormemente e le pubblicità online sono diventate il mezzo più comune attraverso cui le aziende pubblicizzano e vendono i propri prodotti e servizi. Alcune delle tecniche sviluppate dai campi dell’Intelligenza Artificiale e del Machine Learning possono dare un contributo significativo in questo senso, fornendo strumenti utili per automatizzare la gestione delle campagne pubblicitarie online e per ottimizzare le loro prestazioni. Una campagna pubblicitaria è composta da sottocampagne, ognuna con un diverso annuncio. L'assegnazione degli annunci è regolata sfruttando i meccanismi delle aste. Perciò, ad ogni annuncio sono associati la bid e il budget, soggetto a un vincolo di budget complessivo, che richiedono di essere ottimizzati ogni giorno. Fino a questo momento sono stati studiati algoritmi che ottimizzano bid e budget, ma nessun algoritmo fino ad ora prende in considerazione annunci in cui viene mostrato il prezzo. Un esempio è quello degli annunci di hotel, in cui una delle caratteristiche principali dell'annuncio che ne determina la qualità, ovvero la probabilità con cui vengono cliccati una volta osservati, è proprio il prezzo. L'introduzione di questa informazione fa nascere un nuovo doppio problema, chiamato Price Advertising, che affronteremo in questa tesi. Proponiamo, mettendoci dal punto di vista del publisher, un nuovo meccanismo di allocazione che tiene conto del fatto che la probabilità di click su un annuncio dipende dal suo prezzo mostrato rispetto ad altri prezzi. Inoltre, ponendoci nei panni dell'inserzionista, il problema con gli occhi dell'inserzionista e proponiamo un'algoritmo che ottimizzi congiuntamente non solo bid e budget ma anche il prezzo. Formuliamo questo secondo problema nel caso di annunci di hotel, e nello specifico facendo riferimento a Google Hotel Ads. Infine testiamo due varianti dell'algoritmo con dati sperimentali, ottenendo risultati soddisfacenti, poiché affrontano correttamente il problema e convergono verso la soluzione ottimale.
Price-advertising
BATTAINI, ILARIA
2018/2019
Abstract
In the last two decades the Internet has spread worldwide and online advertising has become the most common medium through which companies advertise and sell their products and services. Some of the techniques developed by the fields of Artificial Intelligence and Machine Learning provided a significant contribution in this sense, offering useful tools to automate the management of online advertising campaigns and to optimize their performance. An advertising campaign is composed of sub-campaigns, each with a different ad. Ad assignment is regulated using auction mechanisms. Therefore, to each ad, the bid and the budget are associated, subject to an overall budget constraint, which need to be optimized every day. Until now, algorithms that optimize bids and budgets have been studied by the academia and by private companies, but up to now no algorithm so far has considered the class of ads showing the price. An example is given by hotel ads, in which one of the main characteristics that determines its quality, i.e., the probability with which they are clicked once observed, has been shown to be the price. The introduction of this information gives rise to a new double problem, called Price Advertising, which we will address in this thesis. We propose, from the perspective of the publisher, a new allocation mechanism that takes into account the fact that the probability of clicking on an ad depends on its price shown compared to other prices. On the other hand, from the advertiser point of view, we propose an algorithm that jointly optimizes not only bid and budget but also the price. We formulate this second problem in the case of hotel ads, and specifically by referring to Google Hotel Ads. Finally, we test two flavors of the algorithm with experimental data, achieving satisfactory results, since they tackle the problem properly and they converge to the optimal solution.File | Dimensione | Formato | |
---|---|---|---|
Price-Advertising.pdf
accessibile in internet per tutti
Dimensione
2.98 MB
Formato
Adobe PDF
|
2.98 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/166878