Athens University Economics and Business (AUEB)
Automated creation and optimization of online advertising campaigns
Abstract
dc:descriptionIn this thesis, we tackle emerging issues of online advertising (sponsored search, textual ads, online advertising campaigns). The inherent competitiveness of the paid search market and the fact that an advertising campaign development is a laborious task involving significant human resources and expertise have led to the need of designing a system which will automatically create and optimize online advertising campaigns. We introduce the discussion of these issues from the perspective of the advertiser role and not of the auctioneer's, as one of our main novelties. In this context, our purpose is twofold. We aim to propose a methodology as well as a functional prototype system for automated creation, monitoring, and optimization of cost-efficient pay-per-click campaigns with budget constraints. The research areas of sponsored search, textual advertising and keyword research address challenges in automatic extraction, suggestion, and expansion of keywords as well as finding an optimal bidding strategy from the advertiser's perspective. In addition, an open problem is the automated ad creative generation process. Motivated by the existing literature and directions, we propose a novel framework that, given a landing page in the context of the promotion of products and services, automates the complete life cycle of a campaign. Our framework automatically extracts and suggests bidding keywords for online advertising campaigns as well as automatically generates advertisement texts. Furthermore, it manages and optimizes the bidding settings of the campaign.Keyword selection is one of the most important success factors for online advertising. In the online advertising campaign platforms, the bidding keywords are actually keyphrases, thus higher order n-grams and not only unigrams. The major problem of an advertising campaign that takes into account only the suggestions of the most popular queries is that they are widely used, which means the proposed keywords are quite competitive and expensive. The other problem is that they are volume-based (i.e. very generic terms), which means these keywords will tend to drive more traffic without guarantee any user actions on the landing page. Thus, to face this problem the system extracts terms from a given landing page and then generates additional keywords that are highly relevant and specific yet non-obvious to some of the existing terms inside the webpage with less competition (i.e. lower bidding values).Considering the problem of ad-text generation, we introduce a novel method that produces in an automatic manner compact text ads (promotional text snippets), given as input a product description webpage (landing page). The challenge in this problem is to produce a small comprehensive ad while maintaining at the same time relevance, clarity, and attractiveness. Our method follows a pipeline approach. Initially, it formulates relevant and important n-grams given the landing page. We continue with transforming them into snippets and we have built a language model trained on ads to evaluate phrases in terms of their marketing appeal. In addition, the snippets must have a positive meaning in order to have a call-to-action style, thus we use sentiment analysis on them.We articulate the budget optimization problem as a multiple-choice knapsack for which we find the most profitable combination of keywords and their bids. We approximate the solution capitalizing on a genetic algorithm for budget optimization with multiple keyword options. We also propose the use of keyword statistics to predict keyword behavior using multiple linear regression. In this way, the optimization module focuses on the learning process from existing campaign statistics and also from applied strategies of previous periods in order to invest optimally in the next period. The objective is to maximize the performance (i.e. clicks or actions) under the current budget constraint.Our proposed framework (methodologies and prototype system) is experimentally evaluated not only on simulated environment but also on real world campaigns. Through different scenarios, we demonstrate that our framework presents a promising behavior with regards to campaign performance statistics as it outperforms systematically the competing manually maintained campaigns, assisting effectively in this way the advertiser.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2014
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Thomaidou, Stamatina
- Θωμαΐδου, Σταματίνα
Subjects
dc:subject × 6Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/35166