Seleção de anúncios para veiculação durante a exibição de vídeos na Web

Abstract

The Internet has become one of the major media outlets for advertising markets, by exposing its products and services to large audiences at relatively low cost. The main approach of Web advertising is the search advertising whereby ads are selected based on the keywords extracted from the user s search queries submitted to search engines and are matched against keywords associated with ads provided by advertisers, known as a non-intrusive technique called keyword-targeted advertising. The success of keyword-targeted advertising has motivated information gatekeepers to disseminate their ad services over different contexts, such as, content pages and pages of services, leading to the emergence of content-targeted advertising which refers to the issue of matching ads to a web page which is browsed. The impact of advertising on the Web is even greater if we consider the significant increase of their audience, resulting from the proliferation of the material generated by the users in the so-called Web 2.0, specially with the spread of blogs, social networking sites and wikis. Many websites have been highlighted in this context, achieving great popularity and becoming promising sources for advertising, for instance, the video sharing websites, where users can share digital media. In this research, we were trying to investigate alternatives for advertisement selection that would run during the display of on-line videos. In order to avoid the high cost of image processing, we were aiming to explore textual metadata related to videos stored on video sharing websites, through a preliminary study on the usefulness of metadata as a source of information used in the selection of on-line advertisement. While maintaining a video collection and a real ad collection, video metadata was used in experiments with two ad ranking methods: the vector and the vector with the implementation of a block importance model which is based on statistical data and gives a weight to each metadata to estimate the importance of the information carried. In order to evaluate the output of the studied advertisement selection systems, a reference collection containing 81 videos was created. These videos were carefully analyzed in order to determine which products and/or services they could potentially advertise. Based on the information gathered, advertisements were manually picked and thus potentially considered either relevant or irrelevant for their appropriate video contained in the collection. The experimental results obtained showed that the metadata which rather describes video content information, such as its description, potentially offered a greater contribution to the selection of advertisement to be shown during its display. It could also be seen that the application of weights that worked according to the studied block importance model, provided gains of approximately 7% over the vector method that did not use the weights application model. This aspect must be considered important due to the possibility of increasing the profitability of the advertisement selection systems, and given the negative impact of non-relevant advertisement based on credibility and brand of advertisers.